Patent Application
20230338953
This application claims priority from prior Japanese Patent Applications No. 2022-042965, filed on Mar. 17, 2022, and No. 2022-042964, filed on Mar. 17, 2022, the entire contents of which are incorporated herein by reference.
The present invention relates to a specimen analyzer, a specimen analysis method, and a program that analyze a specimen.
International Publication No. WO2018/203568 discloses a method in which signals obtained by measuring cells by a flow cytometer are analyzed by an artificial intelligence algorithm, and the cells are classified according to the types.
The scope of the present invention is defined solely by the appended claims, and is not affected to any degree by the statements within this summary.
When data is processed by using an artificial intelligence algorithm, the load on a computer that processes the data is increased in accordance with increase in the volume of the data. For example, when the amount of information to be used in classification of components (e.g., cells or particles) in a specimen is increased in order to improve classification accuracy, in a case of a specimen such as blood or urine that contains a plurality of components, the data volume per specimen increases due to increase in the amount of information obtained from a single component. When the number of specimens to be tested has been increased as well, the data volume increases. International Publication No. WO2018/203568 does not disclose any technology that can reduce the load on a computer at the time when data is processed by using an artificial intelligence algorithm.
A specimen analyzer (4000) of the present invention relates to a specimen analyzer configured to analyze an analyte in a specimen. The specimen analyzer (4000) of the present invention includes: a measurement unit (400) including an optical detection part (410, 470) configured to obtain an optical signal (80a, 80b, 80c) from the specimen; and an analysis unit (300, 600) configured to analyze first data and second data (82a, 82b, 82c) that correspond to the optical signal (80a, 80b, 80c). The analysis unit (300, 600) executes, on the first data (82a, 82b, 82c), a first analysis operation according to an artificial intelligence algorithm (60), and executes a second analysis operation of processing a representative value, of the second data (82a, 82b, 82c), that corresponds to a feature of the analyte.
A specimen analysis method of the present invention relates to a specimen analysis method for analyzing an analyte in a specimen. The specimen analysis method of the present invention includes: obtaining (S1, S11, S121, S131, S141, S302) an optical signal (80a, 80b, 80c) from the specimen; and analyzing (S2, S3, S14, S16, S71, S74, S81, S84, S91, S95, S122, S124, S132, S134, S142, S143, S201, S202) first data and second data (82a, 82b, 82c) that correspond to the optical signal (80a, 80b, 80c). The analyzing (S2, S3, S14, S16, S71, S74, S81, S84, S91, S95, S122, S124, S132, S134, S142, S143, S201, S202) includes executing, on the first data (82a, 82b, 82c), a first analysis operation according to an artificial intelligence algorithm (60), and executing a second analysis operation of processing a representative value, of the second data (82a, 82b, 82c), that corresponds to a feature of the analyte.
A computer-readable medium having stored therein a program of the present invention relates to a computer-readable medium having stored therein a program configured to cause a computer (300, 600, 3001, 3002, 6001, 6002) to execute a process of analyzing an analyte in a specimen. The program of the present invention includes a process of analyzing first data and second data (82a, 82b, 82c) that correspond to an optical signal (80a, 80b, 80c) obtained from the specimen. The process executes, on the first data (82a, 82b, 82c), a first analysis operation according to an artificial intelligence algorithm (60), and executes a second analysis operation of processing a representative value, of the second data (82a, 82b, 82c), that corresponds to a feature of the analyte.
A specimen analyzer (4000) of the present invention relates to a specimen analyzer configured to analyze an analyte in a specimen. The specimen analyzer (4000) of the present invention includes: a measurement unit (400) including an optical detection part (410, 470) configured to obtain an optical signal (80a, 80b, 80c) from the specimen; and an analysis unit (300, 600) configured to analyze data (82a, 82b, 82c) that corresponds to the optical signal (80a, 80b, 80c). In accordance with an analysis mode of the data (82a, 82b, 82c), the analysis unit (300, 600) analyzes the data (82a, 82b, 82c) through a first analysis according to an artificial intelligence algorithm (60) or through a second analysis of processing a representative value, of the data (82a, 82b, 82c), that corresponds to a feature of the analyte.
A specimen analysis method of the present invention relates to a specimen analysis method for analyzing an analyte in a specimen. The specimen analysis method of the present invention includes: obtaining (S1, S11, S121, S131, S141, S302) an optical signal (80a, 80b, 80c) from the specimen; and analyzing (S2, S3, S14, S16, S71, S74, S81, S84, S91, S95, S122, S124, S132, S134, S142, S143, S201, S202) data (82a, 82b, 82c) that corresponds to the optical signal (80a, 80b, 80c). The analyzing (S2, S3, S14, S16, S71, S74, S81, S84, S91, S95, S122, S124, S132, S134, S142, S143, S201, S202) includes analyzing, in accordance with an analysis mode of the data (82a, 82b, 82c), the data (82a, 82b, 82c) through a first analysis according to an artificial intelligence algorithm (60) or through a second analysis of processing a representative value, of the data (82a, 82b, 82c), that corresponds to a feature of the analyte.
A computer-readable medium having stored therein a program of the present invention relates to a computer-readable medium having stored therein a program configured to cause a computer (300, 600, 3001, 3002, 6001, 6002) to execute a process of analyzing an analyte in a specimen. The program of the present invention includes a process of analyzing data (82a, 82b, 82c) that corresponds to an optical signal (80a, 80b, 80c) obtained from the specimen. The process analyzes, in accordance with an analysis mode of the data (82a, 82b, 82c), the data (82a, 82b, 82c) through a first analysis according to an artificial intelligence algorithm (60) or through a second analysis of processing a representative value, of the data (82a, 82b, 82c), that corresponds to a feature of the analyte.
Hereinafter, outlines and embodiments of the disclosure will be described in detail with reference to attached drawings. In the following description and drawings, the same reference characters denote the same or similar components, and description of the same or similar components will be omitted for convenience.
The present embodiment discloses a specimen analyzer, a specimen analysis method, and a program that can execute both of analysis according to an artificial intelligence algorithm (AI algorithm) and analysis that does not use the AI algorithm, on data obtained through measurement of a specimen.
In the analysis by the AI algorithm, data is analyzed by a large number of matrix operation processes, for example. Hereinafter, analysis by the AI algorithm will be referred to as “AI analysis”, for convenience. In the AI analysis, a convolution operation according to the AI algorithm is performed, for example.
In the analysis that does not use the AI algorithm, data is analyzed through calculation processing with respect to a representative value that corresponds to a feature of an analyte, for example. Hereinafter, an analysis method that analyzes data through calculation processing with respect to a representative value that corresponds to a feature of an analyte, without using the AI algorithm, will be referred to as “calculation processing analysis” or “non-AI analysis”, for convenience. The representative value that is processed in the calculation processing analysis has a data amount smaller than that of data that is inputted to the AI algorithm in the AI analysis. In the calculation processing analysis, the data amount to be processed and the amount of arithmetic operation processing are smaller than those of the AI analysis, and thus, the load on the computer that performs analysis is smaller than that in the AI analysis. Accordingly, the TAT (Turn Around Time) of analysis of a measurement result can be shortened.
According to the specimen analyzer, the specimen analysis method, and the program of Embodiment 1, analysis of data obtained through measurement of a specimen is apportioned between and executed by the AI analysis and the calculation processing analysis, whereby the load on the computer that performs the analysis can be reduced.
As shown in the drawing in the upper part of
The measurement unit 400 measures a specimen and obtains data regarding the specimen. The analysis unit 300 analyzes the data obtained by the measurement unit 400. The analysis unit 300 may have a function of performing control for setting a measurement condition of a measurement sample and for execution of measurement in the measurement unit 400. The analysis unit 300 is implemented as an apparatus (e.g., computer) different from the measurement unit 400 and is connected to the measurement unit 400. The analysis unit 300 and the measurement unit 400 are connected to each other in a wired or wireless manner.
The measurement unit 400 includes an optical detection part for measuring a measurement sample prepared from a specimen.
The optical detection part is a detection part based on flow cytometry, for example, and is used in measurement of a blood specimen or a urine specimen. The optical detection part applies light to a measurement sample flowing in a flow cell, thereby obtaining optical signals. For example, the optical detection part applies light to a measurement sample containing an analyte (e.g., cell or particle) flowing in the flow cell, thereby causing forward scattered light, side scattered light, and fluorescence to be generated from the analyte. A photodetector provided to the optical detection part receives the lights having been generated, and outputs optical signals corresponding to the intensities of the received lights. The optical signals are analog signals having waveform shapes that correspond to temporal changes in forward scattered light, side scattered light, and fluorescence. An A/D converter provided to the optical detection part performs digital conversion on each optical signal to obtain digital data (hereinafter referred to as “waveform data”) having a waveform shape that corresponds to each of analytes. For example, the waveform data in this case is used for classification of the types of white blood cells in a blood specimen, classification of the numbers of red blood cells and white blood cells in a blood specimen, and classification of particles in a urine specimen, and the like.
The optical detection part may be configured to apply light to a measurement sample contained in a container and to detect, by means of a photodetector, light transmitted through the measurement sample or scattered light scattered by the measurement sample. In this case, the optical detection part applies light to a measurement sample containing an analyte and left to stand in a state of being contained in the container. The photodetector provided to the optical detection part receives, for a predetermined period, transmitted light transmitted through the measurement sample or scattered light generated from the measurement sample, and outputs an optical signal that corresponds to the intensity of the received light. The optical signal in this case is an analog signal having a waveform shape that corresponds to change over time of the transmitted light or the scattered light associated with coagulation of the measurement sample. An A/D converter provided to the optical detection part performs digital conversion on the optical signal and obtains digital data (hereinafter, referred to as “coagulation waveform data”) having a waveform shape that corresponds to change over time of the transmitted light or the scattered light. The coagulation waveform data in this case is used in, for example, analysis or the like of coagulability of the blood specimen.
Next, an example of analysis performed by the analysis unit 300 by using the data obtained by the measurement unit 400 will be described.
In
As shown in the graph in the upper part of
The measurement unit 400 obtains, as waveform data, the region that corresponds to each of the analytes in the specimen, from the digital data obtained through digital conversion of the optical signal. The waveform data is obtained so as to correspond to a plurality of analytes in the specimen. Through calculation processing, the analysis unit 300 calculates a representative value, of the waveform data, that corresponds to a feature of the analyte. As shown in each graph in
In the calculation processing analysis, a representative value corresponding to a feature of each analyte is determined in advance. For example, in a case where classification and counting of blood cells being an analyte are to be performed, the representative value determined in advance in the algorithm of the calculation processing analysis is the peak value. The analysis unit 300 obtains from the waveform data a representative value determined in advance, through a predetermined calculation, and processes the representative value obtained for analyzing the analyte. With respect to each of a plurality of pieces of the waveform data obtained by the measurement unit 400, the analysis unit 300 obtains a representative value determined in advance. That is, through the predetermined calculation by the analysis unit 300, representative values (e.g., peak value) of an identical type are obtained from the plurality of respective pieces of the waveform data. The representative value determined in advance may be obtained by the measurement unit 400 and the obtained representative value and the waveform data may be transmitted to the analysis unit 300.
Meanwhile, in the AI analysis, since the AI algorithm extracts a feature of the waveform data, the representative value is not determined in advance. Since the feature (i.e., a feature corresponding to an analyte) of the waveform data extracted by the AI algorithm may vary in accordance with the learning content of the AI algorithm, it is not necessary to determine in advance a representative value in the AI analysis. Since the AI algorithm can extract various features of the waveform data in accordance with the learning content, not only the representative value but also the waveform data itself is inputted to the AI algorithm. Since the waveform data itself is inputted to the AI algorithm, the computer load for arithmetic operations of data is increased in the AI analysis and the TAT (Turn Around Time) required in the arithmetic operations is also increased, when compared with the calculation processing analysis.
As shown in the left drawing in
As shown in the right drawing in
The types of the analytes classified through the calculation processing analysis and the AI analysis are, for example, the types of blood cells in a blood specimen, the types of particles in a urine specimen, and the like. For example, the analysis unit 300 executes the AI analysis with respect to measurement items for classifying the types of white blood cells in a blood specimen, and executes the calculation processing analysis with respect to the other measurement items.
The measurement unit 400 obtains, as coagulation waveform data, digital data obtained through digital conversion of an optical signal. In a single measurement, one piece of coagulation waveform data is obtained from one measurement sample.
The graph in
Instead of absorbance, transmitted light intensity may be used. In this case, if the percentage (transmitted light intensity) on the vertical axis is set so as to be increased upwardly, the coagulation waveform data has a shape in which the graph decreases in association with a lapse of time, as in
The coagulation waveform data includes, at least, data that corresponds to the optical signal obtained from a timing T2 that indicates start of coagulation of the specimen to a timing T3 that indicates end of coagulation of the specimen. The coagulation waveform data may include data that corresponds to the optical signal obtained from a start timing T1 of the light measurement to an end timing T4 of the light measurement by the measurement unit 400.
In the calculation processing analysis, the analysis unit 300 calculates, through calculation processing, a representative value, of the coagulation waveform data, that corresponds to a feature of an analyte, and executes analysis on the basis of the calculated representative value. In the calculation processing analysis, the analysis unit 300 specifies, as a representative value, coagulation waveform data at a time when the intensity of the detected light satisfies a predetermined condition. For example, the analysis unit 300 obtains, as a representative value, a time (T-T2) required for the absorbance of the coagulation waveform data to decrease to a predetermined value (e.g., 50%), and provides the obtained representative value, as a result that indicates the time taken for the blood specimen to coagulate.
In the AI analysis, the analysis unit 300 analyzes the coagulation waveform data on the basis of the AI algorithm 60 (see
For example, the analysis unit 300 analyzes the presence or absence of an abnormality due to an interference substance in the blood specimen. As a specific example, the analysis unit 300 analyzes the presence or absence of an abnormality by using coagulation waveform data regarding PT (prothrombin time), which is an item for measuring coagulability regarding prothrombin being a coagulation factor.
