Patent Application
20200242329
There is increased focus on digital imagery of culture plates such as for detection of microbial growth, colony counting and/or identification. Systems and techniques for imaging plates for detecting microbes are described in PCT Publication No. WO2015/114121, WO 2016/172527, and WO 2016/172532, the entireties of which are incorporated by reference herein. Using such techniques (also referred to herein as Kiestra Systems), laboratory staff is no longer required to read plates by direct visual inspection. Shifting laboratory workflow and decision-making to analysis of digital images of culture plates can also improve efficiency.
Although significant progress has been made regarding imaging technologies, it is still sought to extend such imaging technologies to support an automated workflow and/or automated diagnostic processes. In this regard, it is desirable to develop techniques that may automate interpretation of culture plate images (e.g., identification of growth, identification of species, susceptibility testing, antibiotic sensitivity analysis etc.) and determine the next steps to be performed based on the automated interpretation. However, development of automated image processing logic (e.g., software) for diagnostic indications can be time consuming given the diversity of specimen types and organism taxa.
Disclosed herein is a system for evaluating biological specimens for the presence of pathogens, the identity of those pathogens and other related analysis and evaluation. The objectives of the systems are speed of analysis, accuracy of the analysis and process automation. Such systems typically obtain digital images of specimens disposed on nutrient media and incubated to discern evidence of microbial growth, such growth providing an indication of the presence of pathogens in the biological specimen. Such systems are described herein as the Kiestra Systems and include apparatus such as a camera and lighting to obtain one or more images of the incubated specimen, bar code readers for the specimen containers (e.g. petri dishes that contain inoculated plated media), etc.
Such systems communicate with and are controlled by one or more customer-centric Imaging Application (Imaging-related Apps or Apps). Such Imaging Apps may be software that utilize data derived from a collection of historic specimen images to analyze new specimen images for automated identification and/or diagnosis of disease state. The Apps can provide a clinical tool for rapid specimen characterization and reporting of results. The Apps link clinical specimens (from a clinical site), to non-patient identifying facts about the specimens (e.g., non-patient identifying specimen origin information like demographics), process conditions (e.g. incubation time and temperature), process materials or environment (e.g. nutrient media) and/or non-patient identifying test results for the specimen (no-growth of pathogens, growth or otherwise positive identification of the pathogens, enumeration of the identified pathogens). The facts and conditions are referred to collectively as analytical information herein. Classifying such analytical information in this manner ensures that only the most relevant historical processing information is used to develop an app. Therefore, each developed app is created for a narrowly defined purpose and only deployed when the specimen classifiers correspond to the app classifiers. It is advantageous if the data is not linked to information that would identify the patient (i.e. patient deidentification). In one embodiment, the system automatically deidentifies the specimen information and information in response to certain conditions. Deidentification includes providing metadata that links to non-patient identifying classifying information (e.g. geographic region from which the specimen was obtained, type of specimen, etc.) that, in effect, permits the data to be used in systems and methods that cannot retain confidential patient information. The system may include Apps that provide time series processing of images, classified/trained and tested diagnostic/evaluation algorithms and/or expert systems. An App may include a module wherein a module may be understood to be one or more processes or algorithms for a particular purpose using, for example, image metadata (metadata serves as the classifications described herein) and rules. In some cases, multiple modules may be implemented as a package. In some embodiments, the data may be stored in a data base that can be used to develop an App, train an App, certify an App, test an App, etc.
The system may provide image analysis processes that may incorporate best practices, addresses relevant specimen types and automated picking.
To significantly expedite time to market for valuable Imaging Apps, a multi-fold or modular system may be employed. Such a multi-fold system may include modules such as:
The Apps described herein can be further developed and refined as specimens, media and organisms are validated and obtain regulatory approval. For example, a quadrant quantitation App could be implemented initially for throat swabs and wound, and later for perianal or other specimen types as more specimens are processed. This ongoing training/development of the Apps provides a robust cadence of new Imaging Apps with increasing usefulness. For example, as more images are evaluated, the App “learns” how to differentiate colonies from background in the image of the plated culture. Processing image information to differentiate pixels associated with the image of the colony from the pixels associated with the image of the background are described in the Kiestra Systems referenced previously herein.
Also described herein is a development system that provides high-value software solutions while reducing time to market and maximizing utilization of resources. An important component of the system may involve initiating a collaboration program that engages select clinical sites running imaging systems (e.g., Kiestra Systems) to gather clinical specimen imaging information that is classified by association with meta-data for development (algorithm training) and validation (submission-ready) purposes. This collection of data may feed automated algorithm refinement, improving performance and decreasing development time. Validation uses a sequestered collection of images in the database. These data can also be reused as additional features are implemented. However, the information is only deployed when the data classification corresponds to the App classification/purpose. This approach creates substantial efficiency, flexibility in App scope (i.e., versioning) and a highly-valuable resource.
