Application development environment for biological sample evaluation processing

Abstract

A system and method for developing an application (App) for automated assessment and analysis of processed biological samples. Such a sample is obtained, combined with a nutrient medium and incubated. The incubated sample is imaged and the image information is classified according to predetermined criteria. The classified image information is then evaluated according to an application derived from classified historical image information in a database. The classified historical image information is compared to the classified image information to provide guidance on further processing of the biological sample via an application tailored to the processing, the tailored sample processing guidance being provided according to the classification assigned to the image information.

Description

[0001] Cross Reference to Related Applications

[0002] This application claims the benefit of the filing date of U.S. Provisional Patent Application No. 62 / 568,579, filed October 5, 2017, the disclosure of which application is hereby incorporated by reference herein in its entirety. BACKGROUND

[0003] Digital imaging of culture plates is increasingly of interest, for example, for microbial growth detection, colony counting, and / or identification. Systems and techniques for imaging plates for detection of microorganisms are described in PCT Publications Nos. WO 2015 / 114121, WO 2016 / 172527, and WO 2016 / 172532, the entireties of which are hereby incorporated by reference. Using such techniques, also referred to herein as the Kiestra system, laboratory personnel are no longer required to read plates by direct visual inspection. Transitioning laboratory workflow and decision making to analyze digital images of culture plates can also improve efficiency.

[0004] Despite the significant developments with respect to imaging technology, there remains an attempt to extend such imaging technology to support automated workflow and / or automated diagnostic processes. In this regard, there is a need to develop techniques that can automate interpretation of culture plate images (e.g., growth identification, species identification, susceptibility testing, antibiotic susceptibility analysis, etc.) and determine subsequent steps to perform based on the automated interpretation. However, development of automated image processing logic (e.g., software) for diagnostic indication can be time consuming in the context of the diversity of specimen types and organism taxa. SUMMARY

[0005] Disclosed herein is a system for assessing whether biological specimens are present for pathogens, the identity of those pathogens, and other related analyses and assessments. Goals of the system are speed of analysis, accuracy of analysis, and process automation. Such systems generally obtain digital images of specimens disposed on nutrient media and incubated to discern evidence of microbial growth, which growth provides an indication of the presence of a pathogen in the biological sample. Such systems are described herein as Kiestra systems and include equipment to obtain one or more images of the incubated specimen, such as a camera and illumination system, a barcode reader for specimen containers (e.g., a petri dish containing inoculated spread media), etc.

[0006] Such systems communicate with and are controlled by one or more client-centric imaging applications (imaging-related Apps or Apps). Such imaging applications can be software that utilizes data derived from a series of historical specimen images to analyze new specimen images in order to automate the identification and / or diagnosis of disease states. The applications can provide clinical tools for rapid specimen characterization and result reporting. The applications link clinical specimens (from a clinical site) to non-patient-identifying facts about the specimen (e.g., non-patient-identifying specimen source information such as demographics), processing conditions (e.g., incubation time and temperature), processing materials or environments (e.g., nutrient media), and / or non-patient-identifying test results for the specimen (no pathogen growth, pathogen growth, or other positive identification of a pathogen, enumeration of identified pathogens). Such facts and conditions are collectively referred to herein as analysis information. Categorizing this analysis information in this way ensures that only the most relevant historical processing information is used to develop the applications. Thus, each developed application is created for a narrow purpose and used only when the specimen classifier corresponds to the application classifier. It is advantageous if the data is not linked to information that would identify the patient (i.e., patient de-identification). In one embodiment, the system automatically de-identifies specimen information and information responsive to certain conditions. De-identification includes providing metadata linked to non-patient-identifying categorical information (e.g., geographic region where the specimen was obtained, specimen type, etc.) that, in effect, allows the data to be used in systems and methods that cannot keep patient information confidential. The system can include applications that provide time series processing of images, classified / trained and tested diagnostic / evaluation algorithms, and / or expert systems. The applications can include a module, where the module can be understood as one or more processes or algorithms that use, for example, image metadata (metadata serves as the categories described herein) and rules for a particular purpose. In some cases, multiple modules can be implemented as one package. In some embodiments, data can be stored in a database that can be used to develop applications, train applications, certify applications, test applications, etc.

[0007] The system can provide image analysis processes that can incorporate best practices, address relevant specimen types, and automate selection.

[0008] To significantly shorten the time to market for valuable imaging applications, a multiple or modular system can be employed. Such a multiple system can include, for example, the following modules:

[0009] a) Define best practice solutions associated with specific media used to culture target microorganisms and taxons of target microorganisms to balance utility and development timelines. For example, combine test results and images associated with a specific media or taxon and use to develop an application to evaluate new specimens classified with the same media and taxon. The application will need enough information to develop a reliable application, but not so much information that development of the application is delayed.

[0010] b) Adjust the algorithm development pace (i.e., tempo or speed) to maximize reuse of existing algorithms.

[0011] c) Establish a clinical collaboration website that continually generates images and associated metadata from a variety of specimen types, organism taxons, and best practice media. That is, obtain specimen information to build a database of information that can be used to train or further train developed applications. This database of fully classified historical specimen information is referred to herein as a data lake.

[0012] d) Generate a database of images with defined criteria that can be applied to algorithm training and validation / clinical submission. The database can include information / data representative of clinical specimen images, and associated manual / standard analysis (quantification, identification, result interpretation) and classification (select patient demographics, imaging time and conditions, media type, etc.) for authenticity. The database infrastructure provides isolated data that can be accessed individually as appropriate for algorithm development, formal proof and validation (V&V), or clinical submission. This data generation allows prioritization and development of applications on-demand for specific specimen or media types.

[0013] e) Employ a bucket classification strategy (i.e., pure colonies, dominant colonies, or complex colonies) that reflects colony complexity for non-selective media; and for identified colonies and sister colonies, data such as by media type (e.g., chromogenic media (i.e., CHROMagar)) for defined taxons (or those defined taxons with specific attributes on specific media types (i.e., hemolysis on BAP)) or other information about conditions and reagents used to obtain images and test results.