It should be noted that, in the AI analysis, the analysis unit 300 may input coagulation waveform data into the AI algorithm 60 and obtain the time taken for the blood specimen to coagulate. Alternatively, in the AI analysis, the analysis unit 300 may input coagulation waveform data to the AI algorithm 60 and obtain a cause for prolongation in a case where the coagulation time has been prolonged.
In step S1, the measurement unit 400 obtains an optical signal by means of the optical detection part, and obtains waveform data from the obtained optical signal.
In step S2, the analysis unit 300 executes the AI analysis on waveform data (first data) to serve as a target of the AI analysis, out of the waveform data obtained by the measurement unit 400. For example, the analysis unit 300 specifies, as the first data, waveform data that corresponds to a measurement item being a target of the AI analysis, and executes the AI analysis on the specified first data.
In step S3, the analysis unit 300 executes the calculation processing analysis on waveform data (second data) to serve as a target the calculation processing analysis, out of the waveform data obtained by the measurement unit 400. For example, the analysis unit 300 specifies, as the second data, waveform data that corresponds to a measurement item being a target of the calculation processing analysis, and executes the calculation processing analysis on the specified second data.
With respect to steps S2 and S3 above, an example case where measurement of classifying the types of white blood cells in a blood specimen is the target of the AI analysis will be described. For example, the measurement unit 400 prepares a blood specimen by using a reagent that corresponds to measurement of white blood cell classification, and measures the prepared measurement sample by means of an optical detection part based on flow cytometry. Since the measurement regarding white blood cell classification is a target of the AI analysis, the analysis unit 300 specifies, as the first data, waveform data based on the measurement sample for white blood cell classification, for example. The analysis unit 300 analyzes the first data by means of the AI algorithm 60, and classifies white blood cells. Meanwhile, for example, the analysis unit 300 specifies, as the second data, waveform data based on a measurement sample other than that for white blood cell classification. The analysis unit 300 specifies a representative value that corresponds to a feature of each analyte from the second data, executes the calculation processing analysis of processing the specified representative value, and classifies blood cells other than white blood cells.
In step S4, the analysis unit 300 provides analysis results obtained in steps S2 and S3. In step S4, for example, the analysis unit 300 performs display of the analysis results on a display part, transmission of the analysis results to another computer, and the like.
It should be noted that, in step S1, the measurement unit 400 may obtain the optical signal by means of the optical detection part from a single measurement sample, and may obtain waveform data from the obtained optical signal. In this case, the first data and the second data are each composed of a plurality of pieces of data, and a part thereof may be the same data between the first data and the second data.
Further, in step S1, from each of a plurality of measurement samples containing a specimen collected from an identical subject, an optical signal may be obtained by the optical detection part, and from each of the obtained optical signals, waveform data may be obtained. In this case, the analysis unit 300 executes, in step S2, the AI analysis on the waveform data (first data) obtained from one measurement sample, and executes, in step S3, the calculation processing analysis on the waveform data (second data) obtained from another measurement sample. The plurality of measurement samples containing a specimen collected from an identical subject may be prepared by using a reagent of the same type with each other, or may be prepared by using reagents of different types from each other.
Further, in step S1, from each of a plurality of measurement samples respectively containing specimens collected from subjects different from each other, an optical signal may be obtained by the optical detection part, and from each of the obtained optical signals, waveform data may be obtained. In this case, the analysis unit 300 executes, in step S2, the AI analysis on the waveform data (first data) obtained from one measurement sample, and executes, in step S3, the calculation processing analysis on the waveform data (second data) obtained from another measurement sample. The plurality of measurement samples respectively containing specimens collected from subjects different from each other may be prepared by using a reagent of the same type with each other, or may be prepared by using reagents of different types from each other.
In Embodiment 1 above, as the calculation processing analysis, in step S3, the analysis unit 300 specifies a representative value that corresponds to a feature of an analyte from the second data, and processes the specified representative value. However, the present disclosure is not limited thereto. For example, in step S1, the measurement unit 400 may obtain a representative value from the waveform data, and output the waveform data and the representative value to the analysis unit 300, and as the calculation processing analysis, in step S3, the analysis unit 300 may process the representative value obtained from the measurement unit 400.
In Embodiment 2, the AI analysis and the calculation processing analysis are selected on the basis of a rule set to the analysis unit 300.
The rule for selecting an analysis operation is set by a user via the analysis unit 300, for example. The user can set, to the analysis unit 300, a rule according to an operation policy of a laboratory, for example. Accordingly, in accordance with the operation policy of the laboratory, apportioning between the AI analysis and the calculation processing analysis can be changed as appropriate.
Since the rule for the analysis operation can be set, apportioning between the AI analysis and the calculation processing analysis can be flexibly changed while the load on the analysis unit 300 is reduced. For example, when the accuracy of the AI analysis has been improved as a result of causing the AI algorithm 60 to additionally learn, it is possible to set the rule such that data to serve as the target of the AI analysis is increased. Further, for example, when shortening of the TAT (Turn Around Time) of analysis of the measurement result is a priority, it is also possible to set the rule such that data to serve as the target of the calculation processing analysis is increased.
In step S11, the measurement unit 400 obtains an optical signal by means of the optical detection part, and obtains waveform data from the obtained optical signal. In step S12, the analysis unit 300 refers to a rule for selecting an analysis operation, and on the basis of the rule referred to, the analysis unit 300 specifies, with respect to the waveform data obtained in step S11, waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis.
In step S13, the analysis unit 300 determines whether or not waveform data to serve as a target of the AI analysis is included in the waveform data specified in step S12. When the waveform data to serve as a target of the AI analysis is included (S13: YES), the analysis unit 300 executes, in step S14, the AI analysis on the waveform data to serve as a target of the AI analysis specified in step S12.
Subsequently, in step S15, the analysis unit 300 determines whether or not there is waveform data to serve as a target of the calculation processing analysis other than the waveform data having been subjected to the AI analysis. When there is waveform data to serve as a target of the calculation processing analysis (S15: YES), the analysis unit 300 executes, in step S16, the calculation processing analysis on the waveform data to serve as a target of the calculation processing analysis specified in step S12.
In some cases, the measurement unit 400 obtains, in step S12, both of waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis. For example, when, in accordance with a measurement order, the measurement unit 400 has executed measurement regarding white blood cell classification and measurement regarding reticulocytes, the measurement unit 400 obtains waveform data for white blood cell classification and waveform data for reticulocyte measurement. When the white blood cell classification is the target of the AI analysis, and the reticulocyte measurement is the target of the calculation processing analysis, the analysis unit 300 determines that waveform data for white blood cell classification being the target of the AI analysis is included (S13: YES), and executes the AI analysis on the waveform data. Further, the analysis unit 300 determines that waveform data for reticulocyte classification being the target of the calculation processing analysis is also included (S15: YES), and executes the calculation processing analysis on the waveform data.
On the other hand, when waveform data to serve as a target of the AI analysis is not included in the waveform data obtained by the measurement unit 400 (S13: NO), the analysis unit 300 executes, in step S16, the calculation processing analysis on the waveform data to serve as a target of the calculation processing analysis specified in step S12. When the AI analysis has been executed and waveform data to serve as a target of the calculation processing analysis is not included (S15: NO), the calculation processing analysis is not executed and the process is advanced to step S17.
In step S17, the analysis unit 300 provides the analysis result.
It should be noted that the analysis unit 300 may determine, in step S13, whether or not waveform data to serve as a target of the calculation processing analysis is included, and may determine, in step S15, whether or not waveform data to serve as a target of the AI analysis is included. In this case, when the analysis unit 300 has determined, in step S13, that waveform data to serve as a target of the calculation processing analysis is included, the analysis unit 300 executes the calculation processing analysis in step S14. Further, when the analysis unit 300 has determined, in step S15, that waveform data to serve as a target of the AI analysis is included, the analysis unit 300 executes the AI analysis in step S16.
In Embodiment 3, various examples of apportioning between the AI analysis and the calculation processing analysis will be described.
For example, apportioning between the AI analysis and the calculation processing analysis is determined by a software program with which the analysis unit 300 executes analysis of waveform data. The software program of the analysis unit 300 specifies waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis, and executes analysis. The software program is designed in accordance with a requirement (e.g., improving the TAT, increasing the analysis accuracy, etc.) regarding the test, for example.
In
In step S21, the analysis unit 300 refers to a rule that includes which of the AI analysis and the calculation processing analysis is to be performed on the basis of measurement items, and on the basis of the rule referred to, the analysis unit 300 specifies, with respect to the waveform data obtained in step S11, waveform data of a measurement item to serve as a target of the AI analysis and waveform data of a measurement item to serve as a target of the calculation processing analysis.
The screen in
Although the user sets either one of the AI analysis and the calculation processing analysis with respect to each measurement item via the screen shown in
With reference back to
For example, in accordance with a measurement order of classifying white blood cells (e.g., five classifications of neutrophil, lymphocyte, monocyte, eosinophil, and basophil), the measurement unit 400 mixes a specimen with a reagent that corresponds to the classification, to prepare a white blood cell measurement sample. The measurement unit 400 obtains optical signals that correspond to the white blood cell measurement sample by means of the optical detection part. The measurement unit 400 obtains waveform data that corresponds to each obtained optical signal. When a measurement item (e.g., the count and proportion of each of neutrophils, lymphocytes, monocytes, eosinophils, and basophils) regarding the white blood cell classification is the target of the AI analysis, the analysis unit 300 executes the AI analysis on the waveform data obtained, by the measurement unit 400, through measurement of the white blood cell measurement sample.
Subsequently, in step S23, the analysis unit 300 determines whether or not there is waveform data of a measurement item to serve as a target of the calculation processing analysis in the waveform data specified in step S12. When there is waveform data to serve as a target of the calculation processing analysis (S23: YES), the analysis unit 300 executes, in step S16, the calculation processing analysis regarding the measurement item on the waveform data to serve as a target of the calculation processing analysis specified in step S21.
For example, in accordance with a measurement order of classifying reticulocytes, the measurement unit 400 mixes a specimen with a reagent that corresponds to the classification, to prepare a reticulocyte measurement sample. The measurement unit 400 obtains optical signals that correspond to the reticulocyte measurement sample by means of the optical detection part. The measurement unit 400 obtains waveform data that corresponds to each obtained optical signal. When a measurement item (e.g., the count and proportion of reticulocytes) regarding classification of reticulocytes is a target of the calculation processing analysis, the analysis unit 300 executes the calculation processing analysis on the waveform data obtained, by the measurement unit 400, through measurement of the reticulocyte measurement sample.
Not limited to a blood cell analyzer, the specimen analyzer 4000 may be a urine analyzer or a blood coagulation measurement apparatus. For example, when the specimen analyzer 4000 is a urine analyzer, the analysis unit 300 executes the AI analysis with respect to some of measurement items and performs the calculation processing analysis with respect to the remaining measurement items. When the specimen analyzer 4000 is a blood coagulation measurement apparatus, the analysis unit 300 executes the calculation processing analysis with respect to all of measurement items, executes, with respect to some measurement items, the AI analysis in addition to the calculation processing analysis, and determines whether or not there is a suspected occurrence of nonspecific reaction.
In
In step S31, on the basis of a measurement order, the analysis unit 300 specifies which of waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis the waveform data obtained in step S11 is. The analysis mode for a measurement order is either an AI analysis mode or a calculation processing analysis mode, and is stored in a storage of the analysis unit 300 in association with the measurement order.
The screen in
Not limited to the configuration in which the analysis mode that is associated with each measurement order is set by the user via the analysis unit 300, the analysis mode may be set in advance at the time of setting of a measurement order at a host computer or the like.
With reference back to
In
In step S41, the analysis unit 300 refers to a rule including the analysis mode of the analysis unit 300, and on the basis of the rule referred to, the analysis unit 300 specifies which of waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis the waveform data obtained in step S11 is. When the AI analysis mode has been set in the above rule, all of the waveform data serves as the target of the AI analysis, and when the calculation processing analysis mode has been set in the above rule, all data serves as the target of the calculation processing analysis.
The screen in
With reference back to
In
In step S51, the analysis unit 300 refers to a rule that includes an analysis mode that corresponds to the type of the measurement order, and on the basis of the type of the measurement order and the rule referred to, the analysis unit 300 specifies which of waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis the waveform data obtained in step S11 is. The type of the measurement order includes “Normal” corresponding to normal measurement such as an initial test, “Rerun” corresponding to a re-test in which a measurement item identical to that of the initial test is set, and “Reflex” corresponding to a re-test in which the measurement item has been changed from that of the initial test. In the above rule, for each type of the measurement order, either one of the AI analysis mode and the calculation processing analysis mode is set.
The screen in
Not limited to the configuration in which the analysis mode that is associated with the type of each measurement order is set by the user via the analysis unit 300, the analysis mode may be set in advance in accordance with the type of the measurement order at a host computer or the like.
With reference back to
In
In step S61, the analysis unit 300 refers to a rule for selecting an analysis operation, and on the basis of the measurement item and the type of the measurement order, the analysis unit 300 specifies, with respect to the waveform data obtained in step S11, waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis.
The screen in
In a case where setting has been performed as shown in
With reference back to
Subsequently, in step S15, the analysis unit 300 determines whether or not waveform data to serve as a target of the calculation processing analysis is included in the waveform data specified in step S61. When there is waveform data to serve as a target of the calculation processing analysis (S15: YES), the analysis unit 300 executes, in step S16, the calculation processing analysis on the waveform data to serve as a target of the calculation processing analysis specified in step S61.