For non-selective media, classification buckets are created that characterize the complexity of the population to identify cultures as no growth, pure, predominate or complex/mixed rather than attempting to identify all sister colonies of all taxa on all media types. Sister colony identification is technically very challenging and time consuming, particularly on non-CHROMagar media. Mixed cultures typically require expert knowledge to interpret, and even with a level of automated image analysis, would likely require review for confirmation and release. By using the pure, predominate and complex/mixed categories most specimens can be characterized and those with pure or predominate colony populations can be automatically worked up with a high degree of confidence. These colonies can then be chosen by the App for automated picking by a picking system. This approach largely renders the need to individually characterize and define specifications of each specimen type (e.g., sputum, wound, swab, etc.) by/verses a particular pathogen, to a more generic classification strategy that efficiently leverages the imaging algorithms and enables most specimen types to be characterized.
Example Imaging Apps are summarized below. Such applications may include
For example, in response to a MRSA (Methicillin-resistant Staphylococcus aureus) screening analysis, the following actions may include: (i.e. for a negative MRSA screening result): i) empiric treatment with a certain set of antibiotics not used with resistant Staph aureus; ii) not sequestering or specifically managing the patient; iii) proceeding with a certain next medical step such as surgery; iv) the ordering of certain additional diagnostic tests; and v) guidance in selection of a certain set of likely effective antibiotics for determination of antibiotic susceptibility testing. An application developed according the method described herein can guide a user to some or all of the above actions based on historical analysis of prior specimens that share certain predetermined criteria with the specimen being evaluated by the App.
Actions from a Rapid Detection Imaging App may include: i) communicating to the treating physician the detection of threshold levels of growth indicating infection many hours sooner than standard methods of practice; ii) initiating diagnostics that rapidly determine the identification of growing pathogens (e.g., via MALDI-tof); iii) communicating the pathogen identification to the physician substantially earlier than standard practice; iv) initiating diagnostics for determination of antimicrobial susceptibility and communicating antibiotic susceptibility profiles substantially sooner than current practices. These objectives are accomplished by using prior images and the results of that image analysis to inform analysis of the present image. The prior images must be classified to develop an App that is sufficiently skilled and reliable to be used to evaluate the present image. Note that, if the image analysis performed by the App results in a positive detection of microbial growth, in one embodiment the App can communicate with an analyzer that will receive the specimen under evaluation and identify the colony or colonies on the specimen to be picked for further analysis. The App can either instruct downstream process or control downstream processing Such downstream processing includes: i) preparing a suspension of the one or more picked colony into a pre-determined buffer or solution; ii) adjustment of said suspended cell suspension to a predetermined cell concentration; iii) the spotting of said cell suspension to a substrate (e.g., MALDI Plate); iv) overlaying of said spots with one or more reagents including, for example, MALDI matrix solution, extraction chemicals; v) distributing one or more aliquots of said cell suspension into wells of an antimicrobial susceptibility plate for determination of susceptibility profiles to a series of antibiotics at various concentration; and vi) distribution into suspensions for analysis by PCR, sequencing or other molecular diagnostic tests. In alternative embodiments the App instructs but does not control follow-on sample processing/analysis.
In order for the Apps developed according to the methods described herein to be deployed to control processes and evaluate specimens, the Apps must be developed using relevant data and analysis. Methods and equipment that can be used to obtain relevant data and analysis include, clinical sites collecting images and reference data, that are used to build data sets for future Imaging Apps training and validations. See
The digital cameras (which are conventional cameras with good megapixel resolution, e.g., 5 MP, 25 MP, etc.) may be implemented for adequate performance for early growth/no growth and presumptive identifications (IDs). For example, for a Colony greater than or equal to 5 mm diameter detected by the imaging module or apparatus that captures the digital image of the inoculated plate, this data could be combined with the Growth/No Growth Detection to indicate when growth occurs with sufficient colony size for further processing. The imaging apparatus may itself be an App that performs analysis of the digital image, such as described in the Kiestra Systems references elsewhere herein. According to those systems and methods, colonies are differentiated from background and from that image analysis the density of the colonies on the inoculate plate is determined and communicated to an App that will determine further actions in response to the image analysis. This information could then be used by the App to determine what colony to pick from the image automatically (without an operator reading the plate or identifying the colony for pick). The Apps are developed and deployed for specific plate environments. For example, pure plates (one colony type) and predominant plates (more than one colony type but predominantly one colony type) can have representatives of each colony type indicated by the purity plate module with associated rules that determine further workup. Plates determined to be complex are processed by a different App (or an App that fires different rules). A pre-determined number of each colony type could be designated for automatic identification (ID) and Antibiotic Susceptibility Testing (AST) workup on an ID/AST module that may involve an automatic picking system/robot.
To decrease product development cost and reduce time to market Best Practices support may be adopted. Specific media (below) may be used to obtain optimal recovery and performance on the platform (e.g., BD Kiestra systems).
The system may involve processing of plated media readings performed in a lab. Plated media are prepared as described elsewhere herein. The media type and sample taxa are examples of specimen classifiers that inform the metadata linked to the specimen information. The selected App may provide presumptive ID with picking recommendations of only the pure or predominant colony types on the plate. Specimens with multiple clinically significant isolates are infrequent and often complex. Specimens identified as complex may be put into a mixed category for manual review and actioning.