[0014] f) Limit certain applications to certain imaging systems / devices (e.g., provide certain system capabilities when using a 25mp camera, but other imaging devices will provide other system capabilities). Likewise, limit applications to use in certain narrowly defined scenarios (e.g., sample type, media type, taxon type, camera type) and only use data to develop applications that correspond to the defined scenarios provides more useful and accurate applications for specimens evaluated by the applications.

[0015] The applications described herein can be further developed and refined as specimens, media, and organisms are validated and approved by regulatory agencies. For example, the quadrant quantification application can initially be implemented for throat swabs and wounds, and later extended to perianal or other specimen types as more specimens become available for processing. This ongoing training / development of the application provides a steady pace of new imaging applications with improved effectiveness. For example, as more images are evaluated, the application "learns" how to distinguish colonies from background in images of plated cultures. The processing of image information to distinguish pixels associated with colony images from pixels associated with background images is described in the Kiestra system previously mentioned herein.

[0016] A development system is also described herein that provides a high value software solution while shortening time to market and maximizing resource utilization. An important component of the system can involve initiating a collaborative program that involves selected clinical sites running an imaging system (e.g., the Kiestra system) to collect clinical specimen imaging information that is categorized by association with metadata for both development (algorithm training) and validation (submission preparation) purposes. This collection of data can provide for automated algorithm optimization, resulting in improved performance and reduced development time. Validation uses a separate set of images from the database. These data can also be reused when implementing additional features. However, the information is only used when the data categorization corresponds to the application categorization / goal. This approach creates extremely high efficiency, flexibility across the application (i.e., version control), and very valuable resources.

[0017] For non-selective media, a categorization bucket is created to characterize the complexity of the population to identify cultures as no growth, pure type, dominant type, or complex / mixed type, rather than attempting to identify all sister colonies of all categorical groups on all media types. Sister colony identification is extremely technically challenging and extremely time consuming, especially on non-CHROMagar media. Mixed cultures often require expert knowledge to interpret, and even at the level of automated image analysis, can require review to confirm and release. By using the pure, dominant, and complex / mixed categories, most specimens can be characterized and those with pure or dominant colony populations can be processed automatically with a higher degree of confidence. These colonies can then be selected by the application for automatic selection by the picking system. This approach requires a separate characterization and definition of the specifications for each specimen type (e.g., sputum, wound, swab, etc.) by specific pathogen / against specific pathogen to a large extent, a more general categorization strategy that efficiently utilizes the imaging algorithms and can characterize most specimen types.

[0018] An example imaging application is outlined below. Such an application can include

[0019] MRSA screening imaging application;

[0020] urine 2.0 imaging application;

[0021] rapid detection imaging application; and

[0022] quadrant quantification imaging application.

[0023] For example, in response to a Methicillin-resistant Staphylococcus Aureus (MRSA) screening analysis, the follow-up actions can include: (i.e., for a negative MRSA screening result): i) empirical treatment using a certain set of antibiotics that are not used in the case of drug-resistant Staphylococcus Aureus; ii) no isolation or special management of the patient; iii) performance of a certain next medical step, such as surgery; iv) ordering of certain additional diagnostic tests; v) selection of a certain set of antibiotics that can be effective to determine the guidance of antibiotic sensitivity tests. The application developed according to the methods described herein can guide the user in some or all of the above actions based on a historical analysis of previous specimens that share certain predetermined criteria with the specimen being evaluated by the application.

[0024] Actions from the rapid detection imaging application can include: i) transmitting to the treating physician a detection indicating a growth threshold level of infection much earlier than standard practice methods; ii) initiating a diagnosis that rapidly determines the identity of the growing pathogen (e.g., via MALDI-tof); iii) transmitting the pathogen identity to the physician substantially earlier than standard practice; iv) initiating a diagnosis to determine antimicrobial susceptibility and transmitting an antibiotic susceptibility curve substantially earlier than current practice. These goals are achieved by using previous images and image analysis results to inform the analysis of the current image. The previous images must be classified to develop an application that is experienced and reliable enough to be used to evaluate the current image. It should be noted that if the image analysis performed by the application results in a positive detection of microbial growth, in one embodiment the application can communicate with an analyzer that will receive the specimen under evaluation and identify one or more colonies on the specimen that are to be taken for further analysis. The application can instruct or control the downstream processing. Such downstream processing includes: i) preparing a suspension of one or more selected colonies in a predetermined buffer or solution; ii) adjusting the suspended cell suspension to a predetermined cell concentration; iii) spotting the cell suspension onto a substrate (e.g., MALDI plate); iv) covering the spots with one or more reagents (including, for example, MALDI matrix solution, extractant); v) dispensing one or more aliquots of the cell suspension into wells of an antimicrobial susceptibility plate to determine a susceptibility curve to a series of antibiotics at various concentrations; and vi) dispensing into a suspension for analysis by PCR, sequencing, or other molecular diagnostic tests. In alternative embodiments, the application instructs but does not control the subsequent sample processing / analysis.

[0025] To use the application developed according to the methods described herein for controlling processes and evaluating specimens, the application must be developed using relevant data and analysis. Methods and equipment that can be used to obtain the relevant data and analysis include collecting images and reference data from clinical sites that are used to construct data sets for future imaging application training and validation. See Figure 1 . This typically involves identifying, initiating, and managing resources for collaboration and data. Software tools collect data and metadata from clinical sites. Technical resources can be needed in the laboratory to generate classifications and assign classifications as metadata, for example from image processing, that are not typically available in a typical clinical examination.