In
In step S71, the analysis unit 300 executes the calculation processing analysis on the waveform data obtained in step S11, and sets a flag that suggests an abnormality regarding an analyte in the specimen, on the basis of the result of the calculation processing analysis. The flag is, for example, a flag that indicates a predetermined abnormal cell has been detected, a flag that indicates the count value of predetermined blood cells is an abnormal value, or the like. In step S72, the analysis unit 300 refers to a rule that includes whether or not to perform the AI analysis on an analysis result having a flag.
The screen in
When the check box of a flag is on, the AI analysis is executed on the waveform data that corresponds to the analysis result. In the example shown in
With reference back to
On the other hand, when the specimen is not an AI analysis target (S73: NO), the analysis unit 300 skips step S74.
In
In step S81, the analysis unit 300 executes the calculation processing analysis on the waveform data obtained in step S11, to classify the analytes. In step S82, the analysis unit 300 refers to a rule that includes whether or not to perform the AI analysis for each type of analyte.
The screen in
When the check box for a type of an analyte is on, the AI analysis is executed on the waveform data classified according to the type. In the example shown in
With reference back to
In the calculation processing analysis, a scattergram of which axes represent two types of representative values calculated from waveform data is used. For example, when the AI analysis has been set to be executed with respect to monocytes and lymphocytes as shown in
With reference back to
In
In step S91, the analysis unit 300 executes the calculation processing analysis on the waveform data obtained in step S11, and classifies analytes. In step S92, the analysis unit 300 refers to a rule that includes whether or not to perform the AI analysis on an analyte of a specific type, e.g., a cell that is not present in peripheral blood of a healthy individual.
The screen in
When a check box for a specific type of analyte is on, the AI analysis is executed on the waveform data classified as the type. In the example shown in
With reference back to
In step S17, the analysis unit 300 provides analysis results obtained through the calculation processing analysis and the AI analysis. At this time, the analysis unit 300 replaces, out of the analysis results obtained through the calculation processing analysis in step S91, the analysis result of the type having served as a target of the AI analysis, with an analysis result obtained through the AI analysis, and provides the analysis result. The analysis result obtained through the calculation processing analysis and the analysis result obtained through the AI analysis may be provided in combination.
In Embodiment 4, a detailed configuration example in which, in the specimen analyzer 4000 that analyzes a specimen according to flow cytometry, analysis is executed by being apportioned between the calculation processing analysis and the AI analysis is shown.
Examples of a specimen measured by the specimen analyzer 4000 of Embodiment 4 can include a biological sample collected from a subject. For example, the specimen can include peripheral blood such as venous blood and arterial blood, urine, and a body fluid other than blood and urine. The body fluid other than blood and urine can include bone marrow aspirate, ascites, pleural effusion, cerebrospinal fluid, and the like, for example. Hereinafter, the body fluid other than blood and urine may be simply referred to as a “body fluid”. The blood sample may be any blood sample that is in a state where the number of cells can be counted and the cell types can be determined. Preferably, blood is peripheral blood. Examples of blood include peripheral blood collected using an anticoagulant agent such as ethylenediamine tetraacetate (sodium salt or potassium salt), heparin sodium, or the like. Peripheral blood may be collected from an artery or may be collected from a vein.
The cell types to be determined in the present embodiment are those according to the cell types based on morphological classification, and are different depending on the type of the specimen. When the specimen is blood and the blood is collected from a healthy individual, the cell types to be determined in the present embodiment include, for example, nucleated cell such as nucleated red blood cell and white blood cell, red blood cell, platelet, and the like. Nucleated cells include, for example, neutrophils, lymphocytes, monocytes, eosinophils, and basophils. Neutrophils include, for example, segmented neutrophils and band neutrophils. When blood is collected from an unhealthy individual, nucleated cells may include, for example, at least one type selected from the group consisting of immature granulocyte and abnormal cell. Such cells are also included in the cell types to be determined in the present embodiment. Immature granulocytes can include, for example, cells such as metamyelocytes, myelocytes, promyelocytes, and myeloblasts.
The nucleated cells may include, in addition to normal cells, abnormal cells that are not contained in peripheral blood of a healthy individual. Examples of abnormal cells are cells that appear when a person has a certain disease, and such abnormal cells are tumor cells, for example. In a case of the hematopoietic system, the certain disease can be a disease selected from the group consisting of, for example: myelodysplastic syndrome; leukemia such as acute myeloblastic leukemia, acute promyelocytic leukemia, acute myelomonocytic leukemia, acute monocytic leukemia, erythroleukemia, acute megakaryoblastic leukemia, acute myeloid leukemia, acute lymphocytic leukemia, lymphoblastic leukemia, chronic myelogenous leukemia, or chronic lymphocytic leukemia; malignant lymphoma such as Hodgkin’s lymphoma or non-Hodgkin’s lymphoma; and multiple myeloma.
Further, abnormal cells can include, for example, cells that are not usually observed in peripheral blood of a healthy individual, such as: lymphoblasts; plasma cells; atypical lymphocytes; reactive lymphocytes; erythroblasts, which are nucleated red blood cells, such as proerythroblasts, basophilic erythroblasts, polychromatic erythroblasts, orthochromatic erythroblasts, promegaloblasts, basophilic megaloblasts, polychromatic megaloblasts, and orthochromatic megaloblasts; megakaryocytes including micromegakaryocytes; and the like.
When the specimen is urine, the cell types to be determined in the present embodiment can include, for example, epithelial cell such as that of transitional epithelium and squamous epithelium, red blood cell, white blood cell, and the like. Examples of abnormal cells include, for example, bacteria, fungi such as filamentous fungi and yeast, tumor cells, and the like.
When the specimen is a body fluid that usually does not contain blood components, such as ascites, pleural effusion, or spinal fluid, the cell types can include, for example, red blood cell, white blood cell, and large cell. The large cell here means a cell that is separated from an inner membrane of a body cavity or a peritoneum of a viscus, and that is larger than white blood cells. For example, mesothelial cells, histiocytes, tumor cells, and the like correspond to the large cell.
When the specimen is bone marrow aspirate, the cell types to be determined in the present embodiment can include, as normal cells, mature blood cells and immature hematopoietic cells. Mature blood cells include, for example, nucleated cells such as nucleated red blood cells and white blood cells, red blood cells, platelets, and the like. Nucleated cells such as white blood cells include, for example, neutrophils, lymphocytes, plasma cells, monocytes, eosinophils, and basophils. Neutrophils include, for example, segmented neutrophils and band neutrophils. Immature hematopoietic cells include, for example, hematopoietic stem cells, immature granulocytic cells, immature lymphoid cells, immature monocytic cells, immature erythroid cells, megakaryocytic cells, mesenchymal cells, and the like. Immature granulocytes can include cells such as, for example, metamyelocytes, myelocytes, promyelocytes, myeloblasts, and the like. Immature lymphoid cells include, for example, lymphoblasts and the like. Immature monocytic cells include monoblasts and the like. Immature erythroid cells include, for example, nucleated red blood cells such as proerythroblasts, basophilic erythroblasts, polychromatic erythroblasts, orthochromatic erythroblasts, promegaloblasts, basophilic megaloblasts, polychromatic megaloblasts, and orthochromatic megaloblasts. Megakaryocytic cells include, for example, megakaryoblasts and the like.
Examples of abnormal cells that can be contained in bone marrow include, for example, hematopoietic tumor cells of a disease selected from the group consisting of: myelodysplastic syndrome; leukemia such as acute myeloblastic leukemia, acute promyelocytic leukemia, acute myelomonocytic leukemia, acute monocytic leukemia, erythroleukemia, acute megakaryoblastic leukemia, acute myeloid leukemia, acute lymphocytic leukemia, lymphoblastic leukemia, chronic myelogenous leukemia, or chronic lymphocytic leukemia; malignant lymphoma such as Hodgkin’s lymphoma or non-Hodgkin’s lymphoma; and multiple myeloma, which have been described above, and metastasized tumor cells of a malignant tumor developed in an organ other than bone marrow.
As the signal obtained from an analyte (e.g., cell or particle) in a specimen shown in the above example, a forward scattered light signal, a side scattered light signal, and a fluorescence signal, which are each an analog optical signal obtained through application of light to each cell flowing in a flow cell, are shown as examples. However, the signal is not limited in particular as long as the signal indicates a feature of each analyte and allows classification of analytes for each type.
Preferably, the optical signal is a light signal obtained as an optical response to application of light to a cell. The light signal can include at least one type selected from a signal based on light scattering, a signal based on light absorption, a signal based on transmitted light, and a signal based on fluorescence.
The signal based on light scattering can include a scattered light signal caused by light application and a light loss signal caused by light application. The scattered light signal represents a feature of an analyte in the specimen in accordance with the light reception angle of scattered light with respect to the advancing direction of application light. The forward scattered light signal is used in calculation of a representative value that indicates the size of the analyte. The side scattered light signal is used in calculation of a representative value that indicates, when the analyte in the specimen is a cell, complexity of the nucleus of the cell.
“Forward” of the forward scattered light means the advancing direction of light emitted from a light source. When the angle of application light is defined as 0°, “forward” can include a forward low angle at which the light reception angle is about 0° to 5°, and/or a forward high angle at which the light reception angle is about 5° to 20°. “Side” is not limited as long as the “side” does not overlap “forward”. When the angle of application light is defined as 0°, “side” can include a light reception angle being about 25° to 155°, preferably about 45° to 135°, and more preferably about 90°. Fluorescence in the present embodiment is detected at a light reception angle similar to that of side scattered light.
The signal based on light scattering may include polarized light or depolarized light as a component of the signal. For example, when scattered light caused by application of light to an analyte in the specimen is received through a polarizing plate, only scattered light polarized at a specific angle can be received. Meanwhile, when light is applied to an analyte in the specimen through a polarizing plate, and the resultant scattered light is received through a polarizing plate that allows passage therethrough of only polarized light having an angle different from that of the polarizing plate for light application, only depolarized scattered light can be received.
A light loss signal indicates the loss amount of received light based on decrease, of the received light amount at a light receiving part, which is caused by application of light to an analyte and scattering of the light. Preferably, the light loss signal is obtained as a light loss (axial light loss) in the optical axis direction of the application light. The light loss signal can be expressed as a proportion of the received light amount at the time of flowing of a measurement sample in the flow cell, when the received light amount at the light receiving part in a state where the measurement sample is not flowing in the flow cell is defined as 100%. Similar to the forward scattered light signal, the axial light loss is used in calculation of a representative value that indicates the size of the analyte, but the signal that is obtained differs depending on whether the analyte has translucency or not.
The signal based on fluorescence may be fluorescence that is excited as a result of application of light to an analyte labeled with a fluorescent substance, or may be an intrinsic fluorescence that is generated from a non-stained analyte. When the analyte in the specimen is a cell, the fluorescent substance may be a fluorescent dye that binds to nucleic acid or membrane protein, or may be a labeled antibody obtained by modifying, with a fluorescent dye, an antibody that binds to a specific protein of the cell.
The optical signal may be obtained in a form of image data obtained by applying light to an analyte in the specimen and capturing an image of the analyte to which the light has been applied. The image data can be obtained by capturing, with an imaging element such as a TDI camera or a CCD camera, an image of each individual analyte flowing in a flow path in a flow cell. Alternatively, a specimen or a measurement sample containing cells is applied, sprayed, or spot-applied on a slide glass, and an image of the slide glass is captured by an imaging element, whereby image data of cells may be obtained.
The signal obtained from an analyte in the specimen is not limited to an optical signal, and may be an electrical signal obtained from a cell. As for the electrical signal, for example, DC current is applied to the flow cell, and change in impedance caused by an analyte flowing in the flow cell may be used as the electrical signal. The thus obtained electrical signal is used in calculation of a representative value that reflects the volume of the analyte. Alternatively, the electrical signal may be the change in impedance at the time of application of a radio frequency to an analyte flowing in the flow cell. The thus obtained electrical signal is used in calculation of a representative value that reflects conductivity of the analyte.
The signal obtained from an analyte in the specimen may be a combination of at least two types of signals out of the above-described signals. Through combination of a plurality of signals, the features of an analyte can be pleiotropically analyzed, and thus, cell classification with a higher accuracy is enabled. As for the combination, for example, at least two out of a plurality of optical signals, e.g., a forward scattered light signal, a side scattered light signal, and a fluorescence signal, may be combined. Alternatively, scattered light signals having different angles, e.g., a low angle scattered light signal and a high angle scattered light signal, may be combined. Still alternatively, an optical signal and an electrical signal may be combined, and the type and number of the signals to be combined are not limited in particular.
The AI algorithm 60 used in the AI analysis in the present embodiment is a deep learning algorithm, for example. The deep learning algorithm is one of artificial intelligence algorithms, and configured as a neural network that includes a middle layer composed of multiple layers. Data inputted to the neural network is processed by a large number of matrix operations. From digital data obtained through A/D conversion of each analog optical signal shown as an example in
In the present embodiment, classifying the type of each analyte in the specimen is not limited to the classification performed by the AI algorithm 60. From individual analytes passing through a predetermined position in a flow path, a signal intensity is obtained, for each of the analytes, at a plurality of time points in a time period while the analyte is passing through the predetermined position, and on the basis of a result in which the obtained signal intensities at the plurality of time points regarding each individual analyte are recognized as a pattern, the type of each analyte may be determined. The pattern may be recognized as a numerical pattern of signal intensities at a plurality of time points, or may be recognized as a shape pattern obtained when signal intensities at a plurality of time points are plotted on a graph. When the pattern is recognized as a numerical pattern, if a numerical pattern of an analyte and a numerical pattern for which the type is already known are compared with each other, the type of the analyte can be determined. For the comparison between the numerical pattern of an analyte and a control numerical pattern, Spearman rank correlation, z-score, or the like can be used, for example. When the pattern of the graph shape of an analyte and the pattern of a graph shape for which the type is already known are compared with each other, the type of the analyte can be determined. For the comparison between the pattern of the graph shape of an analyte and the pattern of the graph shape for which the type is already known, geometric shape pattern matching may be used, or a feature descriptor represented by SIFT Descriptor may be used, for example.