A set of algorithms may be designed that are more generic rather than specific to an App configured for use with a specific type of specimen or the suspected target species contained in the specimen. These algorithm tools are not described in detail herein but can be developed and deployed by those skilled in the art. The algorithms are validated as a process for deployment in Apps designed to evaluate and process a range of specimen types. When such tools are not limited to specific specimen types, they can be used to review a larger number of plates or as building blocks of other Apps.
The system develops a set of tools that provide an overall capability that can be supplemented in the future as more specimen processing data (e.g. imaging data) is obtained and added to the database with the associated classifications. These tools will evaluate plates and provide a specific result independent of specimen type. A compilation of these results along with specimen and patient demographics will be processed by the App to provide an indication of the specimen quantity. This simplifies execution and leads to a faster time to market while providing the user with a level of flexibility. Development of a more generic toolbox of image analysis algorithms (together with the classification metadata, and rules termed a Module) that enable the App can be rapidly matured or optimized (for a particular specimen, for example) by deploying artificial intelligence algorithms such as neural networks, artificial neural networks or deep learning algorithms that use the classified specimen data in the database to make determinations about the subject specimen. As part of the exponential strategy, artificial intelligence may be employed that “automatically” determines what attributes of the image, and with what algorithms, to best provide the desired Module capability. Individual Modules may have sufficient value to be launched as an App, or multiple Modules may be packaged into an App. In some versions, a set of modules may be differentiated to make the following specimen analysis based on the specimen information obtained and its classification: (a) Growth/No Growth; (b) Semi-quantification; (c) Presumptive ID (d) Pure, Predominant, Complex (e) Antibiotic sensitivity (Kirby-Bauer test) (f) Sister colony locator.
Example applications (Apps) may include Key ID, Rapid Detection, CHROMagar ID 2.0, Quadrant Quantitation and Presumptive ID. These may be summarized as follows:
Key ID is a tool that identifies specific species on specific plates where any numbers of colonies are present. Any number of colonies typical of the Key ID app will be designated in pure or mixed cultures. Examples are:
The Rapid Detection App provides rapid detection processing. Growth at any reading point can be used as a flag for growth detected as early as possible on any plate. The user selects the reading point that will serve as a flag, recognizing the trade-off between an earlier reading point that may be less reliable but delivers results more quickly and a later reading point which may be more reliable but take more time to detection. The action taken will be dependent on the reading points chosen and the rules invoked. Early detection may happen as early as 4 hours, but incubation periods of 6 hours, 8 hours or longer are contemplated. Incubation times for a particular specimen are readily ascertained by one skilled in the art. The system and method described herein is not limited to any particular incubation time. The Rapid Detection App is beneficial for rapid positives of critical specimens.
For example, growth is detected at 8 hours on a BAP plate. A rule would take these results and if the specimen type is CSF (Cerebrospinal fluid) the lab would be alerted immediately if the App determined that the specimen had growth on the plates because CSF is typically sterile. This determination requires the App to make some sort of alert in response to a determination that the CSF specimen is positive for pathogens. For other specimen types, i.e., sputum, early detection would have little value since almost all culture have normal flora. In these cases, no rule would be written for specimens so classified and no action would be taken by the App based on rapid detection. Thus, the type of plate and type of detection can trigger different automated processes.
Another embodiment is a CHROMagar ID App that quantitates urines or other specimen types. This App obtains classified image information that will enable the App to presumptively identify the pure or predominate colony type on CHROMagar Orientation. Mixed plates will also be identified but the App may fire different rules in response to the determination of a complex plate. Using the rules engine, the processed images of specimens can be sorted into several categories. Certain categories permit auto reporting of results. Those categories are described elsewhere in detail herein.
The Quadrant Quantitation and Presumptive ID app is a tool that will evaluate all positives to classify each as one of (1) no growth, (2) pure, (3) predominate or (4) complex and the overall quantity on the plate. The pure or predominate isolates will be presumptively identified and colonies identified for picking. In this case all cultures would be analyzed and would fall into several categories with some being auto reported and others sent for review. The plates with significant pathogens could be sent to other systems (e.g., picking or testing) for picking with no customer/clinician intervention. The Table in
The environment (e.g.,
Once the plates are evaluated and results are combined at the specimen level another set of rules would drive auto reporting, user, Laboratory Information System (LIS) or Laboratory Information Management System (LIMS) alerts, and send significant isolates to a worklist or to a picking system such as for ID/AST testing.
By targeting organisms on media independent of the specimen type complete systems may be developed more efficiently and rapidly. For example, E. coli may be considered the same genus and species no matter the specimen source for the E. Coli. Regulatory approval of some Apps by specimen type could be difficult since positives for some specimens (i.e. CSF and other sterile sites) are rare and therefor there would not be sufficient historical sample processing/image data from which to develop an App that would be clinically reliable. However, over a period of time and with the benefit of images and test results from those images obtained from a variety of sources (i.e. clinical trials, regulatory submissions), Apps for even rare specimens can be developed.
