Patent Application
20230253109
The present invention relates to a technology of a cancer screening device and a cancer screening method.
In a conventional cancer screening system using urinary metabolites, biomarkers are narrowed down on the basis of, for example, information of two groups such as a cancer group and a non-cancer group, and a prediction value is calculated using a prediction formula according to the following formula (1). If the prediction value is a value of “+”, it is determined that a possibility of a cancer is high, and if the prediction value is “−”, it is determined that a possibility of a cancer is low. The prediction formula such as formula (1) is appropriately referred to as a cancer screening model.
The cancer screening model according to formula (1) is for identifying whether or not a cancer is present and is commonly used. The biomarkers are urinary metabolites having a causal relationship with onset of a cancer. In other words, although biomarkers are urinary metabolites, not all urinary metabolites are biomarkers. Hereinafter, the biomarkers will be referred to as markers, and the urinary metabolites will be referred to as metabolites.
Prediction value=α×(intensity of marker #1)+β×(intensity of marker #2)+γ×(intensity of marker #3)+δ (1)
α, β, γ, and δ in formula (1) are constants. Here, while formula (1) is a cancer screening model for determining whether or not a cancer is present, as described later, a prediction formula for determining whether or not a predetermined cancer type has developed and a state of a cancer will be also referred to as a cancer screening model.
Patent Literature 1 discloses a method for searching for biomarkers in urinary metabolites, including “searching for a urinary metabolite marker, including the steps of: (a) subjecting a urine specimen to a liquid chromatograph mass spectrometer (LC/MS) and analyzing a urinary metabolite in the urine specimen; (b) quantitatively evaluating an importance level of the urinary metabolite by a random forest method on the basis of analysis data of the urinary metabolite and selecting a urinary metabolite having a high importance level; (c) performing a discrimination analysis method using the analysis data of the selected urinary metabolite; and (d) determining a urinary metabolite associated with a specific disease or condition as a marker candidate on the basis of a result of the discrimination analysis”.
When a screening system using a cancer screening model according to formula (1) is put into practical use, the following problems arise.
(A) The cancer screening model according to formula (1) targets data whose correct answer (whether or not it is a cancer) is known, and the cancer screening model is created so as to have a high correct answer rate. Thus, if a prediction value by the prediction formula of formula (1) becomes even slightly positive, the system possibly determined that there is a possibility of a cancer. Conversely, if the prediction value by the prediction formula of formula (1) is even slightly negative, the system possibly determined that there is no possibility of a cancer. Although the cancer screening model based on the prediction formula of formula (1) is excellent, in the discriminant analysis, a magnitude of the calculated prediction value is not taken into account, and meaning is not given.
(B) When accuracy (sensitivity, specificity, AUC, etc.) of the constructed cancer screening model is verified, evaluation is performed using data for which whether or not there is a cancer is known. However, when this screening system is put into practical use, data for which an answer is unknown is often targeted, but the data is discriminated between two groups of whether or not there is a cancer. On the other hand, in practical use, it is desirable to indicate, as pre-screening, determination of a risk of developing a cancer, whether or not a cancer is in an early stage, or as a prognostic test, whether or not cancers are increased or decreased by treatment. For example, it is desired to obtain screening results other than two groups of whether or not there is a cancer, such as a size of a cancer, a degree of invasion, and the like. The technique described in PTL 1 also needs further improvement from such a viewpoint.
The present invention has been made in view of such a background, and an object of the present invention is to implement a variety of kinds of cancer screening.
In order to solve the above-described problems, a cancer screening device of the present invention includes a first acquisition unit configured to acquire cancer screening data storing cancer screening results that are results of cancer screening for first subjects who are a plurality of subjects including cancer patients and healthy subjects and acquire first metabolite exhaustive data that is a result of analysis performed by LC/MS on first urine specimens collected from the first subjects and is information on amounts of a plurality of metabolites in the first urine specimens; a cancer screening model generation unit configured to construct, as a cancer screening model, a relationship between the cancer screening results in the cancer screening data and the respective amounts of the metabolites in the first metabolite exhaustive data on the basis of the cancer screening data and the first metabolite exhaustive data; and a second acquisition unit configured to acquire second metabolite exhaustive data that is a result of analysis performed by the LC/MS on a second urine specimen collected from a second subject who is a subject different from the first subjects, a cancer state estimation unit configured to estimate a state of a cancer in the second subject by applying an amount of the metabolites in the second metabolite exhaustive data to the cancer screening model, and an output unit configured to output the estimated state of the cancer.