A configuration example of the specimen analyzer 4000 when the measurement unit 400 includes an FCM detection part (detection part based on flow cytometry) for measuring a specimen (e.g., blood specimen, urine specimen, body fluid, bone marrow aspirate) will be described.
As shown in
The specimen suction part 450 suctions a specimen from a specimen container, and discharges the suctioned specimen into a reaction container (e.g., reaction chamber, reaction cuvette), for example. The sample preparation part 440 suctions a reagent for preparing a measurement sample, and discharges the reagent into the reaction container that contains the specimen, for example. The specimen and the reagent are mixed in the reaction container, whereby a measurement sample is prepared. The apparatus mechanism part 430 includes mechanisms in the measurement unit 400.
Light emitted from a light source 4111 is applied via a light application lens system 4112 to each analyte in a measurement sample passing through a flow cell (sheath flow cell) 4113. Accordingly, scattered light and fluorescence are emitted from the analyte flowing in the flow cell 4113.
The wavelength of light emitted from the light source 4111 is not limited in particular, and a wavelength suitable for excitation of the fluorescent dye is selected. As the light source 4111, for example, a semiconductor laser light source, a gas laser light source such as an argon laser light source or a helium-neon laser, a mercury arc lamp, or the like is used. In particular, a semiconductor laser light source is suitable because the semiconductor laser light source is very inexpensive when compared with a gas laser light source.
Forward scattered light generated from an analyte in the flow cell 4113 is received by a light receiving element 4116 via a condenser lens 4114 and a pin hole part 4115. The light receiving element 4116 is a photodiode, for example. Side scattered light generated from the analyte in the flow cell 4113 is received by a light receiving element 4121 via a condenser lens 4117, a dichroic mirror 4118, a bandpass filter 4119, and a pin hole part 4120. The light receiving element 4121 is a photodiode, for example. Fluorescence generated from the analyte in the flow cell 4113 is received by a light receiving element 4122 via the condenser lens 4117 and the dichroic mirror 4118. The light receiving element 4122 is an avalanche photodiode, for example. As the light receiving elements 4116, 4121, 4122, a photomultiplier may be used.
The analog reception light signals (optical signals) outputted from the respective light receiving elements 4116, 4121, 4122 are inputted to the analog processing part 420 via amplifiers 4151, 4152, 4153, respectively.
The analog processing part 420 performs processes such as noise removal and smoothing on the optical signals inputted from the FCM detection part 410, and outputs the processed optical signals to an A/D converter 461.
With reference back to
The A/D converter 461 converts each analog optical signal that is from measurement start to measurement end of the measurement sample and that has been outputted from the analog processing part 420, into digital data. When a plurality of types of optical signals (e.g., optical signals respectively corresponding to forward scattered light intensity, side scattered light intensity, and fluorescence intensity) are generated from a single measurement sample, the A/D converter 461 converts each optical signal from measurement start to measurement end into digital data. For example, as shown in
The A/D converter 461 compares the signal level of each optical signal and a predetermined threshold, and samples the optical signal that has a signal level higher than the threshold. The A/D converter 461 samples the analog optical signal at a predetermined sampling rate (e.g., sampling at 1024 points at a 10 nanosecond interval, sampling at 128 points at an 80 nanosecond interval, sampling at 64 points at a 160 nanosecond interval, or the like). By executing sampling processes on the three types of optical signals that correspond to each analyte, for example, the A/D converter 461 generates digital data (waveform data) of the forward scattered light signal, digital data (waveform data) of the side scattered light signal, and digital data (waveform data) of the fluorescence signal for each analyte. Each piece of digital data (waveform data) corresponds to one analyte in the specimen.
The A/D converter 461 provides an index to each piece of the generated waveform data. The generated pieces of waveform data are pieces of digital data that respectively correspond to N analytes contained in one specimen, for example. Accordingly, for each analyte, three types of waveform data are generated so as to correspond to the three types of optical signals (forward scattered light signal, side scattered light signal, and fluorescence signal).
The waveform data generated by the A/D converter 461 is transmitted to the analysis unit 300 via the IF parts 462, 465 and the bus 463. The apparatus mechanism part 430, the sample preparation part 440, and the specimen suction part 450 are controlled by the analysis unit 300 via the IF parts 464, 465 and the bus 463.
The analysis unit 300 includes a processor 3001, a RAM 3017, a bus 3003, a storage 3004, an IF part 3006, a display part 3011, and an operation part 3012. The analysis unit 300 is implemented by a personal computer, for example. The analysis unit 300 is connected to the measurement unit 400 via the IF part 3006.
The processor 3001 is implemented by a CPU, for example. The processor 3001 executes a program developed on the RAM 3017 from the storage 3004. The RAM 3017 is a so-called main memory. The processor 3001 executes a program for analysis, thereby analyzing waveform data obtained in the measurement unit 400. The processor 3001 executes a program for control, thereby controlling the analysis unit 300 and the measurement unit 400.
The storage 3004 is implemented by a hard disk drive (HDD) or a solid state drive (SSD), for example. The storage 3004 stores waveform data received from the measurement unit 400, a program for controlling the analysis unit 300 and the measurement unit 400, and a program for analyzing waveform data. The program for analyzing waveform data is configured to analyze waveform data on the basis of the calculation processing analysis and the AI analysis described above. The storage 3004 stores a rule for specifying waveform data to serve as a target of each of the AI analysis and the calculation processing analysis, and a rule for selecting an analysis operation.
The display part 3011 is implemented by a liquid crystal display, for example. The display part 3011 is connected to the processor 3001 via the bus 3003 and the IF part 3006. On the display part 3011, an analysis result obtained in the measurement unit 400 is displayed, for example.
The operation part 3012 is implemented by a pointing device and the like including a keyboard, a mouse, and a touch panel, for example. The user such as a doctor or a laboratory technician operates the operation part 3012, to input a measurement order to the specimen analyzer 4000, thereby being able to input a measurement instruction based on the measurement order. By operating the operation part 3012, the user can also input an instruction to display an analysis result. The analysis result includes, for example, a numerical value result, a graph, and a chart that are based on the analysis, flag information provided to the specimen, and the like.
The measurement unit 400 in
The RBC/PLT detection part 4101 is a detection part of an electrical resistance type, and performs measurement of blood cells according to a sheath flow DC detection method on the basis of an RBC/PLT measurement sample. The HGB detection part 4102 performs measurement of hemoglobin according to an SLS-hemoglobin method on the basis of a hemoglobin measurement sample. Data obtained through A/D conversion of an analog signal obtained from each of the RBC/PLT detection part 4101 and the HGB detection part 4102 serves as a target of the calculation processing analysis. From the data based on the RBC/PLT detection part 4101, red blood cells and platelets in the blood specimen are counted. From the data based on the HGB detection part 4102, the hemoglobin content in the blood specimen is obtained.
The data obtained through A/D conversion of the analog signal obtained from each of the RBC/PLT detection part 4101 and the HGB detection part 4102 may serve as a target of the AI analysis. Alternatively, with respect to the data based on the RBC/PLT detection part 4101 and the HGB detection part 4102 as well, the AI analysis and the calculation processing analysis may be selectively used. Accordingly, the load on the analysis unit 300 which processes data can be reduced.
The specimen suction part 450 includes: a nozzle 451 for suctioning a blood specimen (e.g., whole blood) from a collection tube TB; and a pump 452 for providing a negative pressure and a positive pressure to the nozzle. The nozzle 451 is moved upwardly and downwardly by the apparatus mechanism part 430 (see
The sample preparation part 440 includes a WDF sample preparation part 440a, a RET sample preparation part 440b, a WPC sample preparation part 440c, a PLT-F sample preparation part 440d, and a WNR sample preparation part 440e. The sample preparation parts 440a to 440e each include a reaction chamber for mixing a specimen and a reagent (e.g., hemolytic agent and staining liquid). The sample preparation parts 440a to 440e are used in a WDF channel, a RET channel, a WPC channel, a PLT-F channel, and a WNR channel, respectively.
Here, the specimen analyzer 4000 includes a plurality of measurement channels so as to respectively correspond to a plurality of types of measurement samples that are prepared. The specimen analyzer 4000 includes the WDF channel, the RET channel, the WPC channel, the PLT-F channel, and the WNR channel, for example. The WDF channel is a channel for detecting neutrophils, lymphocytes, monocytes, and eosinophils. The RET channel is a channel for detecting reticulocytes. The WPC channel is a channel for detecting blasts and lymphocytic abnormal cells. The PLT-F channel is a channel for detecting platelets. The WNR channel is a channel for detecting white blood cells other than basophils, basophils, and nucleated red blood cells.
The sample preparation parts 440a to 440e each have connected thereto, via flow paths, a hemolytic agent container containing a hemolytic agent being a reagent corresponding to the measurement channel, and a staining liquid container containing a staining liquid being a reagent corresponding to the measurement channel. For example, the WDF sample preparation part 440a has connected thereto, via flow paths, a hemolytic agent container containing a WDF hemolytic agent (e.g., Lysercell WDF II; manufactured by Sysmex Corporation) being a WDF measurement reagent, and a staining liquid container containing a WDF staining liquid (e.g., Fluorocell WDF; manufactured by Sysmex Corporation) being a WDF measurement reagent. Here, a configuration example in which one sample preparation part is connected to a hemolytic agent container and a staining liquid container is shown. However, one sample preparation part need not necessarily be connected to both of a hemolytic agent container and a staining liquid container, and one reagent container may be used in common by a plurality of sample preparation parts. In addition, each sample preparation part and the corresponding reagent container need not be connected by a flow path. A configuration in which a reagent is suctioned by a nozzle from a reagent container, the nozzle is moved, and the suctioned reagent is discharged from the nozzle into a reaction chamber of the sample preparation part, may be adopted.
Through horizontal and up-down movement by the apparatus mechanism part 430, the nozzle 451 having suctioned a blood specimen is positioned above a reaction chamber of a sample preparation part that corresponds to a measurement order, among the sample preparation parts 440a to 440e. In this state, when the pump 452 provides a positive pressure, the blood specimen is discharged from the nozzle 451 to the corresponding reaction chamber. The sample preparation part 440 supplies a hemolytic agent and a staining liquid that correspond to the reaction chamber having discharged therein the blood specimen, and mixes the blood specimen, the hemolytic agent, and the staining liquid in the reaction chamber, thereby preparing a measurement sample.
A WDF measurement sample is prepared in the WDF sample preparation part 440a, a RET measurement sample is prepared in the RET sample preparation part 440b, a WPC measurement sample is prepared in the WPC sample preparation part 440c, a PLT-F measurement sample is prepared in the PLT-F sample preparation part 440d, and a WNR measurement sample is prepared in the WNR sample preparation part 440e. Each prepared measurement sample is supplied from the reaction chamber to the FCM detection part 410 via a flow path, and measurement of cells by flow cytometry is performed.
The measurement channels (WDF, RET, WPC, PLT-F, WNR) described above correspond to measurement items included in a measurement order. For example, the WDF channel corresponds to a measurement item regarding classification of white blood cells, the RET channel corresponds to a measurement item regarding reticulocytes, the PLT-F channel corresponds to a measurement item regarding platelets, and the WNR channel corresponds to a measurement item regarding the number of white blood cells and nucleated red blood cells. The measurement samples prepared in the measurement channels described above are measured by the FCM detection part 410.
The measurement result by the RBC/PLT detection part 4101 corresponds to a measurement item regarding the number of red blood cells. The measurement result by the HGB detection part 4102 corresponds to a measurement item regarding the hemoglobin content.
In the example shown in
When the sample preparation part 440 is configured as shown in
According to the configuration in
In the configuration in
In the configuration in
With respect to the process described with reference to
In
In step S101, the analysis unit 300 refers to a rule that includes which of the AI analysis and the calculation processing analysis is performed on the basis of a measurement channel, and on the basis of the rule referred to, the analysis unit 300 specifies waveform data to serve as a target of the AI analysis and waveform data to serve as a target of the calculation processing analysis, with respect to the waveform data obtained in step S11.
The screen in
With reference back to
In the case of
When the WDF channel is set to be a target of the AI analysis, the AI analysis may be executed with respect to all of the measurement items that correspond to the WDF channel, or the AI analysis may be executed with respect to some measurement items that correspond to the WDF channel and the calculation processing analysis may be executed with respect to the other measurement items.
Next, using an example shown in
As shown in the drawing in the upper part of
The drawing in the middle part of
As shown in the drawing in the middle part of
Here, for convenience, a start point and an end point are set to an analog optical signal and waveform data is obtained. However, as described above, after all of the optical signal has been converted into digital data, a start point and an end point may be set to the digital data, and waveform data may be obtained.
The drawing in the lower part of
Each piece of waveform data generated by the A/D converter 461 may be provided with an index for identifying the corresponding analyte. As the indexes, for example, integers of 1 to N are provided in the sequential order of the generated pieces of waveform data, and the waveform data of forward scattered light, the waveform data of side scattered light, and the waveform data of fluorescence that have been obtained from the same analyte are each provided with an identical index. Since an identical index is provided to the pieces of waveform data that correspond to the same analyte, the AI algorithm 60 can analyze, as one set, the waveform data of forward scattered light, the waveform data of side scattered light, and the waveform data of fluorescence that correspond to each individual analyte, and can classify the type of the analyte.
As a result of measurement of an analyte according to flow cytometry, an optical signal 70a corresponding to forward scattered light, an optical signal 70b corresponding to side scattered light, and an optical signal 70c corresponding to fluorescence are obtained from the analyte. Pieces of waveform data 72a, 72b, 72c corresponding to the analyte are obtained on the basis of the optical signals 70a, 70b, 70c, respectively. As for the training data 75, for example, it is possible to use waveform data 72a, 72b, 72c of an analyte that has been determined, as a result of performing the calculation processing analysis on analytes in a specimen measured according to flow cytometry, to have a high possibility of being a specific type.