One embodiment is a method for processing a biological sample including the steps of obtaining a biological sample, combining the biological sample with nutrient media, incubating the biological sample, obtaining a digital image of the incubated biological sample, classifying the digital image according to analytical criteria selected from the group including of specimen origin information, clinical sample criteria, process materials and process conditions, obtaining data from historical digital images of incubated biological samples on nutrient media that share at least one of the analytical criteria assigned to the digital image and using the historical digital image data to output an instruction to a user for further processing of the biological sample.
The specimen origin information includes geographic information regarding the biological sample source and the type of biological sample. The process materials include the type of nutrient media. The historical digital image data is classified by at least one of specimen type, organism taxa or culture media type. The method further includes analyzing the digital image data and determining, from the analyzed data if the digital image reflects microbial growth. In response to a determination that the digital image does not show microbial growth, the method outputs an indication of no microbial growth. In response to determining that there is an indication of microbial growth, the method performs the step of determining if the specimen is a sterile specimen and identifying one or more coordinates of a colony of microorganisms in the image that was a basis for the indication of microbial growth. In response to determining that the specimen is sterile, the method includes the further step of indicating that the specimen is a high value positive and sending instructions to further process the specimen. Examples of further processing includes identification (ID) testing, antibiotic susceptibility testing or both. The coordinates of an object in the image classified as a colony are communicated to a module that forwards those coordinates to a picking apparatus that will pick the colony from the biological sample. A module may be an App or a combination of Apps as described herein. The biological sample is transferred to the picking apparatus and the colony is picked from the biological sample wherein the steps of transferring and picking are either controlled by the module or the module has issued an instruction to perform such steps. In response to determining that the specimen is not sterile, the historical image data is compared with the image data to identify a specific predetermined species of microorganism in the digital image of the incubated biological sample. The step of comparing is performed by the module, and, if the module determines that the specific predetermined species of microorganism is present in the image data, the module reports the identification of the specific predetermined species. When such a determination is made the module flags the specimen for further review.
The module compares the historical image data with the digital image of the incubated biological sample and determines an amount of microbial growth where the comparing and determining steps are performed in the module in communication with the imaging apparatus that obtained the digital image of the incubated biological sample. According to the method, a module or App determines if the microbial growth is one of pure colonies, predominant colonies, or complex colonies by communicating the digital image of the incubated biological sample to a module that determines a growth level as a vector of three probabilities. In response to a determination that the colony is pure, the module reports that the plate is pure. If the growth on the pure plate exceeds a predetermined threshold growth the biological sample is identified as a high value positive and the module communicates that information to a user of the system. The module, having received the coordinates of the colony from the imaging apparatus, will communicate the coordinates to a picking apparatus that will pick the colony from the biological sample. The method may also include transferring the biological sample to the picking apparatus and picking the colony from the biological sample where the steps of transferring and picking are controlled by the module or requested or required by the module. If the module determines that the growth does not exceed the predetermined threshold the module provides a presumptive identification of the colony based on comparing an image of the colony provided to the module with the historical image data accessed by them module. In response to the determination the module performs the further step of reporting the presumptive ID to a user.
If the module determines that the growth does not exceed the predetermined threshold the module performs the further step of reporting to a user that the complex sample does not meet or exceed the positive growth threshold. In response to a determination that the colony is predominant, the module provides a presumptive identification of the colony based on comparing an image of the colony provided to the module with the historical image data accessed by the module. The module performs the further step of reporting the presumptive ID to a user.
If the module determines that the growth exceeds a predetermined threshold growth, the module identifies the sample as a high value positive. The method also includes alerting a user of the high value positive. The method may also include identifying the coordinates of the high value positive. The method may also include communicating the coordinates of a colony to a module that communicates the coordinates to a picking apparatus that will pick the colony from the biological sample. The method may also include the module controlling or issuing instructions to transfer the biological sample to the picking apparatus and picking the colony from the biological sample where the steps of transferring and picking are controlled by the module. The method may also include alerting a user that further review is required. If the module determines that the growth does not exceed the predetermined threshold the module performs the further step of reporting the presumptive ID to a user. In response to a determination that the colony is complex, the method further includes: reporting, from the module, that the plate is complex. The method may also include determining, by the module, if the growth exceeds a predetermined threshold growth. If the module determines that the growth exceeds the predetermined threshold growth, the method further includes: alerting a user that further review is required.
The present disclosure provides apparatus and methods of an environment for developing imaging applications for identifying and analyzing biological specimens such as microbial growth. Many of the methods described herein can be fully or partially automated, such as being integrated as part of a fully or partially automated laboratory workflow.