Other solutions will be appropriately described in the embodiment.
According to the present invention, it is possible to implement a variety of kinds of cancer screening.
Next, modes for carrying out the present invention (referred to as “embodiments”) will be described in detail with reference to the drawings as appropriate. Note that while the present embodiment is directed to colorectal cancer screening, the present invention is also applicable to screening for other cancers and is also applicable to a plurality of cancer types and general cancers.
<Cancer Screening System 10>
The cancer screening system 10 includes a cancer screening device 1, a liquid chromatography-mass spectrometry (LC/MS) 2, and a user terminal (output unit) 3.
The cancer screening device 1 generates a cancer screening model on the basis of metabolite exhaustive data 131 (see
Furthermore, a urine specimen of a subject for whom cancer screening is desired to be performed is analyzed by the LC/MS 2, and sample data 132 (see
Note that, in the present embodiment, generation of a cancer screening model and cancer screening using the generated cancer screening model are performed by one device. However, the present invention is not limited thereto, and generation of the cancer screening model and the cancer screening using the generated cancer screening model may be performed by different devices. Furthermore, in the example illustrated in
<Cancer Screening Device 1>
The cancer screening device 1 includes a communication device (a first acquisition unit, a second acquisition unit) 101, an input device 102 such as a keyboard and a mouse, and an output device (an output unit) 103 such as a display and a printer. Further, the cancer screening device 1 also includes a memory 110 and a central processing unit (CPU) 104. Still further, the cancer screening device 1 includes a clinical information DB 120, a metabolite DB 130, a screening model DB 140, an analysis condition DB 150, and a screening result DB 160.
The communication device 101 transmits and receives information between the LC/MS 2, a server (not illustrated) provided in the cancer screening institution 4, and the user terminal 3.
A program stored in a storage device (not illustrated) is loaded into the memory 110. As a result of the loaded program being executed by the CPU 104, a pre-processing unit 111, a candidate extraction unit (narrowing unit) 112, a cancer screening model generation unit 113, a screening processing unit (cancer state estimation unit) 114, and an output processing unit (output unit) 115 are embodied.
The pre-processing unit 111 performs pre-processing on clinical data (cancer screening data) 121 (see
The cancer screening model generation unit 113 generates a cancer screening model for determining various symptoms of cancers on the basis of metabolite data extracted by the candidate extraction unit 112 in the clinical data 121 and the metabolite exhaustive data 131.
The screening processing unit 114 performs cancer screening by using the generated cancer screening model and sample data (second metabolite exhaustive data) 132 (see
The output processing unit 115 transmits the result of the cancer screening to the user terminal 3, or the like.
The clinical information DB 120 stores clinical data 121 sent from the cancer screening institution 4. The clinical data 121 will be described later.
The metabolite DB 130 stores metabolite exhaustive data 131 which is a result of analysis performed by the LC/MS 2 and the sample data 132. The metabolite exhaustive data 131 and the sample data 132 will be described later.
In the screening model DB 140, information regarding the cancer screening model generated by the cancer screening model generation unit 113 is stored as cancer screening model data 141 (see
The analysis condition DB 150 stores conditions necessary for analysis performed by the LC/MS 2.
In the screening result DB 160, results of cancer screening performed by using the generated cancer screening model are stored as screening result data 161 (see
As described above, the cancer screening device 1 performs two kinds of processing of processing of generating a cancer screening model and processing of performing actual cancer screening by using the generated cancer screening model. Hereinafter, the two kinds of processing will be described.
<Cancer Screening Model Generation Flowchart>
First, analysis (LC/MS analysis) by the LC/MS 2 is performed on a urine specimen collected from a subject (S101), and cancer screening is performed for the subject (S102). In step S101, metabolites in the urine specimen are comprehensively detected by using a plurality of separation modes. Here, as a plurality of separation modes, in order to detect as many metabolites in the urine specimen as possible, separation in LC such as reversed phase, normal phase, HILIC, or the like, positive or negative ionization in MS by using an electrospray method, or the like, is used. The result of the cancer screening is stored in the clinical data 121, and the result of the analysis by the LC/MS 2 is stored in the metabolite exhaustive data 131.
Then, the clinical data 121 and the metabolite exhaustive data 131 are input to the cancer screening device 1.