[0026] Digital cameras, which are conventional cameras with good megapixel resolution, e.g., 5MP, 25MP, etc., can be implemented for early growth / no growth and presumptive identification (ID) due to adequate performance. For example, for colonies greater than or equal to 5mm in diameter detected by an imaging module or device that captures a digital image of the inoculum plate, this data can be combined with growth / no growth detection to indicate when a colony of sufficient size appears to be growing for further processing. The imaging device itself can be an application that performs analysis of the digital image, such as described in the Kiestra system mentioned elsewhere herein. From those systems and methods, colonies are distinguished from background and the image analyzed, the density of colonies on the inoculum plate is determined and communicated to an application that will determine further action in response to the image analysis. This information can then be used by the application to determine which colonies to automatically pick from the image (without the need for an operator to read the plate or identify colonies for picking). The application is developed and used for a specific plate environment. For example, pure plates (one colony type) and dominant plates (more than one colony type but one colony type dominant) can represent each colony type indicated by a purity plate module and associated rules that determine further review. Plates determined to be complex are handled by a different application (or an application that triggers different rules). A predetermined number of each colony type can be designated for automatic identification (ID) and antibiotic susceptibility testing (AST) review on an ID / AST module that can involve an automatic picking system / robot.

[0027] To reduce product development costs and shorten time to market, best practice support can be employed. Specific media (below) can be used to obtain optimal recovery and performance on a platform (e.g., BD Kiestra system).

[0028] The system can involve processing of streak plate reads performed in the laboratory. Streak plates are prepared as described elsewhere herein. Media type and sample classification group are examples of metadata that inform a specimen classifier linked to specimen information. The selected application can provide a suggestion for presumptive ID to pick only pure or dominant colony types on a plate. Specimens with multiple clinically significant isolates are rare and often complex. Specimens identified as complex can be sorted into a mixed category for manual review and action.

[0029] A set of algorithms can be designed to be more general purpose rather than specific to an application configured for use with a specific type of specimen or suspected target species contained in a specimen. These algorithmic tools are not described in detail herein but can be developed and used by one of skill in the art. The algorithms are validated for use in an application designed to evaluate and process a range of specimen types. When such tools are not limited to a specific specimen type, they can be used to review a large number of plates or as architectural building blocks for other applications.

[0030] The system generates 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 acquired and added to the database along with associated classifications. These tools will evaluate the panel and provide specific results independent of specimen type. These results, along with a compilation of specimen and patient demographics, will be processed by the application to provide an indication of the number of specimens. This simplifies implementation and leads to a faster time to market while providing a degree of flexibility for the user. The development of a more general toolbox of image analysis algorithms (along with classification metadata and rules, referred to as modules) that implement the application can be rapidly matured or optimized (e.g., for specific specimens) 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 identify the subject specimen. As part of the index strategy, artificial intelligence can be employed to "automatically" determine what image attributes and algorithms best provide the desired module capabilities. Individual modules may be sufficiently valuable to launch as applications, or multiple modules can be packaged into an application. In some versions, a set of modules can be distinguished for subsequent specimen analysis based on the specimen information obtained and its classification: (a) Growth / No Growth; (b) Semi-Quantitative; (c) Presumptive ID; (d) Pure, Dominant, Complex; (e) Antibiotic Susceptibility (Kirby-Bauer Test); (f) Sister Colony Locator.

[0031] Sample applications (Apps) may include Key ID, Rapid Detection, CHROMagar ID 2.0, Quadrant Quantification, and Hypothesis ID. These can be summarized as follows:

[0032] Key ID is a tool to identify a specific species on a specific plate where any number of colonies are present. Any number of colonies specific to the Key ID application will be assigned as pure or mixed cultures. Examples are:

[0033] 1. MRSA screening on MRSA II;

[0034] 2. Group A beta Streptococcus on blood agar plates (BAP); and

[0035] 3. Streptococcus pneumoniae on BAP.

[0036] The rapid detection application provides rapid detection processing. Growth at any read point can be used as a flag for growth detected as early as possible on any plate. The user selects the read point that will serve as the flag, recognizing the tradeoff between an earlier read point that can be less reliable but delivers results more quickly and a later read point that can be more reliable but takes more time to detect. The action taken will depend on the selected read point and the rules invoked. Early detection can occur as early as 4 hours, but incubation periods of 6 hours, 8 hours, or longer are expected. One skilled in the art is readily able to determine the incubation time for a particular specimen. The systems and methods described herein are not limited to any particular incubation time. The rapid detection application facilitates rapid positives for critical specimens.

[0037] For example, growth is detected on a BAP plate at 8 hours. A rule will yield these results and if the specimen type is cerebrospinal fluid (CSF), the laboratory will immediately receive an alert if the application determines that the specimen has grown on the plate because CSF is normally sterile. This determination requires the application to issue some kind of warning in response to determining that the CSF specimen is positive for a pathogen. For other specimen types, i.e., sputum, early detection has little value because almost all cultures have normal flora. In these cases, no rules will be written for such classified specimens and the application will not take any action based on rapid detection. Thus, the type of plate and the type of detection can trigger different automated processes.

[0038] Another implementation is a CHROMagar ID application for quantifying urine or other specimen types. This application obtains classified image information that will enable the application to assume identification of pure or dominant colony types in CHROMagar orientation. Mixed plates will also be identified but the application can trigger different rules in response to determination of complex plates. Using a rules engine, the processed images of the specimen can be classified into several categories. Certain categories permit automatic reporting of results. Those categories are described in detail elsewhere herein.

[0039] The quadrant quantification and assumed ID application is a tool that will evaluate all positives to classify each positive into one of the following: (1) no growth, (2) pure, (3) dominant, or (4) complex and evaluate the total number on the plate. Pure or dominant isolates will be assumed to be identified and colonies are identified for picking. In this case, all cultures will be analyzed and sorted into several categories, some of which are automatically reported and others are sent for review. Plates with a large number of pathogens can be sent to other systems (e.g., picking or testing) for picking without customer / clinician intervention. Figure 3 The table in Table 1 is a list of target organisms assumed to be identified on the mentioned plates. Figure 4 The table in Table 2 is a list of plates that are typically inoculated for a particular specimen (best practices).