This document provides a description of the design and implementation of a system that accelerates delivery of automated imaging capabilities to biological imaging systems such as the BD Kiestra™ System. The imaging capabilities of such systems are enabled by a suite of hardware, software, analytical algorithms, and clinical rules. An example of one such commercialized system includes one or more digital cameras (e.g., 4 MP) with multiple illumination configurations that generate an optimized and standardized image based on an appropriate platform. The systems described herein are capable of being implemented in other optical systems for imaging microbiology samples. There are many such commercially available systems, which are not described in detail herein. One example may be the BD Kiestra™ ReadA Compact intelligent incubation and imaging system. The Kiestra™ ReadA compact is an automated incubator with an integrated camera and plate transport system that enables automated imaging of plates. The ReadA compact is commercially available. The ReadA compact also has integrated plate import and plate export devices that couple the incubator to other instruments for manipulation. Therefore, in some embodiments, in response to the analysis of the digital images by one or more of the Apps, the Apps can issue instructions and control the incubation of the relevant specimen under evaluation by the App. Other example systems include those described in PCT Publication No. WO2015/114121 and U.S. Patent Publication 2015/0299639, the entirety of which is incorporated by reference herein. Such optical imaging platforms are well known to those skilled in the art and not described in detail herein.
A series of Apps for the system can provide analysis of the images from most specimens, generated at various predetermined times, such as in the ReadA Compact. The system can enable downstream actioning of these plates-including automated release of no-growth and/or negative plates, and automated characterization of colonies for definitive ID and AST analysis.
At a high level, the imaging analysis tools (Imaging Apps) can be deployed to enable a collection of different results that provide substantial value to the lab and/or actionable clinical results. These Apps utilize one or more image analysis algorithms (Modules), as well as a set of rules that provide information on how to apply the Modules on specific media types and/or specific specimens. Certain Apps can also have associated Expert Systems, which overlay an additional set of rules on more basic App determinations, typically regulatory/clinical guidance, that inform recommendations for actioning and interpreting the result.
Developing Imaging Apps can utilize an iterative approach. A data lake is developed using either specimen information acquisition or images of clinical specimens being processed as normal practice in the clinical lab, or both. Algorithms are developed to model conclusions/instructions/outputs such as those illustrated in
However, a biological image processing application development system 100, such as one illustrated in
Where the Data Lake is stored is largely a matter of design choice. The Data Lake can be stored locally or in a cloud. The data in the Data Lake can be partitioned. For example, the data in the Data Lake could be segmented depending on how the data is accessed and/or used. In one embodiment, one segment of the data in the Data Lake could be for algorithm training, another segment could be used for data verification or validation and yet another could be used for clinical submission.
The database of the system can store and provide classified data of these clinical images and also linked data, that can be pulled individually as appropriate for algorithm development, formal verification and validation, or clinical submission on demand. The App can be established with a level of global standardization including lab protocols, media types, imaging time, quantitation scoring, etc. Resources can supplement the generation of images and reference data (specimens and/or spiked/contrived samples) for specimens/organisms/conditions that are infrequently encountered in the clinical setting. This proactive strategy for data generation allows an almost on-demand prioritization and development of particular Apps for specific specimen or media types. The system architecture permits integration of the new modules into existing laboratory site software to ease release of new imaging diagnostic functionalities (Apps/modules/packages). This may be achieved by standardizing the interface between new Apps/modules/packages with existing/previous imaging software systems. In this regard, the Apps/modules/packages may be add-ons to system software such as a system software for an automated imaging system (e.g., in a processing system that controls any one or more of a plate/sample conveyor, incubator, camera, picking machine and/or related robotics, for moving such samples/plates within such automated laboratory cell/equipment etc.). Certain specimens determined to be negative for pathogens can be spiked with known pathogens and the subsequent image of the incubated, spiked specimen characterized as described herein. The images of the spiked specimens can then be used as a training set for Apps that can be used to evaluate and process less frequently occurring pathogens.
This approach can provide maximal flexibility in prioritization and the cadence of App development. It would also make it possible to provide early App performance metrics to help better understand the added value of an App, the synergies at the solution level, and enable earlier recognition. The Data Lake may be generated such as with one or more hospital systems that meet a list of Corporate Clinical Development (CCD) criteria (i.e. technical and medical information). Implementation provides the ability to store and classify the data, query, and audit the database; lab protocols that define the program and processes; training, monitoring, compliance, quality metrics etc. as is typical for a clinical trial- and may be done as part of the development of the Data Lake, rather than at the end of a typical product development process. The approach may implement dedicated clinical lab resources to determine certain plate results outside standard protocols, and to link images with analysis results to the Data Lake. Certain specimens, plate types or image acquisition time points may need to be run specifically to develop the Data Lake in special processes that may be independent of normal/typical lab practice. Thus, in some versions, the Data Lake database may comprise images of clinical specimens with linked manual/standard analysis of truth (e.g., quantitation, IDs, result interpretation) and metadata associated with classifications for the image (e.g., select patient demographics, imaging time and conditions, media types, etc.).
Establishment of both the algorithms and Data Lake may include a certain level of standardization. This will also define what any given App is validated for, and the analysis/instruction/output that may ultimately be obtained. Given the diversity of media types, vendors of media, and incubation times in use across labs globally, a “Best Practices” approach may be used to initiate this effort. Additional conditions may be added in the future by stocking the Data Lake with appropriate data. An example of a Media x Specimen matrix is summarized in
To ensure database accuracy, population of the data into the database may involve independent human image analysis of images performed by several individuals, or plate analysis done manually by human technologists. Human readings may be compared with clinical laboratory reports. In some cases, a further image review may be involved for discrepant readings. Database input may involve de-identification of patient information from image related data. Review of images for data entry may involve human scoring of the growth on plates for pure, predominant, complex and no growth; quadrant quantitation.