Here, specific examples of the clinical data 121 and the metabolite exhaustive data 131 will be described with reference to
In the present embodiment, 30 urine specimens are collected from colorectal cancer patients and healthy subjects as a control group and analyzed by the LC/MS 2 as described above (S101). As a result, 1000 or more intensities of ions of the metabolites can be obtained.
(Clinical Data 121)
The clinical data 121 includes fields of “specimen ID”, “donor ID”, “collection date”, “pathological name”, “details”, “age”, “gender”, “stage”, “T factor (T)”, “N factor (N)”, and “M factor (M)”.
Here, the “specimen ID” is an ID for uniquely distinguishing a urine specimen.
The “donor ID” is an ID for uniquely distinguishing a subject (donor).
The collection date is date on which the urine specimen is collected.
“Pathological name” is a name of a cancer found as a result of cancer screening. Note that “NA” means that the cancer is benign or no cancer has been detected.
The “details” stores detailed information of the detected cancer (colorectal cancer). In the example of
The “T factor (T)” is an index indicating a size of the tumor and a degree of invasion, and is T1a, T1b, T2a, T2b, T3, and T4 in the order of ascending symptoms from the mildest symptom.
“N factor (N)” is an index indicating a degree of lymph node metastasis of a tumor, and “NO” indicates a case where there is no lymph node metastasis, and “N3” indicates a case where there is the most metastasis.
“M factor (M)” is an index indicating distant metastasis, “M1a” or “M1b” indicates a case where there is distant metastasis and a location of metastasis, and “M0” indicates a case where there is no distant metastasis.
As illustrated in
(Metabolite Exhaustive Data 131)
As illustrated in
The “specimen ID” is the same as the “specimen ID” in FIG. 4.
Each field of “metabolite A”, “metabolite B”, “metabolite C”, . . . stores information of ion intensity (hereinafter, referred to as intensity) of each metabolite in a urine specimen measured by the MS. Metabolites can be discriminated by a metabolite database (not illustrated), or the like, and also include unknown metabolites with an unknown chemical structure because only an m/z (mass-to-charge ratio) is known at the time of MS. In addition, as illustrated in
The explanation is back to
Next, in step S103, the pre-processing unit 111 performs pre-processing to the input clinical data 121 and metabolite exhaustive data 131. The pre-processing unit 111 performs data association, data integration, unnecessary data cleaning, format conversion, normalization, normalization by osmolality or creatinine concentration, standardization, missing value complementation, outlier exclusion, autoscaling, and the like, as necessary. In this process, drugs that are not included in the cancer screening model, exogenous metabolites derived from foods, and the like, are also excluded. Note that it is not necessary to perform all the pre-processing described here. Note that the processing in step S103 may be performed by the pre-processing unit 111 on the basis of information input by the user via the input device 102 on the basis of experience or may be automatically performed by the pre-processing unit 111.
Subsequently, the pre-processing unit 111 divides the urine specimen data in each of the pre-processed clinical data 121 and metabolite exhaustive data 131 into training data 171 for generating a cancer screening model and test data 172 for verifying the generated cancer screening model, as necessary. Here, the urine specimen data is a record having a specimen ID common to the clinical data 121 and the metabolite exhaustive data 131. The urine specimen data is randomly divided into the training data 171 and the test data 172. Note that the training data 171 is teacher data for generating a cancer screening model. The test data 172 is data for verifying the generated model. As verification, cross verification is performed.
Next, the candidate extraction unit 112 performs marker candidate extraction processing. Here, the candidate extraction unit 112 first performs a significance test (t-test, f-test, Wilcoxon rank sum test, etc.) on an amount of each metabolite in a urine specimen for two groups of cancer patients and healthy persons (S111). Then, the candidate extraction unit 112 extracts metabolites having a significant difference between cancer patients and healthy subjects as marker candidates. Further, the candidate extraction unit 112 performs correlation analysis and a random forest method which is one of machine learning (S112), calculates importance levels of the marker candidates and ranks the marker candidates. The processing in steps S111 and S112 may be executed at the time of generating each cancer screening model (S121 to S124). However, the number of metabolite types in the metabolite exhaustive data 131 is as large as several thousands, and thus, by narrowing down the number of marker candidates from several tens to several hundred in step S111 in advance, a calculation amount and a calculation period are reduced. Note that both the significance test (S111) and the random forest method (S112) do not need to be performed, and either one may be performed.