[0040] The environment (e.g., Figure 1 ) can provide a software development workflow. This development attempts to focus on completing workflows and implementing organism group detection in Figure 4 on target media in Figure 3 The flowchart in FIG. 2 outlines a general strategy for algorithm and software development of one or more applications that support the analytical workflow of digital image evaluation. All images of inoculated and incubated plates are evaluated by the modules outlined in FIG. 2 and described below. Each module evaluates a specific outcome or discrete group of outcomes and is largely independent of specimen type. However, metadata associated with the specimen (which is used to classify the specimen) can also guide further processing (e.g., incubation instructions, imaging instructions, etc.). Such metadata can be read from a barcode on the specimen container (i.e., plate or petri dish). As results become available, the results are evaluated by an expert system and appropriate actions are suggested or taken. For example, when an application indicates that a specimen should be evaluated by AST, the expert system will provide guidance rules for the AST panel. The guidance typically takes into account regulatory or guidance positions (i.e., as established by the FDA or CLSI) as well as limitations on the validated capabilities of the AST testing platform (i.e., limitations). The expert system has a set of base rules. The application itself can also invoke rules based on information about the subject specimen's subject image that the application knows about, even if the application itself is not an expert system. These rules can be edited or additional rules can be developed by the user specific to their institution.

[0041] Once the plates are evaluated at the specimen level and the results are combined together, another set of rules will drive automatic reporting, user, laboratory information system (LIS) or laboratory information management system (LIMS) alerts, and sending of large numbers of isolates to worklists or pick systems (e.g., for ID / AST testing).

[0042] By targeting organisms on media independent of specimen type, the entire system can be developed more efficiently and quickly. For example, E. coli can be treated as the same genus and species regardless of the specimen source of the E. coli. Regulatory approval of some applications by specimen type can be difficult because some specimens (i.e., CSF and other sterile sites) have very low positive rates, so there is not enough historical sample processing / image data to develop clinically reliable applications. However, over time and with benefit of images obtained from a variety of sources and test results from those images (i.e., clinical trials, regulatory submissions), applications can even be developed for rare specimens.

[0043] One embodiment is a method for processing a biological sample comprising the steps of: obtaining a biological sample; combining the biological sample with a nutrient medium; incubating the biological sample; obtaining a digital image of the incubated biological sample; classifying the digital image according to an analysis criterion selected from the group comprising specimen source information, clinical sample guidelines, processing materials, and processing conditions; obtaining data from historical digital images of incubated biological samples from the nutrient medium that share at least one of the analysis criteria assigned to the digital image; and outputting instructions to a user for further processing of the biological sample using the historical digital image data.

[0044] The specimen source information comprises geographic information about the source of the biological sample and the type of biological sample. The processing materials comprise the type of nutrient medium. The historical digital image data is classified by at least one of the specimen type, organism taxon, or medium type. The method further comprises analyzing the digital image data and determining from the analyzed data whether the digital image reflects microbial growth. In response to determining that the digital image does not exhibit microbial growth, the method outputs an indication of no microbial growth. In response to determining an indication of microbial growth, the method performs the steps of determining whether the specimen is a sterile specimen and identifying one or more coordinates of the microbial colony in the image that is the basis for the indication of microbial growth. In response to determining that the specimen is sterile, the method comprises 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 include an identification (ID) test, an antibiotic susceptibility test, or both. The coordinates of the targets in the image that are classified as colonies are transmitted to a module that relays those coordinates to a picking device that will pick a colony from the biological sample. The module can be an application or combination of applications as described herein. The biological sample is transferred to the picking device and a colony is picked from the biological sample, where the transfer and picking steps are controlled by the module or the module has issued instructions to perform such steps. In response to determining that the specimen is not sterile, the historical image data is compared to the image data to identify a particular predetermined species of microbe in the digital image of the incubated biological sample. The comparison step is performed by the module, and if the module determines that the particular predetermined species of microbe is present in the image data, the module reports the identification of the particular predetermined species. When this determination is made, the module flags the specimen for further review.

[0045] The module compares historical image data to the digital image of the incubated biological sample and determines the amount of microbial growth, wherein the comparing and determining steps are performed in the module in communication with the imaging device that obtained the digital image of the incubated biological sample. According to the method, the module or application determines whether the microbial growth is a pure colony, a dominant colony, or a complex colony by transmitting the digital image of the incubated biological sample to a module that determines the level of growth as a vector of three probabilities. In response to determining that the colony is a pure colony, 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 transmits information to a user of the system. The module that has received the colony coordinates from the imaging device transmits the coordinates to a picking device that will pick the colony from the biological sample. The method can also include transferring the biological sample to the picking device and picking the colony from the biological sample, wherein the transferring and picking steps are controlled by 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 a comparison of an image of the colony provided to the module to historical image data accessed by the module. In response to the determination, the module performs another step of reporting the presumptive ID to a user.

[0046] If the module determines that the growth does not exceed the predetermined threshold, the module performs another step of reporting to the user that the complex sample does not meet or exceed the positive growth threshold. In response to determining that the colony is a dominant colony, the module provides a presumptive identification of the colony based on a comparison of an image of the colony provided to the module to historical image data accessed by the module. The module performs another step of reporting the presumptive ID to a user.

[0047] If the module determines that the growth exceeds the predetermined threshold growth, the module identifies the sample as a high value positive. The method further includes alerting a user of the high value positive sample. The method can also include identifying coordinates of the high value positive sample. The method can also include transmitting the coordinates of the colony to a module that transmits the coordinates to a picking device that will pick the colony from the biological sample. The method can also include a module that controls or issues instructions to transfer the biological sample to the picking device and pick the colony from the biological sample, wherein the transferring and picking steps are controlled by the module. The method can also include alerting a user that further review is needed. If the module determines that the growth does not exceed the predetermined threshold, the module performs another step of reporting the presumptive ID to a user. In response to determining that the colony is a complex colony, the method further includes reporting by the module that the plate is complex. The method can also include determining by the module whether 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 needed. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A block diagram of an example system environment to illustrate development of a biological image processing application.