In a typical example, the Data Lake may contain media plate images, linked to lab-determined quantitation (e.g., no growth, +, ++, +++). It may also contain identification (ID) of organisms determined to be of significance for the kind or type of specimen (such as ones that may be significant a trained clinical microbiologist). The Data Lake may also contain image-based metadata, and in some cases, patient demographic information. Organisms not routinely identified as pathogens (i.e., normal flora) may also be requested to be identified to facilitate algorithm development. Once populated, a portion of images will be leveraged to train and test appropriate algorithms toward App development.
At a high level, the imaging analysis tools can be deployed to enable a collection of different results that provide substantial value to the lab and/or actionable clinical results. These Modules (Mods) are comprised of one or more image analysis algorithms, as well as a set of rules that provide information on how to apply the Modules on specific media types and specific specimens. Some Modules, such as Screening-MRSA, may be implemented as an App. Other modules may more often be packaged with other Modules to provide a synergistic capability (for example, quadrant quantitation and purity will often be packaged as an App.) Some Apps can also have associated Expert Systems, which overlay an additional set of rules, typically regulatory/clinical guidance, that inform recommendations for actioning and interpreting the result (e.g. KB Zone described herein). Based on technical and clinical considerations, one or more Apps may also be bundled as components of a Launch Package depending on different clinical lab needs. It is also anticipated that some Apps will have versions (e.g., UCA V1.8 with FDA approval will become UCA 2.0).
Some examples of the algorithms and Modules (collections of synergistic algorithms) are generic (generally working across specimens, pathogens, media) and are summarized below and may be considered in relation to the processes of the illustration of
Growth App 1010 (See
A Key Id Module 1020 (See,
In some versions, the system may include a Screening and Critical Pathogens module(s) developed on a pathogen basis. These modules may provide detection of specific pathogens, groups of pathogens or those with specific properties. The Screening Apps may be implemented to enable identification of specific pathogens on CHROMagar, and can be used for patient management as well as pathogen characterization e.g., MRSA, ESBL, CPE, VRE etc. CHROMagar may enable a collection of pathogens to be presumptively identified i.e., CHROMagar orientation for both Gram Negative (GN) and Gram Positive (GP) bacteria. Pathogen specific media may be used for specimens—such as SS media for Salmonella, Shigella from stool. Certain organisms may be presumptively identified or flagged on more generic media based on, for example, Hemolytic properties on blood agar, or unique morphologic properties on specific media. A potential collection of modules with pathogen x media capabilities is summarized in the Table of
Based on a streaking pattern, e.g., a BD Kiestra™ InoqulA™ quadrant streaking pattern, this Quadrant Quantitation module 1030 (See,
With respect to growth type, the images/plates may be characterized as pure, predominant and complex. This may be based on a minimal number of isolated colonies of each type. For example, predominant growth may be greater than (or equal two) two colony types, where one type is greater than a factor (e.g., 10 times) the other(s). Complex may be greater than two colony types with no predominate isolate or if the isolate is not identifiable within a presumptive ID table such as the example of
For example, a colony forming unit (CFU)/mL Quantitation App/Module 1040 (See,
In general, the Quantification module may, in some versions, determine if growth is due to a single growing organism, a predominant organism or a mixture of (multiple) organisms. A pure organism may be deemed to be an organism responsible for ≥99% of the observable/imageable growth. A predominant organism may be an organism responsible for (90%, 99%) of the observable/imageable growth. A purity level may be returned as a vector of probabilities (e.g., 3 probabilities as pure, predominant, complex) ranging on [0,1] and summing up to 1. In the case of pure or predominant growth up to five colony locations for the main organism will be given in decreasing probabilities.
Examples of responses to determined quantification are as follows:
1) An image of a specimen on a plated media is determined to have greater than a predetermined threshold (100,000 CFU/mL) of mixed flora. The response of the App to this determination is to flag this plate as a complex plate, because the plate has over the threshold amount of mixed flora and recommend manual review of the plate. Such a determination is not made in the context of media or taxa so this is an App of broad applicability and not limited in deployment to specific media or taxa on the plate.
2) An image is evaluated and determined to exhibit no growth for 24 hrs. If the specimen is classified as a critical specimen the App issues a preliminary report of no growth and either recommends or controls the re-incubation of the plate for another 24 hours. If a subsequent image detects no growth in 48 hours, the App sends a final report to the user of no growth after 48 hours. The App either recommends or controls discarding the plate
3) An image of a specimen on a plated media is determined to have greater than a predetermined threshold (100,000 CFU/mL) of a pure growth and a colony size that exceeds 0.5 mm. The response of the App is to issue an instruction or control the pick of colony for ID and AST testing. The App will flag the specimen for review by the technician and will send the greater than 100,000 CFU/mL report to the physician associated with the specimen.