(Marker Candidate Extraction Result)
As a result of the processing in steps S111 and S112, the top 20 marker candidates obtained are illustrated in
The marker candidate extraction result data includes fields of “rank”, “importance level”, “LS/MS separation mode”, and “m/z (mass-to-charge ratio)”.
Here, the “importance level” is a degree of importance calculated by random forest. In addition, in the example of
The explanation is back to
Next, the cancer screening model generation unit 113 performs first screening model generation processing (S121). In step S121, the cancer screening model generation unit 113 uses OPLS-DA (orthogonal partial least squares discriminant analysis) to generate a first cancer screening model (first cancer screening model) 142 that is a cancer screening model for determining whether or not it is a cancer. In the present embodiment, data of colorectal cancer patients/healthy subjects are handled, and thus, whether or not it is a colorectal cancer is determined by the first cancer screening model 142. Note that not only OPLS-DA but also other discrimination analysis may be used.
For example, the cancer screening model generation unit 113 selects top 20 marker candidates among 10 marker candidates indicated in the marker candidate extraction result illustrated in
Then, the cancer screening model generation unit 113 first generates the first cancer screening model 142 for discriminating between a colorectal cancer (cancer) and healthy subjects by OPLS-DA using the training data 171. In other words, the first cancer screening model 142 determines whether or not a colorectal cancer has developed.
Specifically, the cancer screening model generation unit 113 temporarily sets a linear expression having 10 variables as intensity of 10 markers. Next, the cancer screening model generation unit 113 uses OPLS-DA to calculate a coefficient of each variable that can be discriminated between colorectal cancer patients and healthy subjects. As a result, the first cancer screening model 142 represented by the following formula (2) is generated.
y0=a1·x1+a2x2+ . . . +a9x9+a10x10+a0 (2)
Here, x1, x2, . . . , and x10 are intensity of the selected top 10 markers among the 20 marker candidates indicated in the marker candidate extraction result illustrated in
Thereafter, the cancer screening model generation unit 113 verifies the generated first cancer screening model 142 using 30 pieces of test data 172. In other words, the cancer screening model generation unit 113 applies the first cancer screening model 142 to the data with answers of colorectal cancers/healthy subjects and verifies a correct answer rate. Note that step S121 includes generation of the first cancer screening model 142 by using the OPLS-DA to verification of the generated first cancer screening model 142.
A prediction value y0 generated by using the first cancer screening model 142 indicated in formula (2) is a first-order polynomial obtained by multiplying intensity of 10 markers by a coefficient for each marker and the cancer screening model identifies a cancer patient when the prediction value y0 is positive and identifies a healthy subjects when the prediction value y0 is negative.
According to verification using 30 pieces of test data 172 illustrated in
The explanation is back to
After step S112 in
It is supposed that the second cancer screening model 143 is a scheme, for example, in which a subject collects urine at home, a risk is easily assessed and a thorough examination is suggested. First, the cancer screening model generation unit 113 divides 60 pieces of urine specimen data used in the first cancer screening model 142 into 30 pieces of training data 171 and test data 172, the same as the first cancer screening model 142 and performs logistic analysis on the training data 171. In this process, the cancer screening model generation unit 113 temporarily sets the following formula (3).
y1=1/[1+exp{−(b1·x1+b2x2+ . . . +b20·x20+b0)}] (3)
Here, x1, x2, . . . , x20 are intensity of 20 marker candidates used in the first cancer screening model 142. A suffix of x is “rank” indicated in
Next, the cancer screening model generation unit 113 calculates an odds ratio (exp(b1), exp(b2), . . . , exp(b20)) for each marker candidate. Thereafter, the cancer screening model generation unit 113 selects top 7 markers in descending order of the odds ratio. The number of markers to be selected is not limited to seven. As a result, seven markers in positions 1, 2, 5, 7, 11, 12, and 20 are selected in the “rank” illustrated in
Then, when the cancer screening model generation unit 113 applies the selected 7 markers to formula (3), the second cancer screening model 143 of the following formula (4) is obtained. Here, the cancer screening model is reconstructed with seven markers, and thus, each of the coefficients b1, b2, b5, b7, b11, b12, b20, and b0 has a value different from that in formula (3).