[0049] FIG. 2 is an example flowchart of an implementation of an example application and its processes in a development environment, according to an aspect of the disclosure, where FIG. 2 includes Figure 2A and Figure 2B .

[0050] Figure 3 A table to illustrate examples of target organisms and the assumed identification and associated media.

[0051] Figure 4 A table to illustrate example target media and associated specimen types.

[0052] Figure 5 A schematic diagram of processes that can be implemented in some versions of the present technology, for example, in a processing system 101 in Figure 1 , where the processing system has access to data and / or algorithms of a data lake as discussed herein to develop an application with a training process, a validation process, and / or a clinical submission process. These processes in turn can employ information (e.g., updated data and algorithms) that is also derived from the data lake. DETAILED DESCRIPTION

[0053] The present disclosure provides apparatuses and methods for developing imaging applications for identifying and analyzing biological specimens, such as environments of microbial growth. Many of the methods described herein can be fully or partially automated, such as integrated as part of a fully or partially automated laboratory workflow.

[0054] This document provides a description of the design and implementation of a system that will speed up the delivery of automated imaging capabilities to biological imaging systems such as the BD Kiestra TM System. The imaging capabilities of such systems are enabled by a series of hardware, software, analytical algorithms, and clinical rules. An example of one such commercialized system includes one or more digital cameras (e.g., 4MP) with multiple illumination configurations that generate optimized and standardized images based on a suitable platform. The system described herein is capable of being implemented in other optical systems for imaging microbial specimens. There are many such commercially available systems that are not described in detail herein. One example can be the BD Kiestra TM ReadA compact smart incubation and imaging system. Kiestra TMThe ReadA Compact System is an automated incubator with integrated camera and plate transport system that enables automated imaging of plates. The ReadA Compact System is commercially available. The ReadA Compact System also has integrated plate import and plate export devices that couple the incubator to other instruments for manipulation. Thus, in some embodiments, in response to analysis of digital images by one or more applications, the applications can issue instructions and control incubation of relevant specimens in evaluation by the applications. Other example systems include those described in PCT Publication No. WO 2015 / 114121 and U.S. Patent Publication 2015 / 0299639, the entireties of which are incorporated by reference herein. Such optical imaging platforms are well known to those of skill in the art and are not described in detail herein.

[0055] A suite of applications for the system can provide, for example, in the ReadA Compact System, analysis of images generated from most specimens at various predetermined times. The system can enable downstream operations of these plates, including automated release of no growth and / or negative plates and automated characterization of colonies for definitive ID and AST analysis.

[0056] At a high level, imaging analysis tools (imaging applications) can be deployed to enable a range of different results and / or actionable clinical outcomes to be provided to the laboratory. These applications utilize one or more image analysis algorithms (modules), as well as a set of rules providing information on how to apply the modules to particular media types and / or particular specimens. Certain applications can also have associated expert systems that overlay with more basic application determinations regarding notification of recommendations for action and interpretation of results, often another set of rules for regulatory / clinical guidance.

[0057] Development of imaging applications can utilize an iterative approach. A data lake is developed using specimen information acquisition or clinical specimen images or both that are processed as routine practice in a clinical laboratory. Algorithms are developed to model conclusions / instructions / output such as illustrated in FIG. 2 that can be extracted from the facts and categorical information associated with the specimen images evaluated by the application. Validation and verification (V&V) of the application is performed by using a range of predefined images from the data lake that are designated for V&V analysis along with certain metadata requirements. The application is then subjected to a clinical study (which includes processing specimens to obtain images of those samples and evaluating the images of the specimens using the application) to prove the application and train the application. The clinical study does not require that a particular clinical site be used for the clinical study.

[0058] However, the biological image processing application development system 100, for example Figure 1The system illustrated in the middle can be implemented to take advantage of an exponential strategy with several components. The system 100 includes one or more of the following:

[0059] a) A set of tool box applications, e.g., for a processing system 101 with one or more processors, with general algorithms and modules that implement the applications and can be quickly matured or optimized (e.g., for a type of specimen) by applying artificial intelligence algorithms and other deep learning algorithms such as neural networks, artificial neural networks, etc. Artificial intelligence can be implemented to "automatically" determine what image attributes and use what algorithms to best provide the desired module capabilities.

[0060] b) A developed database 102, named herein as a data lake, is critical to the development and deployment of the applications described herein, including images of clinical specimens and related manual / standard analysis of "truth" (i.e., facts about the image such as colony quantification, ID, result interpretation, etc.) and image conditions and (in some cases) patient demographics (appropriately de-identified). The images carry classification information in the form of metadata so that only relevant image data is used to develop a particular application. The data lake can be populated by, for example, a clinical laboratory in collaboration with the imaging system 104, which can optionally supply data to the database via a network.

[0061] The storage location of the data lake depends largely on design choices. The data lake can be stored locally or in the cloud. Data in the data lake can be partitioned. For example, data in the data lake can be segmented depending on how the data is accessed and / or used. In one implementation, one segment of data in the data lake can be used for algorithm training, another segment can be used for data proofing or validation, and yet another segment can be used for clinical submission.

[0062] The system's database can store and provide classification data for these clinical images and related data, which can be individually extracted on demand for algorithm development, formal proof and validation, or clinical submission as appropriate. The application can be built with a certain level of global standardization including lab protocols, media types, imaging times, quantification scores, etc. Resources can be supplemented to generate images and reference data (specimens and / or spiked / artificial specimens) for samples / organisms / conditions that are rarely encountered in clinical settings. This proactive strategy for data generation allows for almost on-demand prioritization and development of specific applications for specific specimen or media types. The system architecture permits integration of new modules into existing lab site software to ease release of new imaging diagnostic functionality (applications / modules / packages). This can be achieved by standardizing the interface between new applications / modules / packages and existing / previous imaging software systems. In this regard, the applications / modules / packages can be add-ons to system software such as system software for automated imaging systems (e.g., in any one or more of a control board / specimen transporter, incubator, camera, picking machine, and / or related robotic technology for moving such specimens / plates within processing systems for such automated lab units / equipment, etc.). Known pathogens can be spiked in certain specimens that are negative for pathogen determination, and subsequent images of the incubated, spiked specimens are characterized as described herein. The images of the spiked specimens can then be used as a training set for applications that can be used to evaluate and handle less frequently occurring pathogens.