Hereinafter, an example in which the specimen analyzer 4000 serving as a blood cell counter that analyzes a blood specimen is used, will be described.
An operator measures a blood specimen by the FCM detection part 410, and accumulates waveform data of forward scattered light, side scattered light, and fluorescence of each individual analyte contained in the specimen. Subsequently, for example, on the basis of the peak value of the waveform data based on side scattered light and the peak value of the waveform data based on fluorescence, the operator classifies each analyte (cell) in the specimen into a group of neutrophil, lymphocyte, monocyte, eosinophil, basophil, immature granulocyte, or abnormal cell. The operator provides a label value 77 that corresponds to the classified cell type, to the waveform data of the cell, thereby obtaining the training data 75. Since the training data 75 is generated for each cell type, the label value 77 is different in accordance with the cell type, as shown in
At this time, the operator obtains the mode, the average value, or the median of the peak values of the pieces of waveform data based on side scattered light and fluorescence of cells included in the neutrophil group, specifies a representative cell on the basis of the value, and provides waveform data of the specified cell with a label value “1” which corresponds to neutrophil.
The generation method of the training data 75 is not limited thereto. For example, the operator collects only specific cells by a cell sorter, measures each cell according to flow cytometry, and provides the obtained waveform data with a label value for the cell, whereby the operator may obtain the training data 75.
The pieces of the waveform data 72a, 72b, 72c are each combined with a label value 77 that indicates the type of the cell being the source of the data. The training data 75 includes the three pieces of waveform data (waveform data based on the optical signals 70a, 70b, 70c) that correspond to each cell in a state of being associated with each other. Then, the training data 75 is inputted to the AI algorithm 50.
An outline of training of a neural network will be described using
The AI algorithm 50 is configured as a neural network that includes a middle layer composed of multiple layers. The neural network in this case is a convolutional neural network having a convolution layer, for example. The number of nodes of an input layer 50a in the neural network corresponds to the number of elements of the array included in the waveform data 72a, 72b, 72c of the training data 75 to be inputted. The number of elements of the array is equal to the sum of the number of elements of the waveform data 72a, 72b, 72c of forward scattered light, side scattered light, and fluorescence that correspond to one analyte.
In the example in
In the analysis method of waveform data shown as an example in
Preferably, the analysis data 85 and the training data 75 at least have the same obtaining condition. The obtaining condition includes conditions for measuring an analyte in a specimen according to flow cytometry, e.g., a preparation condition for a measurement sample, the flow speed at which the measurement sample is caused to flow in a flow cell, the intensity of light applied to the flow cell, the amplification factor at light receiving parts that receive scattered light and fluorescence, and the like. The obtaining condition further includes a sampling rate at the time of performing A/D conversion on an analog optical signal.
The analysis data 85 includes the three pieces of waveform data (waveform data based on the optical signals 80a, 80b, 80c) that correspond to each analyte, in a state of being associated with each other. Then, the analysis data 85 is inputted to the trained AI algorithm 60. The AI algorithm 60 is configured as a neural network that includes a middle layer composed of multiple layers.
When the analysis data 85 has been inputted to an input layer 60a of the neural network forming the AI algorithm 60, classification information 82 regarding the type of the analyte that corresponds to the analysis data 85 is outputted from an output layer 60b. A middle layer 60c is positioned between the input layer 60a and the output layer 60b. The classification information 82 includes a probability at which the analyte corresponds to each of a plurality of types. Further, the type having the highest probability is determined to be the type to which the analyte belongs, and a label value 83 being an identifier representing the type and an analysis result 84 being a character string representing the type are outputted.
In the example in
The analysis method for waveform data in the example shown in
In the case of the example shown in
When having been classified as a predetermined cell through the calculation processing analysis, the cell is identified by an index for the waveform data in
The analysis method for waveform data described with reference to
In the analysis method described with reference to
In step S111, the measurement unit 400 obtains optical signals from a measurement sample prepared in the WDF channel, and obtains waveform data from each obtained optical signal. In step S112, the analysis unit 300 executes the AI analysis on the waveform data obtained in step S111. In step S113, the analysis unit 300 provides an analysis result of the waveform data of the WDF channel, and analysis results of waveform data of other channels in combination. How the analyses on the waveform data of the other channels are apportioned between and executed by the AI analysis and the calculation processing analysis is determined on the basis of one of the rules shown as examples in the embodiments described above, for example.
In another analysis method based on the configuration shown in
The analysis unit 300 executes the calculation processing analysis on waveform data that corresponds to cells that have been classified as neither nucleated red blood cells nor basophils. For example, the peak value of waveform data that corresponds to each cell that has been classified as neither a nucleated red blood cell nor a basophil is extracted, the cell type is classified on the basis of a two-dimensional graph (scattergram) generated from peak values that correspond to side scattered light and peak values that correspond to fluorescence. For example, on the basis of the two-dimensional graph, which of an eosinophil, a neutrophil, a lymphocyte, a monocyte, and other than these the cell is, is classified. A cell that has been classified as other than an eosinophil, a neutrophil, a lymphocyte, and a monocyte through the analysis based on the two-dimensional graph is classified as debris, for example.
In step S121, the measurement unit 400 obtains optical signals from a measurement sample prepared in the WDF channel, and obtains waveform data from each obtained optical signal. In step S122, the analysis unit 300 executes the AI analysis on the waveform data obtained in step S121. Accordingly, nucleated red blood cells and basophils are classified. In step S123, the analysis unit 300 specifies waveform data that corresponds to cells that are classified as neither nucleated red blood cells nor basophils.
In step S124, the analysis unit 300 executes the calculation processing analysis on the waveform data specified in step S123. Accordingly, lymphocytes, monocytes, eosinophils, and neutrophils are classified. In step S125, the analysis unit 300 provides the analysis result of the waveform data of the WDF channel and analysis results of the waveform data of other channels in combination.
In another analysis method based on the configuration shown in
For example, from the counting result of cells classified as either neutrophils or basophils through the calculation processing analysis, a counting result of cells classified as basophils through the AI analysis is subtracted, whereby a counting result of neutrophils and a counting result of basophils are calculated. Cells that have been classified as neither nucleated red blood cells nor basophils through the AI analysis are classified as debris, for example.
In step S131, the measurement unit 400 obtains optical signals from a measurement sample prepared in the WDF channel, and obtains waveform data from each obtained optical signal. In step S132, the analysis unit 300 executes the calculation processing analysis on the waveform data obtained in step S131. Accordingly, groups of lymphocytes, monocytes, eosinophils, and neutrophils and basophils are classified. In step S133, the analysis unit 300 specifies waveform data that corresponds to (1) cells that have been classified as none of lymphocytes, monocytes, eosinophils, and neutrophils or basophils, and (2) cells that have been classified as neutrophils or basophils.
In step S134, the analysis unit 300 executes the AI analysis on the waveform data specified in step S133. Accordingly, neutrophils and basophils are classified. In step S135, the analysis unit 300 provides an analysis result of waveform data of the WDF channel and analysis results of waveform data of other channels in combination.
In Embodiment 5, a detailed configuration example in which, in the specimen analyzer 4000 that analyzes coagulability of a blood specimen, analysis is executed by being apportioned between the calculation processing analysis and the AI analysis is shown.
An example of a specimen to be measured by the specimen analyzer 4000 of Embodiment 5 is a biological sample collected from a subject. The specimen can include whole blood, plasma, and the like, for example. The specimen analyzer 4000 of Embodiment 5 analyzes the presence or absence of an abnormality due to an interference substance in a specimen, on the basis of a coagulation method, a synthetic substrate method, an immunonephelometry, an agglutination method, a chemiluminescent enzyme immunoassay (CLEIA), or the like. Similar to the configuration example of Embodiment 1 shown in
When compared with the measurement unit 400 shown in
The detection part 470 includes a light source part 471 and a detection block 476. The light source part 471 includes a halogen lamp, for example. The light source part 471 is configured to be able to emit, for example, light having a wavelength of 660 nm for blood coagulation time measurement, light having a wavelength of 405 nm for synthetic substrate measurement, and light having a wavelength of 800 nm for immunonephelometric measurement. The sample preparation part 440 mixes a specimen with a blood coagulation reagent to prepare a measurement sample. The detection part 470 applies light from the light source part 471 to the measurement sample composed of the blood coagulation reagent and the specimen, and detects light transmitted through the specimen. The detection part 470 may apply light from the light source part 471 to the measurement sample and detect light scattered by the specimen.
The controller 466 is implemented by an FPGA, for example. The controller 466 is connected to the analysis unit 300 via the bus 463 and the IF part 465. The controller 466 controls each component of the measurement unit 400 on the basis of an instruction from the analysis unit 300.
The detection block 476 includes a holding part 472, an optical fiber 473, a condenser lens 474, and a light receiving part 475.
A reaction container C1 containing a measurement sample prepared from a specimen and a reagent corresponding to a measurement item is held in the holding part 472. Accordingly, the measurement sample is left to stand. Light emitted from the light source part 471 (see
With reference back to
In step S141, the measurement unit 400 obtains an optical signal in the detection part 470 and obtains coagulation waveform data from the obtained optical signal. In step S142, the analysis unit 300 executes the calculation processing analysis on the coagulation waveform data obtained in step S141. For example, as described with reference to
In step S143, the analysis unit 300 executes the AI analysis on the coagulation waveform data obtained in step S141. Accordingly, the analysis unit 300 obtains information on the presence or absence of an abnormality regarding the measurement, on the basis of the feature amount extracted by the AI algorithm 60 from the coagulation waveform data. On the basis of the presence or absence of an abnormality regarding the measurement, the analysis unit 300 determines whether or not there is a suspected occurrence of nonspecific reaction.
In step S144, the analysis unit 300 provides the result that indicates the time for the blood specimen to coagulate obtained in step S142, and the result that indicates the presence or absence of an abnormality regarding the measurement obtained in step S143.
In
Although the specimen analyzer 4000 of Embodiment 5 is a blood coagulation measurement apparatus that optically measures change in the turbidity, of a measurement sample, that is associated with coagulation of a blood specimen, the present disclosure is not limited thereto. For example, the specimen analyzer 4000 of Embodiment 5 may be a blood coagulation measurement apparatus that measures change, in oscillation of a steel ball in a measurement sample, that is associated with change, in the viscosity of the measurement sample, that is associated with coagulation of the blood specimen, on the basis of the received frequency of a high frequency transmitted from a high frequency transmission coil. Although the specimen analyzer 4000 of Embodiment 5 is a blood coagulation measurement apparatus, the specimen analyzer 4000 of Embodiment 5 may be an immunoassay apparatus, a biochemical measurement apparatus, or a gene measurement apparatus.
In Embodiment 6, a configuration example of the specimen analyzer 4000 that includes a host processor and a parallel-processing processor is shown. In Embodiment 6, in a parallel-processing processor 3002, parallel processing is executed on waveform data, and on the basis of the result of the parallel processing, information regarding the type of each of analytes is generated.
According to Embodiment 6, even when data having a huge volume of several hundred megabytes to several gigabytes per specimen is analyzed, processing regarding waveform data can be executed in parallel by the parallel-processing processor provided separately from the host processor. Therefore, for example, even when data having a huge volume is processed by the AI algorithm 60, processing of the data is concluded in the specimen analyzer 4000. Therefore, for example, it is not necessary to transmit data via the Internet or an intranet to an analysis server that stores the AI algorithm 60. Therefore, according to Embodiment 6, it is not necessary to transmit a large volume of data from the specimen analyzer 4000 to the analysis server, and to obtain an analysis result returning from the analysis server. Thus, the throughput of the specimen analyzer 4000 can be maintained at a high level while the classification accuracy of analytes in the specimen is improved.
With reference to
The specimen analyzer 4000 of Embodiment 6 includes the measurement unit 400, and the analysis unit 300 provided in the measurement unit 400. In the measurement unit 400 of Embodiment 6, when compared with the measurement unit 400 of Embodiment 4 shown in
When compared with the analysis unit 300 of Embodiment 4 shown in
The parallel-processing processor 3002 is configured to be able to process, instead of a master processor, arithmetic processes by the AI algorithm 60. By using the parallel-processing processor 3002 suitable for the processes of a matrix operation executed by the AI algorithm 60, it is possible to improve the TAT necessary for the AI analysis. However, although the TAT is improved by the parallel-processing processor 3002, if the data amount of the analysis target is increased, the computer load necessary for the AI analysis is increased. With regard to this, as described above, since data analysis is apportioned between the calculation processing analysis and the AI analysis, the computer load can be reduced, and improvement of the test efficiency can be realized.
Using the parallel-processing processor 3002, the processor 3001 executes an analysis process on waveform data by the AI algorithm 60. That is, the processor 3001 executes analysis software 3100, thereby executing the AI analysis of waveform data based on the AI algorithm 60. The analysis software 3100 is used in order to analyze waveform data that corresponds to each analyte in a specimen, on the basis of the AI algorithm 60.
The analysis software 3100 may be stored in the storage 3004. In this case, the processor 3001 executes the analysis software 3100 stored in the storage 3004, thereby executing the AI analysis of waveform data based on the AI algorithm 60.
In the present embodiment, for example, the AI analysis is executed by the processor 3001 and the parallel-processing processor 3002, and the calculation processing analysis is executed by the processor 3001 without using the parallel-processing processor 3002.
The processor 3001 is a CPU (Central Processing Unit), for example. For example, Core i9, Core i7, or Core i5 manufactured by Intel Corporation, or Ryzen 9, Ryzen 7, Ryzen 5, or Ryzen 3 manufactured by AMD, or the like may be used as the processor 3001.