y1=1/[1+exp{−(b1·x1+b2x2+b5·x5+b7·x7+b11·x11+b12·x12+b20·x20+b0)}] (4)
x1, x2, . . . , b1, b2, . . . are the same as those in formula (3), and thus, description thereof is omitted here. A portion of y2 in exp{-(y2)} in formula (4), that is, the following formula (5) is set as a prediction value of the second cancer screening model 143.
y2=b1·x1+b2·x2+b5·x5+b7·x7+b11·x11+b12·x12+b20·x20+b0 (5)
Thereafter, the cancer screening model generation unit 113 verifies the second cancer screening model 143. Specifically, the cancer screening model generation unit 113 substitutes intensity of the selected marker in the 30 pieces of test data 172 into the second cancer screening model 143 (formula (4)). Then, the cancer screening model generation unit 113 compares the probability (formula (4)) obtained from the probability with onset of a cancer in the test data 172, thereby verifying the second cancer screening model 143. Step S122 includes steps from generation of the second cancer screening model 143 by logistic analysis to verification of the generated second cancer screening model 143.
When logistic analysis was performed by using actual urine specimen data to obtain an odds ratio, it was found that the rank by the odds ratio did not always coincide with the rank by the random forest indicated in
In general, a degree of matching of ranks between different analysis methods, such as the rank in the random forest and the rank obtained through the logistic analysis as described above, is improved as the number of markers to be used is larger. In addition, as the number of markers to be used is larger, a cancer screening model with higher accuracy may be generated. On the other hand, as the number of markers to be used is smaller, time and cost for cancer screening using a cancer screening model tend to be decreased. Thus, the number of markers to be used is determined by the user on the basis of these balances.
In
Here, the prediction value is a value of y2 in formula (5).
Then, a vertical axis is a probability of onset of a cancer (colorectal cancer) with respect to the prediction value y2 of the second cancer screening model 143 (y1 in formula (4)).
In
As indicated in
The explanation is back to
After step S122 in
Here, the user first divides a size of the tumor into five classes from “1” to “5” and sets a class of “0” for a person without tumor. Next, the cancer screening model generation unit 113 performs multiple regression analysis on the training data 171 used at the time of generating the first cancer screening model 142. Specifically, the cancer screening model generation unit 113 temporarily sets the following formula (11).
y4=c1·x1+c2·x2+ . . . +c20·x20+c0 (11)
Here, x1, x2, . . . , x20 are intensity of 20 marker candidates used in the first cancer screening model 142. A suffix of x is “rank” indicated in
Then, the cancer screening model generation unit 113 sets, as a third cancer screening model 144, the following formula (12) obtained by applying the selected six markers with the “ranks” of 1, 2, 5, 8, 9, and 10 illustrated in
y4=c1·x1+c2·x2+c5·x5+c8·x8+c9·x9+c10·x10+c0 (12)
Next, the cancer screening model generation unit 113 verifies the third cancer screening model 144. Specifically, the cancer screening model generation unit 113 substitutes the intensity of metabolites in the test data 172 into the third cancer screening model 144 indicated in formula (12). Then, the cancer screening model generation unit 113 compares the result obtained by substituting the intensity of metabolites in the test data 172 into the third cancer screening model 144 with the size of the tumor in the test data 172, thereby verifying the third cancer screening model 144. Note that the processing in step S123 includes generation and verification of the third cancer screening model 144 by multiple regression analysis.
In
As indicated in
Note that the prediction value on the horizontal axis in
The explanation is back to
After step S123 in
The generation procedure of the fourth cancer screening model 145 is the same procedure as the generation of the second cancer screening model 143, and thus, description thereof is omitted here. In the present embodiment, the fourth cancer screening model 145 estimates a malignant/benignancy probability of a tumor of a colorectal cancer, but the present invention is not limited thereto, and a probability of metastasis to other sites of the colorectal cancer may be used, or other qualitative probabilities may be estimated.
The generated respective cancer screening models are stored in the cancer screening model data 141 illustrated in
Here, whether or not a cancer has developed is estimated by the first cancer screening model 142, a probability of onset of a cancer (colorectal cancer in the present embodiment) is estimated by the second cancer screening model 143, a size of a tumor is estimated by the third cancer screening model 144, and malignant/benign probability of a tumor of cancer (colorectal cancer in the present embodiment) is estimated by the fourth cancer screening model 145. In addition, by performing multiple regression analysis as in the third cancer screening model 144, it is possible to generate a cancer screening model for estimating a therapeutic effect, a degree of invasion of cancer, and the like.