[0063] This approach can provide the most flexibility in terms of prioritization and pace of application development. It will also make it possible to provide early application performance metrics, helping to better understand the added value of the application, the synergy at the solution level, and enabling earlier recognition. The data lake can be generated, for example, using one or more hospital systems that comply with the list of company clinical development (CCD) guidelines (i.e., technical and medical information). The implementation provides the ability to store and classify data, query and audit databases; define procedures and processes for lab protocols; training, monitoring, compliance, quality metrics, etc. typical for clinical trials - and can be done as part of the data lake development, not at the end of the typical product development process. The approach can implement dedicated clinical lab resources to determine certain plate results outside of standard protocols, and link images and analysis results to the data lake. Certain specimens, plate types, or image acquisition time points can need to be run specifically, developing the data lake with special processes that can be unrelated to regular / typical lab practices. Thus, in some versions, the data lake database can include images of clinical specimens along with related manual / standard analysis for authenticity (e.g., quantification, ID, result interpretation) and metadata associated with image classification (e.g., select patient demographics, imaging time and conditions, media types, etc.).

[0064] The establishment of both the algorithm and the data lake can include a level of standardization. This will also define the validation purposes of any given application, as well as the analysis / instructions / output that can ultimately be obtained. Given the diversity of media types used by global laboratories, media suppliers, and incubation times, a "best practices" approach can be used to initiate this work. Additional conditions can be added in the future by storing appropriate data in the data lake. An example of a media x specimen matrix is outlined in Table Figure 3 Figure 3 The table in Table 1 includes 12 media types. It should be noted that blood agar, trypticase soy broth (TSA), and Columbia are grouped together as one media type. XLD is xylose lysine deoxycholate agar, SS agar is Salmonella, Shigella agar, CNA is Columbia Nalidixic Acid agar (CNA), and CLED is cystine lactose electrolyte deficient agar. The skilled artisan is well aware of the media listed and the microorganisms known to be identifiable on the listed media. Thus, standardization can include laboratory protocols, media types, time of imaging, streaking pattern, quantification scoring, etc., which are a classification of data that is applied to database development and image analysis with reference to the database / data lake / historical image information. This focuses on validation work and minimizes development time and allows for inter-lab metrics, data sharing, etc. The use of the data lake as training data, validation data, clinical submission data, etc. for developing applications is illustrated in Figure 5

[0065] To ensure database accuracy, data population into the database can involve independent human image analysis of images performed by several individuals, or plate analysis performed manually by human technicians. The human readings can be compared to clinical laboratory reports. In some cases, another image review can be involved for discrepant readings. Database input can involve de-identifying patient information from image related data. The review of data input images can involve human scoring of plates for pure growth, predominant growth, complex growth, and no growth; quadrant quantification.

[0066] Example software modules for the toolbox

[0067] In a typical example, the data lake can contain media plate images associated with laboratory determined quantification (e.g., no growth, +, ++, +++). It can also contain identification (ID) of organisms that are significant to the species or type of specimen (e.g., can be apparent to a trained clinical microbiologist). The data lake can also contain image based metadata, and in some cases patient demographic information. Organisms that are routinely not identified as pathogens (i.e., normal flora) can also be requested to facilitate algorithm development. Once populated, a portion of the images will be leveraged to train and test appropriate algorithms for application development. ​​

[0068] At the high level, imaging analysis tools can be deployed to implement a range of different results and / or actionable clinical outcomes that provide significant value to the laboratory. These modules (Mods) include one or more image analysis algorithms, and a set of rules that provide information on how to apply the module to a particular media type and a particular specimen. Some modules, such as screening for MRSA, can be implemented as an application. Other modules can more often be packaged together with other modules to provide synergistic capabilities (e.g., quadrant quantification and purity are often packaged as an application). Some applications can also have an associated expert system that overlaps with another set of rules that inform recommendations for action and interpretation of results (e.g., the KB zones described herein). Based on technical and clinical considerations, one or more applications can also be bound as a component of a launch pack depending on different clinical laboratory needs. It is also expected that some applications will have versions (e.g., FDA approved UCA V1.8 will become UCA 2.0).

[0069] Some examples of algorithms and modules (collections of synergistic algorithms) are generic (often across specimens, pathogens, media work) and are outlined below and can be considered in relation to the process illustrated in FIG. 2. As mentioned previously, the categorization of the functionality of various applications facilitates the development of detection applications for various species and media.

[0070] 1. Growth Application / Module 1010A

[0071] The growth application 1010 (see FIG. 2) can be directed to answering the simple question: at this particular incubation time, can anything be detected for growth on this plate? The answer to this question will be a growth probability that ranges between 0 and 1. Growth can be a module targeted at any media, regardless of dispense volume or streak pattern. Growth can be detected as early as possible from the preset imaging point. Rules specify whether to issue a warning based on specimen type and / or media. In some versions, Gram stain results can also be integrated with the application / module where appropriate. In some cases, this can implement an early detection or early growth application / module. Growth can be detected as early as possible (e.g., 4-14 hour or longer detection window) from the preset imaging point. Rules specify whether to issue a warning based on specimen type and / or media. In some versions, Gram stain results can also be integrated here where appropriate.

[0072] 2. Key ID Application / Module 1020

[0073] The key ID module 1020 (see FIG2 ) can be designed to identify species that are potentially growing on a given culture medium. For each requested key ID organism, the module can provide a list of colony locations for each key ID sorted by descending probability (with probability). These colonies can then be picked manually or by an automated picking system.