The processor 3001 controls the parallel-processing processor 3002. The parallel-processing processor 3002 executes parallel processing regarding, for example, a matrix operation in accordance with control by the processor 3001. That is, the processor 3001 is a master processor of the parallel-processing processor 3002, and the parallel-processing processor 3002 is a slave processor of the processor 3001. The processor 3001 is also referred to as a host processor or a main processor. The processor 3001 executes the matrix operation according to the AI algorithm 60, through parallel processing performed by the parallel-processing processor 3002.
The parallel-processing processor 3002 executes in parallel, a plurality of arithmetic processes being at least a part of processes regarding analysis of waveform data. The parallel-processing processor 3002 is a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), or an ASIC (Application Specific Integrated Circuit), for example. When the parallel-processing processor 3002 is an FPGA, the parallel-processing processor 3002 may have programed therein in advance an arithmetic process regarding the trained AI algorithm 60, for example. When the parallel-processing processor 3002 is an ASIC, the parallel-processing processor 3002 may have incorporated therein in advance a circuit for executing the arithmetic process regarding the trained AI algorithm 60, or may have built therein a programmable module in addition to such an incorporated circuit, for example.
As the parallel-processing processor 3002, GeForce, Quadro, TITAN, Jetson, or the like manufactured by NVIDIA Corporation may be used, for example. In the case of the Jetson series, Jetson Nano, Jetson Tx2, Jetson Xavier, or Jetson AGX Xavier is used, for example.
The processor 3001 executes calculation processing regarding control of the measurement unit 400, for example. The processor 3001 executes calculation processing regarding control signals transmitted/received between the apparatus mechanism part 430, the sample preparation part 440, and the specimen suction part 450, for example. The processor 3001 executes calculation processing regarding transmission/reception of information to/from the computer 301, for example.
The computer 301 has a function of displaying an analysis result transmitted from the analysis unit 300 on the basis of the processing performed by the processor 3001, for example. The computer 301 transmits a measurement order to the analysis unit 300, for example. The measurement order is transmitted from a host computer to the computer 301, for example. It is also possible for the user to input a measurement order via an input device of the computer 301.
The processor 3001 executes processes regarding reading-out of program data from the storage 3004, developing a program onto the RAM 3017, and transmission/reception of data to/from the RAM 3017, for example. The above-described processes executed by the processor 3001 are required to be executed in a predetermined sequential order, for example. For example, when processes needed for control of the apparatus mechanism part 430, the sample preparation part 440, and the specimen suction part 450 are assumed to be A, B, and C, respectively, the processes are required to be executed in a sequential order of B, A, and C, in some cases. The processor 3001 often executes such continuous processes that depend on a sequential order, and thus, even when the number of arithmetic units (each may be referred to as a “processor core”, a “core”, or the like) is increased, the processing speed is not always increased.
Meanwhile, the parallel-processing processor 3002 executes a large number of regular calculation processes such as arithmetic operations on matrix data including a large number of elements, for example. In the present embodiment, the parallel-processing processor 3002 executes parallel processing in which at least a part of processes of analyzing waveform data in accordance with the AI algorithm 60 are parallelized. The AI algorithm 60 includes a large number of matrix operations, for example. For example, the AI algorithm 60 may include at least 100 matrix operations, or may include at least 1000 matrix operations.
The parallel-processing processor 3002 has a plurality of arithmetic units, and the respective arithmetic units can simultaneously execute matrix operations. That is, the parallel-processing processor 3002 can execute, in parallel, matrix operations by a plurality of respective arithmetic units, as parallel processing. For example, a matrix operation included in the AI algorithm 60 can be divided into a plurality of arithmetic processes that are not dependent on a sequential order with each other. The thus divided arithmetic processes can be executed in parallel by a plurality of arithmetic units, respectively. These arithmetic units may be each referred to as a “processor core”, a “core”, or the like.
As a result of execution of such parallel processing, speed up of arithmetic processing in the entirety of the specimen analyzer 4000 can be realized. A process such as a matrix operation included in the AI algorithm 60 may be referred to as “Single Instruction Multiple Data (SIMD) processing”, for example. The parallel-processing processor 3002 is suitable for such an SIMD operation, for example. Such a parallel-processing processor 3002 may be referred to as a vector processor.
As described above, the processor 3001 is suitable for executing diverse and complicated processes. Meanwhile, the parallel-processing processor 3002 is suitable for executing in parallel a large number of regular processes. Through parallel execution of a large number of regular processes, the TAT required for calculation processing is shortened.
The parallel processing to be executed by the parallel-processing processor 3002 is not limited to matrix operations. For example, when the parallel-processing processor 3002 executes a learning process with respect to the AI algorithm 50, differential operations or the like regarding the learning process can be the target of the parallel processing.
As for the number of arithmetic units of the processor 3001, a dual core (the number of cores: 2), a quad core (the number of cores: 4), or an octa core (the number of cores: 8) is adopted, for example. Meanwhile, the parallel-processing processor 3002 has, for example, at least ten arithmetic units (the number of cores: 10), and can execute in parallel ten matrix operations. The parallel-processing processor 3002 that has several ten arithmetic units also exists. The parallel-processing processor 3002 that has, for example, at least 100 arithmetic units (the number of cores: 100) and that can execute in parallel 100 matrix operations also exists. The parallel-processing processor 3002 that has several hundred arithmetic units also exists. The parallel-processing processor 3002 that has, for example, at least 1000 arithmetic units (the number of cores: 1000) and that can execute in parallel 1000 matrix operations also exists. The parallel-processing processor 3002 that has several thousand arithmetic units also exists.
When compared with the specimen analyzer 4000 in
The parallel-processing processor 3002 includes a plurality of arithmetic units 3200 and a RAM 3201. The respective arithmetic units 3200 execute arithmetic processes on matrix data in parallel. The RAM 3201 stores data regarding arithmetic processes executed by the arithmetic units 3200. The RAM 3201 is a memory that has a capacity of at least 1 gigabyte. The RAM 3201 may be a memory that has a capacity of 2 gigabytes, 4 gigabytes, 6 gigabytes, 8 gigabytes, 10 gigabytes, or more. Each arithmetic unit 3200 obtains data from the RAM 3201 and executes an arithmetic process. The arithmetic unit 3200 may be referred to as a “processor core”, a “core”, or the like.
In the example shown in
In the example shown in
A plurality of USB devices each having the parallel-processing processor 3002 mounted thereon may be connected to the IF part 467, whereby a plurality of the parallel-processing processors 3002 may be installed to the specimen analyzer 4000. The parallel-processing processor 3002 mounted on one USB device has a smaller number of arithmetic units 3200 than a GPU or the like in some cases. Therefore, if a plurality of USB devices are connected to the measurement unit 400, scale-up of the number cores can be realized.
Next, with reference to
The parallel-processing processor 3002 includes a plurality of the arithmetic units 3200 and the RAM 3201. The processor 3001, which executes the analysis software 3100, issues an order to the parallel-processing processor 3002, and causes the parallel-processing processor 3002 to execute at least a part of arithmetic processes necessary for analysis of waveform data according to the AI algorithm 60. The processor 3001 orders the parallel-processing processor 3002 to execute arithmetic processes regarding waveform data analysis based on the AI algorithm 60.
All or at least a part of waveform data is stored in the RAM 3017. The data stored in the RAM 3017 is transferred to the RAM 3201 of the parallel-processing processor 3002 by a DMA (Direct Memory Access) method, for example. The plurality of arithmetic units 3200 of the parallel-processing processor 3002 respectively execute, in parallel, arithmetic processes with respect to the data stored in the RAM 3201. Each of the plurality of arithmetic units 3200 obtains necessary data from the RAM 3201, to execute an arithmetic process. Data corresponding to the arithmetic result is stored into the RAM 3201 of the parallel-processing processor 3002. The data corresponding to the arithmetic result is transferred from the RAM 3201 to the RAM 3017 by a DMA method, for example.
Prior to analyzing waveform data in accordance with the AI algorithm 60, calculation of the product of a matrix (matrix operation) is executed. The parallel-processing processor 3002 executes in parallel a plurality of arithmetic processes regarding the matrix operation, for example.
The drawing in the upper part of
As shown in
Through the arithmetic operations shown as an example in
The arithmetic operations of the probability at which an analyte in the specimen belongs to each of a plurality of classification types may be performed by a processor different from the parallel-processing processor 3002. For example, the arithmetic results by the parallel-processing processor 3002 may be transferred from the RAM 3201 to the RAM 3017, and on the basis of the arithmetic results read out from the RAM 3017, the processor 3001 may perform arithmetic operations of the information regarding the probability at which the analyte corresponding to each piece of waveform data belongs to each of a plurality of classification types. Alternatively, the arithmetic results by the parallel-processing processor 3002 may be transferred from the RAM 3201 to the analysis unit 300, and a processor installed in the analysis unit 300 may perform arithmetic operations of the information regarding the probability at which the analyte corresponding to each piece of waveform data belongs to each of a plurality of classification types.
The processes shown in
The drawing in the upper part of
The drawing in the lower part of
In Formula 1, the suffixes of x are variables that indicate the row number and the column number of the waveform data. The suffixes of h are variables that indicate the row number and the column number of the filter. In the example shown in
The parallel-processing processor 3002 executes in parallel the matrix operation represented by Formula 1, by means of the plurality of respective arithmetic units 3200. On the basis of the arithmetic processes executed by the parallel-processing processor 3002, classification information regarding the type of each analyte in the specimen is generated. The generated classification information is used in generation and display of a test result of the specimen based on the classification information.
As shown in
The computer 301 may include an operation part implemented by a pointing device including a keyboard, a mouse, or a touch panel. The user such as a doctor or a laboratory technician operates the operation part to input a measurement order to the specimen analyzer 4000, thereby being able to input a measurement instruction in accordance with the measurement order. The user can input an instruction for displaying a test result, to the computer 301 via the operation part. By operating the operation part, the user can view various types of information regarding the test result, such as a numerical value result, a graph, a chart, and flag information provided to the specimen that are based on the analysis.
With reference to
In step S200, when the processor 3001 of the analysis unit 300 has received a measurement order, the processor 3001 instructs the measurement unit 400 to execute measurement. For example, through the instruction issued to the measurement unit 400, the analysis unit 300 controls operation of each detection part (the FCM detection part 410, the RBC/PLT detection part 4101, the HGB detection part 4102), the specimen suction part 450, and the sample preparation part 440 of the measurement unit 400. The measurement unit 400 starts measurement of a specimen in accordance with the instruction from the analysis unit 300.
In step S300, in accordance with the measurement instruction from the analysis unit 300, the specimen suction part 450 suctions a specimen from a collection tube and discharges the suctioned specimen into a reaction chamber. The measurement instruction from the analysis unit 300 includes information of a measurement channel with respect to which measurement is requested by the measurement order. On the basis of the information of the measurement channel included in the measurement instruction, the specimen suction part 450 discharges the specimen into the reaction chamber of the corresponding measurement channel.
In step S301, in accordance with the measurement instruction from the analysis unit 300, the sample preparation part 440 prepares a measurement sample. Specifically, on the basis of the information of the measurement channel included in the measurement instruction, the sample preparation part 440 supplies a reagent (hemolytic agent and staining liquid) to the reaction chamber having the specimen discharged therein, and mixes the specimen and the reagent. Accordingly, a measurement sample (e.g., WDF measurement sample, RET measurement sample, WPC measurement sample, PLT-F measurement sample, WNR measurement sample) is prepared.
The sample preparation part 440 supplies a reagent to a reaction chamber having the specimen discharged therein, and mixes the specimen and the reagent, to prepare an RBC/PLT measurement sample. The sample preparation part 440 supplies a reagent to a reaction chamber having the specimen discharged therein, and mixes the specimen and the reagent, to prepare a hemoglobin measurement sample.
In step S302, in accordance with the measurement instruction from the analysis unit 300, the FCM detection part 410 measures the prepared measurement sample. Specifically, in accordance with the measurement instruction from the analysis unit 300, the apparatus mechanism part 430 sends the measurement sample in the reaction chamber of the sample preparation part 440 to the FCM detection part 410. The measurement sample sent from the reaction chamber is caused to flow in the flow cell 4113, and is irradiated with laser light by the light source 4111 (see
The RBC/PLT detection part 4101 performs measurement of blood cells by a sheath flow DC detection method on the basis of the RBC/PLT measurement sample. The HGB detection part 4102 performs measurement of hemoglobin by an SLS-hemoglobin method on the basis of the hemoglobin measurement sample. An analog signal detected by the RBC/PLT detection part 4101 is processed by the analog processing part 4201, and then outputted to the A/D converter 4611. An analog signal detected by the HGB detection part 4102 is processed by the analog processing part 4202, and then outputted to the A/D converter 4612 (see
In step S303, as described above, the A/D converter 461 generates digital data by sampling at a predetermined rate the analog optical signal, and generates waveform data that corresponds to each of analytes on the basis of the digital data. The waveform data generated by the A/D converter 461 is transferred directly to a RAM by, for example, DMA transfer, not via the processor 3001 of the analysis unit 300. Accordingly, waveform data based on a forward scattered light signal, waveform data corresponding to side scattered light, and waveform data corresponding to fluorescence, which have been obtained from each analyte, are taken into the RAM 3017.
The A/D converter 4611 generates digital data by sampling at a predetermined rate the analog signal from the RBC/PLT detection part 4101. The A/D converter 4612 generates digital data by sampling at a predetermined rate the analog signal from the HGB detection part 4102. These pieces of digital data may also be taken into the RAM 3017.
In step S201, the processor 3001 of the analysis unit 300 executes the AI analysis on the waveform data by using the AI algorithm 60, and executes the calculation processing analysis with respect to a representative value, of the waveform data, that corresponds to a feature of the analyte. Apportioning between the AI analysis and the calculation processing analysis has been described above. Accordingly, the analyte in the specimen is classified. Although the process of the AI analysis in step S201 will be described later, the processor 3001 obtains, as a result of the process using the parallel-processing processor 3002, classification information 82 of each individual analyte in the specimen, for example, and obtains a label value 83 and an analysis result 84 (see
In step S202, the processor 3001 analyzes the label value 83 and the analysis result 84 by using a program stored in the storage 3004, and generates a test result of the specimen. In step S202, for example, on the basis of the label value 83 and the analysis result 84 of each individual analyte, the number of analytes is counted for each type of analyte.