(Cancer Screening Model Data 141)
The cancer screening model data 141 illustrated in
As illustrated in
In the “model number”, the number of the cancer screening model is stored. For example, “model number: 1” indicates the first cancer screening model 142 described above, and “model number: 2” indicates the second cancer screening model 143 described above. The same applies to “model number: 3” and “model number: 4”. Note that information indicating what is to be estimated by using each cancer screening model is also preferably stored in the cancer screening model data 141. For example, the third cancer screening model 144 is a model for estimating “a size of a tumor”.
In the “model generation method”, the name (OPLS-DA, logistic analysis, multiple regression analysis, and the like) of the analysis method used when each cancer screening model is generated is stored.
In the “coefficient #0”, a value of a zero-order coefficient in each cancer screening model is stored. The zero-order coefficient is b0 in formula (5) and c0 in formula (12).
“Marker #1” is x1 in formula (5) or (12), and “coefficient #1” is b1 in formula (5) or c1 in formula (12).
Hereinafter, the same applies to “marker #2”, “marker #3”, . . . , “coefficient #2”, “coefficient #3”, —. Note that the number after “#” is a number in the cancer screening model and is not the “rank” in
As qualitative variables determined by the cancer screening model, in addition to those described in the present embodiment, presence or absence of cancer metastasis, presence or absence of invasion, presence or absence of angiogenesis, presence or absence of metabolic reprogramming (reflection on metabolites), and the like, are also possible. In addition, as quantitative variables determined by the cancer screening model, a degree of activity, a cancer stage, a degree of invasion, the number of angiogenesis, a degree of metabolic reprogramming, and the like, are also possible. In addition, a location of the cancer, the name of the disease, and the like, can also be determined by complexly determining the qualitative variable and the quantitative variable. These cancer screening models are generated as exhaustively as possible, and the most suitable one is used in cancer screening to be performed.
Note that there is a set of markers suitable for each cancer screening model. In each cancer screening model, the marker to be used is made common as much as possible, and the number of markers is reduced, so that efficiency of cancer screening in
In
First, analysis by the LC/MS 2 (LC/MS analysis) is performed on a urine specimen of a subject (S201), thereby measuring intensity of each metabolite.
Then, the sample data 132 is input to the cancer screening device 1. The sample data 132 may be similar to the metabolite exhaustive data 131 illustrated in
Then, the pre-processing unit 111 performs pre-processing on the input sample data 132 (S202). The processing in step S202 is similar to that in step S104 in
Subsequently, the screening processing unit 114 calculates a probability P of onset of a colorectal cancer for the urine specimen to be screened by using the second cancer screening model 143 (S211, second screening model processing). In other words, the screening processing unit 114 calculates the prediction value by substituting the intensity of the marker in the sample data 132 into formula (5). Further, the screening processing unit 114 calculates the probability P of onset (that is, y1 in formula (4)) of a colorectal cancer by substituting the calculated prediction value (y2 in formula (5)) into formula (4). Here, the second cancer screening model 143 is used, but presence or absence of onset of a cancer may be determined by using the first cancer screening model 142.
Then, the screening processing unit 114 determines whether or not the probability P of onset of a colorectal cancer calculated in the second screening model processing of step S211 is equal to or less than a predetermined value P1 (P≤P1) (S212). Here, P1=10%, but the probability is not limited thereto. Here P1=10% is set, but by setting P1=0%, the screening processing unit 114 may determine whether or not P=P1 is met in step S212.
As a result of step S212, when the probability P of onset of a colorectal cancer is equal to or less than the predetermined value P1 (here, 10%) (S212: Yes), the screening processing unit 114 outputs the result of the cancer screening (here, colorectal cancer screening) to the user terminal 3 as determination of a low risk (for example, “D” for the ABCD grading scale) of a colorectal cancer (S221).
As a result of step S212, when the probability P of onset of a colorectal cancer is higher than the predetermined value P1 (here, 10%) (S212: No), that is, when the probability of onset of a colorectal cancer is high or moderate in the cancer screening model, the screening processing unit 114 executes the next cancer screening model and calculates and outputs a more detailed state.