[0074] In some versions, the system may include one or more screening and key pathogen modules generated based on pathogens. These modules may provide detection for specific pathogens, groups of pathogens, or pathogens with specific properties. Screening applications may be implemented to enable identification of specific pathogens on CHROMagar and may be used for patient management as well as pathogen characterization, such as MRSA, ESBL, CPE, VRE, etc. CHROMagar may enable a range of pathogens to be presumptively identified, i.e., CHROMagar directed for both Gram-negative (GN) and Gram-positive (GP) bacteria. Pathogen-specific media may be used for specimens - for example SS media for Salmonella, Shigella in stool. Certain organisms may be presumptively identified or marked on more general media based on, for example, hemolytic properties on blood agar or unique morphological properties on specific media. Possible collections of modules with pathogen x media capabilities are outlined in Figure 4 in the table.

[0075] 3. Quadrant Quantification Application / Module 1030

[0076] Based on a stroke pattern, e.g. BD Kiestra TM InoqulA TM Quadrant quantification module 1030 (see FIG2 ) provides a growth level as slight, moderate, or vigorous if growth is detected. BD Kiestra TM InoqulA TM The processing of both liquid and non-liquid bacteriological specimens is automated, thereby helping to simplify the workflow, realize standardization process and ensure constant and high-quality streaking for inoculating solid growth medium. Growth level will be returned as a vector of three probabilities (slight, moderate or vigorous), ranging from [0,1], summing to 1. For example, the module can evaluate all plates to assess whether there is no growth or there are different growth amounts (e.g., +, ++, +++) and whether any growth is pure growth, dominant growth or complex growth. No growth determination can optionally result in automatic release or batch release. In some cases, growth quantification (e.g., +, ++, +++) can be determined by three or more bacterium colonies in any particular quadrant.

[0077] With respect to growth type, images / plates can be characterized as pure, dominant, and complex. This can be based on a minimum number of isolated colonies per type. For example, dominant growth can be greater than (or equal to) two colony types, with one type being greater than (e.g., 10 times) one or more other types. Complex growth can be greater than two colony types, and there is no dominant isolate or if the isolate cannot be identified in the assumed ID table, for example Figure 3 Examples of complex plates can be automatically flagged / passed for manual interpretation. Pure and dominant plate types can be automatically passed for further automated processing, for example, assigning a representative of each colony type, with rules to push further examination (e.g., pick, ID, and / or AST).

[0078] For example, a colony forming unit (CFU) / mL quantification application / module 1040 (see FIG. 2) can be implemented. Based on InoqulA TM Streak pattern #4 (single plate) or #6 (double plate), this module can provide a growth level of <1, 1-9, 10-99, 100-999, >1000 CFU / medium on a plate. The growth level can be returned as a vector of 5 probabilities ranging from [0,1] and summing to 1. To obtain equivalent CFU / mL bucket units, the dispense volume can need to be taken into account.

[0079] In general, in some versions, the quantification module can determine whether growth is due to a single growing organism, a dominant organism, or a mixture of organisms. A pure organism can be considered to be an organism that accounts for >99% of the observable / imageable growth. A dominant organism can be an organism that accounts for (90%, 99%) of the observable / imageable growth. The purity level can be returned as a vector of probabilities (e.g., 3 probabilities, such as pure, dominant, complex) ranging from [0,1] and summing to 1. In the case of pure or dominant growth, at most five colony locations of the primary organism will be given in decreasing probability.

[0080] Examples of responses to the determined quantification are as follows:

[0081] 1) An image of a specimen plated on a medium determines that there is a mixed flora with greater than a predetermined threshold (100,000 CFU / mL). The application's response to this determination is to flag the plate as a complex plate because the plate has a mixed flora greater than the threshold amount and suggests manual review of the plate. This determination is not made in the case of a medium or taxon, so this is an application with broad applicability and is not limited to a particular medium or taxon deployed on the plate.

[0082] 2) The image is evaluated and determined to be no growth within 24 hours. If the specimen is classified as a critical specimen, the application issues a preliminary report of no growth and recommends or controls re-incubation of the plate for 24 hours. If the subsequent image detects no growth within 48 hours, the application sends a final report of no growth after 48 hours to the user. The application recommends or controls discarding the plate.

[0083] 3) The image of the specimen on the streaked media determines pure growth with greater than a predetermined threshold (100,000 CFU / mL) and colony size of more than 0.5 mm. The response of the application is to issue an order or control the picking of the colony for ID and AST testing. The application will flag the specimen for review by a technician and send a report of greater than 100,000 CFU / mL to the physician associated with the specimen.

[0084] 4) The image of the specimen is determined by the application to reveal the presence of MRSA. The application sends a report of detection of MRSA and adds the specimen to a positive MRSA work list. If the application determines that the size of the MRSA colony exceeds a threshold (e.g., greater than 0.5 mm), the application will issue an order or control to send the sample for ID and AST testing. As mentioned elsewhere herein, ID and AST have their own specimen review and evaluation. As such, the ID and AST system and equipment are typically downstream from the incubation / imaging equipment (e.g., Kiestra ReadA Compact system). TM

[0085] 5) The image of a specimen classified as ESBL reveals no growth. The application will issue a final report of no detection of ESBL isolate and will issue an order to discard or control discarding of the plate.

[0086] 6) The image of a specimen classified as sputum is identified as having a mixed flora that exceeds a threshold amount (thus a complex plate). When the application determines that the plate is complex, the application issues an order for a technician to review the plate. It should be noted that different thresholds of mixed flora that trigger the requirement for manual review can be used depending on the specimen classification.