For example, in a case of an example in which a test of blood cells in a blood specimen is performed, if, in one specimen, there are N pieces of classification information provided with a label value “1” which indicates neutrophil, a counting result that the number of neutrophils = N is obtained as a test result of the specimen. The processor 3001 obtains the counting result regarding the measurement item corresponding to the measurement channel on the basis of the analysis results 84, and stores the counting result, together with identification information of the specimen, into the storage 3004.
Here, the measurement item corresponding to the measurement channel is an item of which the counting result is requested by the measurement order. For example, a measurement item corresponding to the WDF channel includes a measurement item of the number of five classifications of white blood cells, i.e., monocytes, neutrophils, lymphocytes, eosinophils, and basophils. A measurement item corresponding to the RET channel includes a measurement item of the number of reticulocytes. A measurement item corresponding to PLT-F includes a measurement item of the number of platelets. A measurement item corresponding to WPC includes a measurement item of the number of hematopoietic progenitor cells. A measurement item corresponding to WNR includes a measurement item of the number of white blood cells and nucleated red blood cells.
The counting result is not limited to that of an item (also referred to as “reportable item”) for which measurement as listed above is requested, and can include a counting result of another cell of which measurement can be performed in the same measurement channel. For example, in the case of the WDF channel, as shown in
Further, the processor 3001 analyzes the obtained counting result to generate a test result of the specimen, and stores the result into the storage 3004. The analysis of the counting result includes performing determination as to, for example, whether the counting result is in a normal value range, whether any abnormal cell has been detected, whether separation from the previous test result is in an allowable range, and the like.
In step S203, the computer 301 displays the test result generated by the analysis unit 300, on a display part.
Step S201 is executed by the processor 3001 in accordance with operation of the analysis software 3100.
In step S2010, the processor 3001 causes the waveform data taken into the RAM 3017 in step S303, to be transferred to the parallel-processing processor 3002. As shown in
In step S2011, the processor 3001 instructs the parallel-processing processor 3002 to execute parallel processing on the waveform data. The processor 3001 instructs the execution of parallel processing by calling a kernel function of the parallel-processing processor 3002, for example. The process executed by the parallel-processing processor 3002 will be described later with reference to
In step S2012, the processor 3001 receives results of arithmetic operations executed by the parallel-processing processor 3002. The arithmetic results are DMA-transferred from the RAM 3201 to the RAM 3017 as shown in
Step S2011 is executed by the parallel-processing processor 3002 on the basis of an instruction from the processor 3001.
In step S2100, the processor 3001, which executes the analysis software 3100, causes the parallel-processing processor 3002 to execute assignment of arithmetic processes to the arithmetic units 3200. For example, the processor 3001 causes the parallel-processing processor 3002 to execute assignment of arithmetic processes to the arithmetic units 3200, by calling a kernel function of the parallel-processing processor 3002. As shown in
In step S2101, the arithmetic processes are processed in parallel by a plurality of arithmetic units 3200. The arithmetic processes are executed on the plurality of pieces of waveform data. In step S2102, arithmetic results generated through the parallel processing by the plurality of arithmetic units 3200 are transferred from the RAM 3201 to the RAM 3017. The arithmetic results are DMA-transferred from the RAM 3201 to the RAM 3017 as shown in
In step S201 in
Next, with reference to
In the example shown in
In the example shown in
The analysis unit 300 includes a connection port 3007, an A/D converter 3008, and an IF part 3009.
The connection port 3007 is connected to the connection port 421 (see
The A/D converter 3008 is connected to the connection port 3007. As described above, the A/D converter 3008 samples each analog optical signal outputted from the measurement unit 400, to generate waveform data that corresponds to each analyte in the specimen. The generated waveform data is stored into the storage 3004 or the RAM 3017 via the IF part 3009 and the bus 3003. The transmission path from the connection port 3007 to the A/D converter 3008 may have wires of which the number corresponds to the types of optical signals transmitted to the analysis unit 300.
The processor 3001 and the parallel-processing processor 3002 execute arithmetic processes on the waveform data stored in the storage 3004 or the RAM 3017. The analysis software 3100, which operates on the processor 3001, is similar to the analysis software 3100 shown in
Next, with reference to
The measurement unit 400 shown in
The IF part 4631 is an interface serving as a dedicated line having a communication band of not less than 1 gigabit/second, for example. For example, the IF part 4631 is an interface according to Gigabit Ethernet, USB 3.0, or Thunderbolt 3. When the IF part 4631 is of Gigabit Ethernet, the transmission line 4632 is a LAN cable. When the IF part 4631 is of USB 3.0, the transmission line 4632 is a USB cable according to USB 3.0. The transmission line 4632 is a dedicated transmission line for transmitting digital data between the measurement unit 400 and the analysis unit 300, for example.
The analysis unit 300 shown in
The analysis software 3100, which operates on the processor 3001, has functions similar to those of the analysis software 3100 described above. The analysis software 3100 analyzes the type of each analyte in the specimen through operations similar to those in the related description above.
In the configuration shown in
The measurement unit 400 and the analysis unit 300 are connected to each other in a one-to-one relationship via the transmission line 4632, for example. The transmission line 4632 in this case is a transmission line that provides no transmission of data related to an apparatus other than components (e.g., the measurement unit 400 and the analysis unit 300) forming the specimen analyzer 4000. The transmission line 4632 is a transmission line different from an intranet or the Internet, for example. Accordingly, even when waveform data generated in the measurement unit 400 is transmitted to the analysis unit 300, bottleneck in the communication speed of transmission of digital data can be avoided.
Next, with reference to
In the present configuration example, an analysis unit 600 is provided between the measurement unit 400 and the computer 301. That is, in the configuration in
In the measurement unit 400 in
The analysis unit 600 includes a processor 6001, the parallel-processing processor 6002, a bus 6003, a storage 6004, a RAM 6005, and IF parts 6006, 6007. Each component of the analysis unit 600 is connected to the bus 6003.
The bus 6003 is a transmission line having a data transfer rate of not less than several hundred MB/s, for example. The bus 6003 may be a transmission line having a data transfer rate of not less than 1 GB/s. The bus 6003 performs data transfer on the basis of the PCI-Express or PCI-X standard, for example. The analysis unit 600 may be connected to a plurality of the measurement units 400 via a plurality of the IF parts 6006. When a plurality of the measurement units 400 are provided, an analysis unit 600 may be connected to each of the measurement units 400. In this case, for example, a plurality of the measurement units 400 and a plurality of the analysis units 600 are connected in a one-to-one relationship, respectively.
The processor 6001 and the parallel-processing processor 6002 have configurations and functions similar to those of the processor 3001 and the parallel-processing processor 3002 described above. The parallel-processing processor 6002 includes a plurality of arithmetic units 6200 and a RAM 6201. Analysis software 6100, which analyzes the type of each analyte in the specimen, operates on the processor 6001. The analysis software 6100 operating on the processor 6001 has functions similar to those of the analysis software 3100 shown in
The computer 301 in
The analysis software 3100 need not necessarily operate on the processor 3501. The computer 301 receives, via the IF part 3506, analysis results obtained by the analysis unit 600. The IF part 3506 is of Ethernet or USB, for example. The IF part 3506 may be an interface capable of performing wireless communication.
In the configuration of
As described above, the IF part 4631 is a dedicated interface that connects the measurement unit 400 and the analysis unit 600, and the IF part 4631 connects the measurement unit 400 and the analysis unit 600 in a one-to-one relationship. In other words, the transmission line 4632 is a transmission line that provides no transmission of data related to an apparatus other than components (e.g., the measurement unit 400 and the analysis unit 300) forming the specimen analyzer 4000, for example. The transmission line 4632 is a transmission line different from an intranet or the Internet. Accordingly, even when waveform data generated in the measurement unit 400 is transmitted to the analysis unit 600, bottleneck in the communication speed of transmission of the waveform data can be avoided.
In this case, steps S200 to S202 in
Next, with reference to
In the measurement unit 400 in
When compared with the configuration in
The analog optical signal transmitted from the measurement unit 400 via the connection cable 4210 is inputted to the A/D converter 6009 via the connection port 6008. The A/D converter 6009 generates waveform data from the optical signal through a process similar to that of the A/D converter 461.
The analysis unit 600 may be connected to a plurality of the measurement units 400 via a plurality of the connection ports 6008. When a plurality of the measurement units 400 are provided, an analysis unit 600 may be connected to each of the measurement units 400. In this case, for example, a plurality of the measurement units 400 and a plurality of the analysis units 600 are connected in a one-to-one relationship, respectively.
In the configuration in
Next, with reference to
When compared with the configuration in
When compared with the configuration in
In the configuration in
Next, with reference to
When compared with the configuration in
When compared with the configuration in
Next, with reference to
In the measurement unit 400 in
When compared with the configuration in
Next, the data sizes of waveform data and digital data will be described.
In the present embodiment, for example, with respect to one analyte in the specimen, sampling is performed for each of an analog optical signal (FSC) based on forward scattered light, an analog optical signal (SSC) based on side scattered light, and an analog optical signal (FL) based on fluorescence.
Examples of the sampling rate include sampling at 1024 points at a 10 nanosecond interval, sampling at 128 points at an 80 nanosecond interval, sampling at 64 points at a 160 nanosecond interval, and the like. The data amount is 2 bytes per sampling, for example. With respect to each of FSC, SSC, and FL, data (in the case of the rate of 1024 points, 2 bytes×1024=2048 bytes) of an amount corresponding to the sampling rate is obtained. This data amount is the data amount per analyte in the specimen.
In a single measurement, FSC, SSC, and FL are measured with respect to at least 100 analytes, for example. In a single measurement, FSC, SSC, and FL may be measured with respect to at least 1000 analytes, for example. In a single measurement, FSC, SSC, and FL may be measured with respect to about 10000 to about 140000 analytes, for example. Therefore, when the number of analytes measured in a single measurement is 100000 and the sampling rate is 1024, the data amount of digital data of each of FSC, SSC, and FL is 2 bytes×1024×100000=204,800,000 bytes, and the data amount in total of FSC, SSC, and FL is 614,400,000 bytes.
Further, FSC, SSC, and FL are measured for each measurement channel. When the number of analytes measured in a single measurement is 100000, the sampling rate is 1024, and the number of measurement channels is 5, the data amount of each of FSC, SSC, and FL is 2 bytes×1024×100000×5=1,024,000,000 bytes, and the data amount in total of FSC, SSC, and FL is 3,072,000,000 bytes.
Thus, the volume of digital data is several hundred megabytes to several gigabytes per specimen, for example, and is at least 1 gigabyte depending on the number of analytes, the sampling rate, and the number of measurement channels.
According to the present embodiment, when digital data having a huge volume of several hundred megabytes to several gigabytes per specimen is analyzed, the analysis process using the AI algorithm 60 is concluded inside the specimen analyzer 4000 as described above, and the digital data is not transmitted, via the Internet or an intranet, to an analysis server provided outside the specimen analyzer 4000. Therefore, decrease in the throughput associated with increase in communication load caused by transmission of the digital data from the specimen analyzer 4000 to the analysis server can be avoided.
The configuration of a measurement unit 400a is similar to that of the measurement unit 400 described above. In the measurement unit 400a, a measurement sample prepared on the basis of a specimen is sent to the flow cell 4113. The light source 4111 (see
The deep learning apparatus 100 is a vendor-side apparatus. The deep learning apparatus 100 receives training waveform data obtained by the measurement unit 400a. The generation method of the training waveform data has been described above. The AI algorithm 50 stored in the deep learning apparatus 100 is a deep learning algorithm. The deep learning apparatus 100 causes the AI algorithm 50 configured as a neural network before being trained, to learn by using training data, and provides the user with the AI algorithm 60 having been trained by the training data. The AI algorithm 60 configured as a learned neural network is provided to the specimen analyzer 4000 from the deep learning apparatus 100 through a storage medium 98 or a communication network 99. The storage medium 98 is a computer-readable non-transitory tangible storage medium such as a DVD-ROM or a USB memory, for example.
The deep learning apparatus 100 is implemented as a general-purpose computer, for example, and performs a deep learning process on the basis of a flowchart described later.
The specimen analyzer 4000 executes the AI analysis on waveform data that corresponds to each analyte, by using the AI algorithm 60 configured as a learned neural network.
The deep learning apparatus 100 includes a processing part 10, an input part 16, and an output part 17.
The input part 16 and the output part 17 are connected to the processing part 10 via an IF part 15. The input part 16 is an input device such as a keyboard or a mouse, for example. The output part 17 is a display device such as a liquid crystal display, for example.
The processing part 10 includes a CPU 11, a memory 12, a storage 13, a bus 14, the IF part 15, and a GPU 19.
The CPU 11 performs data processing described later. The memory 12 is used as a work area for the data processing. The storage 13 stores a program and processing data described later. The bus 14 transmits data between components. The IF part 15 performs input/output of data to/from an external apparatus. The GPU 19 functions as an accelerator that assists arithmetic processes (e.g., parallel arithmetic processes) performed by the CPU 11. That is, in the description below, the processes performed by the CPU 11 also include processes performed by the CPU 11 using the GPU 19 as an accelerator. The GPU 19 has a function equivalent to that of the parallel-processing processor 3002, 6002 described above. Instead of the GPU 19, a chip suitable for calculation in a neural network may be used. Examples of such a chip include FPGA, ASIC, and Myriad X (Intel).