In the example of the flowchart illustrated in
Further, the screening processing unit 114 performs fourth screening model processing (S214). In this processing, the screening processing unit 114 calculates a malignant/benign probability of a tumor of a cancer (here, a colorectal cancer) by using the fourth cancer screening model 145 and stores the result in the screening result data 161 (see
If “No” is determined in step S212, the screening processing unit 114 performs the fifth screening model processing (S215). In this processing, the probability of metastasis of a cancer to other parts is calculated by using the fifth cancer screening model, and the result is stored in the screening result data 161. As described above, although not illustrated in
In
Finally, the output processing unit 115 generates result output data 181 illustrated in
(Screening Result Data 161)
The screening result data 161 illustrated in
As illustrated in
The “sample ID” is an ID for uniquely distinguishing a urine specimen. The urine specimen here is a urine specimen in the sample data 132.
The “colorectal cancer probability” is a probability of onset of a colorectal cancer calculated by the second cancer screening model 143.
In addition, the “benign probability” is a malignant/benign probability of a tumor of a colorectal cancer calculated by the fourth cancer screening model 145.
“Tumor size” is a class of a size of a tumor calculated by the third cancer screening model 144.
The “metastasis probability” is calculated by the fifth cancer screening model.
(Result Output Data 181)
The result output data 181 illustrated in
As illustrated in
The data stored in the result output data 181 is data stored in the screening result data 161 of
There are a wide variety of conditions (cancer type, stage, TNM, malignant/benign, tumor size, activity, angiogenesis, invasion, metastasis, etc.) in a cancer, and it is predicted that there is metabolic reprogramming. Until now, as practical cancer screening, only whether or not it is a cancer (I/O discrimination) has been determined by formula (1). However, in actual cancer screening, it is necessary to present not only such information but also malignancy/benignancy, risk (qualitative probability), therapeutic effect (quantitative variable), and the like. According to the present embodiment, it is possible to generate a cancer screening model capable of determining a probability of onset of a cancer (colorectal cancer), a size of a tumor, and the like, on the basis of intensity of metabolites in a urine specimen. As a result, it is possible to screen various cancer conditions on the basis of the intensity of metabolites in a urine specimen, and it is possible to greatly improve cost and efficiency of cancer screening. In addition, such a cancer screening model may be used as an auxiliary means such as therapeutic assistance such as correspondence with an image diagnosis result, quantitative clarification by being used for interpolation of data, and discovery of an undiscoverable minute tumor by being used for extrapolation of image data.
In addition, in steps S111 and S112, the candidate extraction unit 112 narrows down metabolites to be subjected to generation of the cancer screening model, so that the cancer screening model may be efficiently generated. When narrowing down metabolites, the candidate extraction unit 112 extracts metabolites having a significant difference between cancer patients and healthy subjects by a significance test. Alternatively, the candidate extraction unit 112 ranks metabolites on the basis of an importance level of the random forest and extracts metabolites ranked high. As a result, metabolites related to onset of a cancer may be narrowed down, so that a cancer screening model can be efficiently generated.
In addition, it is possible to determine whether or not a cancer or a predetermined cancer type (for example, colorectal cancer) has developed by the first cancer screening model 142.
Furthermore, the probability of onset of a cancer or a predetermined condition in a predetermined cancer type (for example, colorectal cancer) is estimated by the second cancer screening model 143 and the fourth cancer screening model 145.
Then, a degree of a state of a predetermined phenomenon in a cancer (for example, a size of the tumor, or the like) is estimated by the third cancer screening model 144.
In addition, by generating a cancer screening model as in the present embodiment, the inventor has found that there is a useful marker for each cancer screening model. For example, among the top metabolite candidates for determining whether or not a cancer is present (
In the present embodiment, marker candidates are narrowed down by performing a significance test and further performing random forest in step S111 in
The present invention is not limited to the above-described embodiment and includes various modifications. For example, the above-described embodiment has been described in detail for explaining the present invention in a lucid way and are not necessarily limited to those having all the described configurations.
In addition, some or all of the above-described configurations, functions, units 111 to 115, DBs 120, 130, 140, 150, 160, and the like, may be implemented by hardware, for example, by being designed with an integrated circuit. In addition, as illustrated in
In addition, in each embodiment, control lines and information lines considered to be necessary for description are illustrated, and not all control lines and information lines in a product are necessarily illustrated. In practice, it may be considered that almost all the components are connected to each other.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2020-121399 | Jul 2020 | JP | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2021/010804 | 3/17/2021 | WO |