4) An image of a specimen is determined to reveal the presence of MRSA by the App. The App sends a report that MRSA is detected and adds the specimen to a positive MRSA work list. If the App determines that the size of the MRSA colony exceeds a threshold (e.g. greater than 0.5 mm) the App will issue an instruction or control sending he sample for ID and AST testing. As noted elsewhere herein, ID and AST have their own specimen work up and evaluation. As such, ID and AST systems and apparatus are typically downstream of the incubation/imaging apparatus (e.g. Kiestra™ ReadA compact).
5) An image of a specimen classified as ESBL is determined to show no growth. The App will issue a final report that no ESBL isolate was detected and will issue an instruction to discard or control discard of the plate.
6) An image of specimen classified as sputum is identified as having mixed flora (therefore a complex plate) that exceeds the threshold amount. When the App determines that the plate is complex it issues an instruction for a technician to review the plate. Note that different thresholds for mixed flora that trigger the requirement for manual review might be deployed depending on specimen classification.
7) An image of a specimen classified as critical on blood agar is determined to reveal growth. In such an instance, the critical specimen App would fire and send an alert to the physician associated with the specimen and cause the specimen to be added to the critical sample work list. If the App determines that a colony greater than a threshold size (i.e. greater than 0.5 mm) and specimen is classified as being disposed on MacConkey agar, then the App causes the specimen to be sent to auto pick for MALDI and for Gram negative AST. The App will also send a report indicating that a Gram-negative specimen has been isolated.
8) An image of a specimen reveals a colony number that is greater than 100,000 CFU/mL and classifies the image as pure and having a colony size greater than a predetermined threshold (e.g. greater than 0.5 mm). In response, the App causes the specimen to be auto picked for AST, causes the specimen to be added to the positive review list by a technician and causes a report to be sent to the physician associated with the sample that indicates that more than 100,000 CFU/mL of a colony was detected from the sample. Further, if the AST results reveal that the picked colony is resistant to Carbapenem, then the App causes a molecular confirmatory test to be performed.
9) An image of a specimen is determined to possess heavy predominant growth. In response, the App causes the growth to be auto picked and a suspension prepared for performing MALDI on the picked sample. If the MALDI identifies the colony as E. Coli, then the App causes the sample to be further evaluated for Gram Negative AST (using either the MALDI suspension or a new pick of the colony).
In some versions, growth detection may be implemented as two modules where one evaluates plates to determine growth/no growth and another module evaluates growth quantity (+, ++, +++) in 3 or more colonies in any particular quadrant. For example, a first module evaluates an image of a critical, normally sterile specimen. If that evaluation reveals growth at a predetermined time point, and determines that a colony size is greater that a predetermined threshold size, then the App identifies coordinates of representative colonies and issues instructions to the imaging apparatus (e.g. ReadA) that the plate is to be moved to an apparatus that will auto pick the identified colony. The picked colony is resuspended in a solution to a predetermined density for further testing in, e.g., a molecular diagnostic apparatus or test (e.g., PCR, Sequencing).
In case of pure or predominate growth, a Presumptive ID module 1050 (See,
For example, the media and colony identities may be those indicated in the Table of
In some versions, the system may implement a purity plate module. Pure, predominant and complex plates may require a minimal number of isolated colonies in each. Predominant growth has typically greater than two colony types, where one colony type is greater than ten times the other colony type. Complex type typically has greater than two colony types with no predominate isolate or if the isolate is not identifiable (presumptive ID table below). Complex plates are typically manually interpreted. Thus, the module may classify the plate image according to whether it is pure, predominant (which can be slightly mixed) and/or complex.
Pure and predominant plates can have representatives of each colony type indicated by the Purity Plate Mod with associated Rules that drive further workup as set forth in the examples above. A pre-determined number of each colony type could be designated for automatic identification (ID) and AST workup on an ID/AST module that may involve an automatic picking system/robot.
Some embodiments may utilize an optional measurement App. Such an App may leverage existing imaging capabilities and the AST Expert Systems to provide zone measurements. Optionally these measurements may be linked to expert systems to provide interpretations. Opportunities also exist for a version of this App for Early zone measurements for particular drug/organism combinations and for zones on media plated directly from positive blood culture. Such algorithms may be based on metadata and/or images of the Data Lake.
In some versions, this App would leverage existing imaging capabilities and an AST Expert System to provide zone measurements and Abx Disk identification. Optionally these measurements could be linked to expert systems to provide guidance on an antibiotic susceptibility profile for a pathogen isolated from the patient and guidance on treatment/response. In some versions, implementations of this App may provide early zone measurements for particular drug/organism combinations and for zones on media plated directly from positive blood culture. Some Apps may include expert systems (interpretations) and could be facilitated significantly with the Data Lake being stocked, monitored, audited appropriately and with required metadata and images.
Although a system 100 may include any one or more of the above imaging related modules/Apps, in some versions a particular segmentation of functionality of the modules may be implemented by the following set of discrete modules/Apps: (a) Quadrant-Quantitation; (b) Detection of No Growth; (c) Purity Plate: # Colony Types (e.g., pure, Predominant, Complex) (d) Screening (e.g., CHROMagar i.e. MRSA); (e) Critical pathogens; (f) Early Growth Detection; (g) Auto Select ID/AST and (h) Zone measurement Kirby-Bauer.