In order to perform the process of each step described later with reference to
In the description below, unless otherwise specified, the processes performed by the processing part 10 mean processes performed by the CPU 11 on the basis of the program and the AI algorithm 50 stored in the storage 13 or the memory 12. The CPU 11 temporarily stores necessary data (intermediate data being processed, etc.) using the memory 12 as a work area, and stores, as appropriate in the storage 13, data to be saved for a long time such as arithmetic results.
The configuration of the specimen analyzer 4000 (see
In order to perform the process of each step described in the waveform data analysis process below, the specimen analyzer 4000 has previously stored, in the storage 3004 (see
The AI algorithm 60 stored in the storage 3004, 6004 may be updated via a communication network. The deep learning apparatus 100 transmits the AI algorithm 60 to the specimen analyzer 4000 via a communication network (e.g., Internet, intranet). The specimen analyzer 4000 updates, by the received AI algorithm 60, the AI algorithm 60 already stored in the storage 3004, 6004.
The processing part 10A of the deep learning apparatus 100 includes a training data generation part 101, a training data input part 102, and an algorithm update part 103. A program that causes a computer to execute a deep learning process is installed in the storage 13 or the memory 12 of the processing part 10 shown in
A training data database (DB) 104 and an algorithm database (DB) 105 are stored in the storage 13 or the memory 12 of the processing part 10 shown in
The processes of steps S401, S404, and S406 in
First, the processing part 10A obtains the training waveform data 72a, 72b, 72c. The pieces of the training waveform data 72a, 72b, 72c are pieces of waveform data based on forward scattered light, side scattered light, and fluorescence, respectively. The training waveform data 72a, 72b, 72c may be obtained, for example, from the measurement unit 400a, from the storage medium 98, or via the communication network 99, through operation by the operator. When the training waveform data 72a, 72b, 72c is obtained, information regarding which cell type the training waveform data 72a, 72b, 72c indicates is also obtained. The information regarding the cell type may be associated with the training waveform data 72a, 72b, 72c, or may be inputted by the operator through the input part 16.
In step S401, the processing part 10A generates the training data 75 from the training waveform data 72a, 72b, 72c and the label value 77 as shown in
In the cell type analysis method according to the present embodiment, a convolutional neural network is used, and a stochastic gradient descent method is used. Therefore, in step S403, the processing part 10A determines whether or not training results of a previously set predetermined number of times of trials have been accumulated. When the predetermined number of training results have been accumulated (S403: YES), the processing part 10A advances the process to step S404. On the other hand, when the predetermined number of training results have not been accumulated (S403: NO), the processing part 10A skips the process of step S404.
When the predetermined number of training results have been accumulated (S403: YES), the processing part 10A updates, in step S404, the connection weight w of the neural network forming the AI algorithm 50, by using the training results accumulated in step S402. In the cell type analysis method according to the present embodiment, since the stochastic gradient descent method is used, the connection weight w of the neural network is updated at the stage where the predetermined number of times of training results have been accumulated. Specifically, the process of updating the connection weight w is a process of performing calculation by the gradient descent method, represented by Formula 12 and Formula 13 described later.
In step S405, the processing part 10A determines whether or not the AI algorithm 50 has been trained by a prescribed number of pieces of training data 75. When the AI algorithm 50 has been trained by the prescribed number of pieces of training data 75 (S405: YES), the deep learning process ends. On the other hand, when the AI algorithm 50 has not been trained by the prescribed number of pieces of training data 75 (S405: NO), the processing part 10A takes in another piece of training waveform data 72a, 72b, 72c in step S406, and returns the process to step S401.
Through the processes as above, the processing part 10A trains the AI algorithm 50 and obtains the AI algorithm 60.
The upper part of
In the neural network of the AI algorithm 50, a plurality of nodes 89 arranged in a layered manner are connected between the layers. Accordingly, information is propagated only in one direction indicated by an arrow D in the drawing, from the input layer 50a to the output layer 50b.
The middle part of
Each input is multiplied by a different weight. In Formula 2, b is a value referred to as bias. The output (z) of the node serves as an output of a predetermined function f with respect to the total input (u) represented by Formula 2, and is represented by Formula 3 below. The function f is referred to as an activation function: [Math 3]
The lower part of
When Formula 4-1 to Formula 4-3 are generalized, Formula 4-4 below is obtained. Here, i=1,..., I, and j=1,..., J. I is the total number of inputs, and J is the total number of outputs: [Math 5]
When Formula 4-4 is applied to the activation function, an output represented by Formula 5 below is obtained: [Math 6]
In the cell type analysis method according to the embodiment, a rectified linear unit function is used as the activation function. The rectified linear unit function is represented by Formula 6 below: [Math 7]
Formula 6 is a function obtained by setting u=0 to the part u<0 in the linear function with z=u. In the example shown in the lower part of
If a function expressed by use of a neural network is defined as y(x:w), the function y(x:w) varies when a parameter w of the neural network is varied. Adjusting the function y(x:w) such that the neural network selects a more suitable parameter w with respect to the input x is referred to as neural network training/learning. It is assumed that a plurality of pairs of inputs and outputs of a function expressed by use of a neural network are given. When a desirable output for an input x is defined as d, the pairs of the input/output are given as {(x1,d1), (x2,d2),..., (xn,dn)}. The set of pairs expressed as (x,d) is referred to as training data. Specifically, as shown in
The neural network learning means adjusting the weight w such that, with respect to any input/output pair (xn,dn), the output y(xn:w) of the neural network when given the input xn becomes close to the output dn as much as possible, as shown in the Formula below: [Math 9]
An error function is a measure for measuring closeness between the training data and the function expressed by use of the neural network. The error function is also referred to as a loss function. An error function E(w) used in the cell type analysis method according to the embodiment is represented by Formula 7 below. Formula 7 is referred to as cross entropy: [Math 10]
A method for calculating the cross entropy of Formula 7 will be described. In the output layer 50b of the neural network used in the cell type analysis method according to the embodiment, i.e., in the last layer of the neural network, an activation function for classifying the input x into a finite number of classes in accordance with the contents, is used. The activation function is referred to as a softmax function, and is represented by Formula 8 below. It is assumed that, in the output layer 50b, nodes are arranged by the same number as the number of classes k. It is assumed that the total input u of each node k (k=1,..., K) of an output layer L is given as uk(L) from the outputs of the previous layer L-1. Accordingly, the output of the k-th node in the output layer is represented by Formula 8 below: [Math 11]
Formula 8 is the softmax function. The sum of output y1,..., yK determined by Formula 8 is always 1.
When each class is expressed as C1,..., CK, output yK of node k in the output layer L (i.e., uk(L)) represents the probability at which the given input x belongs to class CK. The input x is classified to a class at which the probability represented by Formula 9 below is highest: [Math 12]
In the neural network learning, a function expressed by the neural network is considered as a model of the posterior probability of each class. The likelihood of the weight w with respect to the training data is evaluated under such a probability model, and a weight w that maximizes the likelihood is selected.
It is assumed that the target output by the softmax function of Formula 8 is 1 only when the output is a correct class, and otherwise, the target output is 0. When the target output is expressed in a vector form dn=[dn1,..., dnK], if, for example, the correct class of input xn is C3, only target output dn3 becomes 1, and the other target outputs become 0. When coding is performed in this manner, the posterior distribution is represented by Formula 10 below: [Math 13]
Likelihood L(w) of the weight w with respect to the training data {(xn,dn)}(n=1,..., N) is represented by Formula 11 below. When the logarithm of the likelihood L(w) is taken and the sign is inverted, the error function of Formula 7 is derived: [Math 14]
Learning means minimizing the error function E(w) calculated on the basis of the training data, with respect to the parameter w of the neural network. In the cell type analysis method according to the embodiment, the error function E(w) is represented by Formula 7.
Minimizing the error function E(w) with respect to the parameter w has the same meaning as finding a local minimum point of the error function E(w). The parameter w is a weight of connection between nodes. The local minimum point of the weight w is obtained by iterative calculation of repeatedly updating the parameter w from an arbitrary initial value used as a starting point. An example of such calculation is the gradient descent method.
In the gradient descent method, a vector represented by Formula 12 below is used: [Math 15]
In the gradient descent method, a process of moving the value of the current parameter w in the negative gradient direction (i.e., -∇E) is repeated many times. When the current weight is w(t) and the weight after the moving is w(t+1), the arithmetic operation according to the gradient descent method is represented by Formula 13 below. The value t means the number of times the parameter w is moved: [Math 16]
The symbol used in Formula 13 and shown in Formula 14 below is a constant that determines the magnitude of the update amount of the parameter w, and is referred to as a learning coefficient:
As a result of repetition of the arithmetic operation represented by Formula 13, an error function E(w(t)) decreases in association with increase of the value t, and the parameter w reaches a local minimum point.
It should be noted that the arithmetic operation according to Formula 13 may be performed on all of the training data (n=1,..., N), or may be performed on only a part of the training data. The gradient descent method performed on only a part of the training data is referred to as a stochastic gradient descent method. In the cell type analysis method according to the embodiment, the stochastic gradient descent method is used.
The specimen analyzer 4000 includes: the measurement unit 400 that includes the FCM detection part 410 or the detection part 470 (optical detection part) for obtaining an optical signal from a specimen; and the analysis unit 300 or the analysis unit 600 that analyzes first data and second data that correspond to the optical signal. The analysis unit 300, 600 executes the AI analysis (a first analysis operation according to an artificial intelligence algorithm) on the first data out of all of waveform data having been obtained, and executes the calculation processing analysis (a second analysis operation that processes a representative value that corresponds to a feature of an analyte) on the second data out of all of the waveform data having been obtained.
According to this configuration, since the analysis process on data corresponding to the optical signal obtained from the specimen is apportioned between the AI analysis and the calculation processing analysis, the load on the analysis unit 300, 600 being the computer that processes the data can be reduced when compared with a case where all of the data corresponding to the optical signal is analyzed by using only the artificial intelligence algorithm.
When the specimen analyzer 4000 is a blood cell analyzer or a urine analyzer, the first data and the second data are digital data (waveform data) that corresponds to the intensity of the optical signal based on light generated from each analyte (cell or particle). The optical signal in this case is analog signals outputted from light receiving elements on the basis of forward scattered light, side scattered light, and fluorescence. The optical signal is a signal that has a region corresponding to each of analytes in the specimen and that reflects the presence of the analyte in the specimen. The waveform data (the first data and the second data) is generated, corresponding to the region of the optical signal. In other words, the waveform data corresponds to the optical signal obtained while an analyte passes through an application position of light from the light source 4111. The representative value that corresponds to a feature of an analyte is a value (see
When the specimen analyzer 4000 is a blood coagulation measurement apparatus, the first data and the second data are digital data (coagulation waveform data) that corresponds to the intensity of the optical signal based on transmitted light or scattered light. The optical signal in this case is an analog signal from start of light measurement to end of light measurement (e.g., after 180 seconds from the start) based on the intensity of transmitted light or scattered light. The optical signal may be an analog signal from start of coagulation reaction (timing T2 in
The first data and the second data may be data identical to each other or may be data completely different from each other. For example, in a case where, on the basis of waveform data obtained through a single measurement according to the WDF channel, the AI analysis is executed with respect to nucleated red blood cells and basophils and the calculation processing analysis is executed with respect to the other white blood cells, the first data and the second data are data identical to each other. In a case where, out of waveform data obtained through two measurements, the calculation processing analysis is executed on waveform data that is based on the first measurement and the AI analysis is executed on waveform data that is based on the second measurement, the first data and the second data are data different from each other.
The representative value of waveform data (the second data) to be subjected to the calculation processing analysis is specified on the basis of the magnitude of the waveform data (the second data). Specifically, a representative value such as the peak value, the area, or the width, and a representative value such as the time taken for the absorbance to become 50% are specified on the basis of the magnitude of the second data. Accordingly, the representative value can be smoothly specified.
When the specimen analyzer 4000 is a blood cell analyzer or a urine analyzer, the optical signal has a region that corresponds to each of analytes in the specimen. In the calculation processing analysis, the analysis unit 300, 600 specifies a representative value to serve as a target of the calculation processing analysis, on the basis of waveform data (the second data) that corresponds to each of regions of the optical signal. Since the optical signal includes regions that correspond to the respective analytes, a representative value such as the peak value, the area, or the width that corresponds to each analyte can be smoothly specified on the basis of the waveform data that corresponds to each region of the optical signal.
When the specimen analyzer 4000 is a blood cell analyzer or a urine analyzer, the optical signal has a region that corresponds to each of analytes in the specimen. In the AI analysis, the analysis unit 300, 600 inputs waveform data (the first data) that corresponds to each of the regions of the optical signal, into the artificial intelligence algorithm. Since the optical signal includes a region that corresponds to each of analytes, the AI analysis can be smoothly executed by inputting waveform data that corresponds to each region of the optical signal, into the artificial intelligence algorithm.
As described above, when the optical signal has a region that corresponds to each of analytes in the specimen, the measurement unit 400 obtains waveform data (the first data and the second data) on the basis of a signal that is greater than a predetermined threshold and that corresponds to the intensity of the optical signal, as shown in the drawing in the upper part of
As shown in
As shown in
As shown in
As shown in
As shown in
As shown in
As shown in
The representative value that is subjected to the calculation processing analysis (the second analysis operation) has a data amount smaller than that of waveform data (the first data) that is inputted to the AI algorithm 60 in the AI analysis (the first analysis operation). That is, since, in the calculation processing analysis, the data amount to be processed is smaller than that of the AI analysis, the load on the computer that performs the analysis is smaller than that in the case of the AI analysis. Accordingly, the TAT (Turn Around Time) of the analysis of the measurement result can be shortened.
Various modifications can be made as appropriate to the embodiments of the present disclosure, without departing from the scope of the technological idea defined by the claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2022-042964 | Mar 2022 | JP | national |
| 2022-042965 | Mar 2022 | JP | national |