The system 100, providing the combination of algorithms/modules/rules, the Data Lake, and the ability to extract predetermined subsets of data, can enable an on-demand ability to rapidly mature algorithms and develop Apps. As an example, Apps may be developed based on specimen type and may be implemented with a collection of Apps. The strategy to implement a collection of Apps is influenced by many factors: supporting software launch cadence; value of individual Apps vs Apps being together; the availability of certain algorithms or specimen/plate/organism types in the Data Lake, etc.
An example may be considered in relation to the following table:
In this specimen-based example, a series of 5 different Modules concerns one specimen type. Apps validated against a particular specimen type is one way to package functionality, however, an Exponential approach will also allow other options. For example, Surveillance Apps will allow launch by particular targeted organism (MRSA, Streptococcus, Shigella); Quantitation Apps could be packaged by media type (quantity of the sample on blood agar, independent of specimen), etc. In this sense, however, certain Apps will likely have minimal value for certain Specimens (i.e., quantity of the sample on nonselective media with sputum, given high normal flora levels). Note that the Apps can be delimited by region with specific rules and specific functions limited to the geographic region from which the specimen under evaluation was obtained. Table 1 identifies functions specific to clinical requirements particular to the United States (US) and Europe (EU).
Thus, potential Apps may be segregated into two-high-level buckets. A first bucket collection may be considered screening and key identification Apps. Such Apps typically target specific organisms on CHROMagars, or for high-value pathogens on, for example, Blood agar. Each of these are discrete and can be prioritized for development with minimal impact to other Apps or specimen types and launched individually if desired. Similarly, the Kirby-Bauer zone App is generally independent of the next generation Apps (“Next Gen App”), which have associated algorithms that can be independently prioritized. Additionally, as new CHROMagars (i.e., vancomycin resistant enterococci (VRE)) become available, appropriate specimens can be run to populate the Data Lake and be added to this list. If targeted isolates are relative rare, the Data Lake may be supplemented with contrived (spiked specimen) samples. Additional example screening and Key ID Apps are illustrated in the following Table.
Salmonella and
Shigella on
Salmonella and
Shigella
N. gonorrhea
N. gonorrhea
Haemophilus
Haemophilus
It can be observed from Table 2 that Apps can be quite specific and that the outputs can depend on specimen classification (i.e. type of specimen, US region or EU region, etc.). Table 2 also illustrates at a high level, the type of data used to train the App.
A second bucket collection of Apps use more generic algorithms and can be prioritized and grouped for launch by several criteria. A summary of examples of these Apps is provided in Table 3 below. In essence, each cell in Table 3 represents an App. Cells in the Table below that share the same number are appropriate capabilities for that specimen type where the shared cell numbers are reasonably packaged together in a common module. With this model, there are 8 additional launch packages/modules.
Example Imaging Modules May be Considered with Reference to the Following Table:
EUCAST is the European Committee on Antimicrobial Susceptibility Testing. In one example of a process integrated with one or more Apps for specimen evaluation and process control, a specimen is inoculated onto a plated media using BD Kiestra™ InoqulA. The specimen is streaked onto the media using a predetermined pattern which is tracked as part of metadata via the bar code. The streaked sample is conveyed into BD Kiestra™ ReadA compact where it is incubated and imaged at times determined by the App. The images obtained by ReadA at the appointed time is analyzed by the App to determine if the specimen on the plate is pure. Further work up of the specification is performed based on the results of the determination. The image is evaluated to identify the coordinates of the selected colonies. The App can convey those coordinates to an apparatus (or technologist). The App can issue instructions to convey the specimen to the apparatus in which the colony will be picked. The App can further coordinate or control the picking of the colonies and conveying the picked colonies to another platform where ID of the pathogen is performed. In one example ID is performed by MALDI. As described elsewhere herein, a sample is evaluated by MALDI by placing the picked sample into suspension and inoculating a MALDI plate with the suspension. The App can also coordinate or control transfer of the colony suspension to BD Kiestra™ InoqulA. There the suspension can be inoculated onto another type of culture media (e.g., Mueller Hinton) using a “spread pattern” and then moved to an AST testing apparatus where predetermined antibiotic disks (e.g. BD BBL™ Sensi-Discs™) are place on the culture. The plate carrying the inoculated specimen and the antibiotic disks is then conveyed to a ReadA compact under coordination and control of the App. The ReadA obtains images and provides those images to the App, the results of which are conveyed from the App to an Expert System for analysis of the resulting antibiotic disk zones and interpretation of the results. The Expert System then conveys the results of the analysis to the clinical lab staff
Although the invention herein has been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present invention. It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims.
This application claims the benefit of the filing date of U.S. Provisional Patent Application No. 62/568,579, filed Oct. 5, 2017, the disclosure of which is hereby incorporated by reference herein.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/US2018/054368 | 10/4/2018 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 62568579 | Oct 2017 | US |