[0087] 7) The image of a specimen classified as more critical on blood agar reveals growth. In this case, the critical specimen application will trigger and send a warning to the physician associated with the specimen and cause the specimen to be added to a critical sample work list. If the application determines that a colony greater than a threshold size (i.e., greater than 0.5 mm) and the specimen is classified as being placed on MacConkey agar, the application causes the specimen to be sent for automatic picking for MALDI and Gram-negative AST. The application will also send a report indicating that a Gram-negative specimen has been isolated. ​

[0088] 8) The image of the specimen reveals a number of colonies greater than 100,000 CFU / mL and the image is classified as pure and with colony size greater than a predetermined threshold (e.g., greater than 0.5 mm). In response, the application causes the specimen to be automatically selected for AST, causes the specimen to be added to a positive review list by a technician, and causes a report to be sent to the physician associated with the specimen indicating that more than 100,000 CFU / mL of colonies were detected from the sample. In addition, if the AST results reveal that the selected colonies are resistant to carbapenems, the application causes a molecular confirmatory test to be performed.

[0089] 9) The image of the specimen is determined to have profuse growth. In response, the application causes the growth to be automatically selected and a suspension prepared for MALDI to be performed on the selected sample. If the MALDI identifies the colonies as E. coli, the application causes the sample to be further evaluated for Gram-negative AST (using the MALDI suspension or a new selected colony).

[0090] In some versions, the growth detection can be implemented as two modules, where one module evaluates the plate to determine growth / no growth and the other module evaluates the number of growth (+,++,+++) in 3 or more colonies in any particular quadrant. For example, a first module evaluates the image of a critical, typically sterile specimen. If the evaluation reveals growth at a predetermined time point and determines that the colony size is greater than a predetermined threshold size, the application identifies the coordinates of a representative colony and issues instructions to the imaging device (e.g., ReadA) to move the plate to a device that will automatically pick the identified colony. The picked colony is resuspended in solution to a predetermined density for further testing (e.g., PCR, sequencing) in a molecular diagnostic device or test.

[0091] 4. Hypothesis ID Application / Module 1050

[0092] In the case of pure or profuse growth, the Hypothesis ID module 1050 (see FIG. 2) can identify the primary organism(s) potentially growing on a given medium using a set of recognition algorithms based on training using the data lake. This module can provide / output the name of the highest ranked (e.g., probability) organism(s) (or group of organisms) and up to five colony locations ranked from highest to lowest probability for the identification. These algorithms enable the identification of any number of colonies and specific species on specific medium types that are considered to be of clinical significance.

[0093] For example, medium and colony identification can be Figure 3those indicated in the table above. Rules for manipulation checks on these colonies can be included in the module. As examples of specific ID applications, the Urine Culture Application (UCA) and the Chrom ID application can provide presumptive ID on CHROMagar from urine for those organisms required by the media. Rules will be used for automatic reporting / automatic release (or batch release) and downstream checks (e.g., automatic picking, testing, etc.). In some cases, rules can determine high value positives to flag the plate for fast review and guidance for further processing by worklist or automated picking, etc.

[0094] 4.1 Purity Panel Application / Module

[0095] In some versions, the system can implement a purity plate module. Pure, dominant, and complex plates can each require a minimum number of isolated colonies. Dominant growth typically has more than two colony types, with one colony type more than ten times greater than the other. Complex types typically have more than two colony types without dominant isolates or isolates that are not identifiable (presumptive ID table below). Complex plates are typically interpreted manually. Thus, the module can classify a plate image as pure, dominant (may be slightly mixed), or complex from the plate image.

[0096] 4.2 Auto Select ID / AST Module

[0097] Pure and dominant plates can have a representation of each colony type indicated by the purity plate module and associated rules that drive further checks, as set forth in the examples above. A predetermined number of each colony type can be specified for automatic identification (ID) and AST checks on the ID / AST module that can involve an automated picking system / robot.

[0098] 5. Kirby-Bauer (KB) zone diameter measurement application module

[0099] Some embodiments can utilize an optional measurement application. This application can leverage existing imaging capabilities and AST expert systems to provide zone measurements. Optionally, these measurements can be correlated with the expert system to provide interpretation. One version of this application also has the opportunity to make early zone measurements on specific drug / organism combinations and on each zone on media plated directly from positive blood cultures. Such algorithms can be based on metadata and / or images from the data lake.

[0100] In some versions, this application will leverage existing imaging capabilities and AST expert systems to provide zone measurements and Abx disk identification. Optionally, these measurements can be correlated with expert systems to provide guidance on antibiotic susceptibility curves for pathogens isolated from patients and guidance on treatment / response. In some versions, embodiments of this application can provide early zone measurements for specific drug / organism combinations and for each zone on media plated directly from positive blood cultures. Some applications can include expert systems (interpretation) and can be greatly aided by the application using a data lake that is curated, monitored, properly audited and the required metadata and images.

[0101] While the system 100 can include any one or more of the above imaging related modules / applications, in some versions, specific segmentation of the functionality of the modules can be implemented by the following discrete set of modules / applications: (a) quadrant quantification; (b) detection of no growth; (c) purity plate: # colony type (e.g., pure, dominant, complex); (d) screening (e.g., CHROMagar, i.e., MRSA); (e) key pathogens; (f) early growth detection; (g) automatic select ID / AST; and (h) zone measurement paper method.

[0102] Imaging applications and launch packs

[0103] The system 100 providing the combination of algorithms / modules / rules, data lake, and the ability to extract predetermined subsets of data can enable the on-demand ability to mature algorithms and develop applications quickly. As an example, an application can be developed based on specimen type and can be implemented with a series of applications. The strategy for implementing a series of applications is influenced by many factors: the support software launch cadence; the value of individual applications versus the value of the applications together; the availability of certain algorithms or specimen / plate / organism types in the data lake; etc.

[0104] Examples can be considered in relation to the following table:

[0105] Table 1: Exemplary modules and their functionality

[0106] Module Number Module Name Function 1 Urine Culture Quantification Quantification into 5 buckets; application of user threshold rules 2 Urine Culture Auto Negative Release Auto release of no significant growth panels (in Europe) 3 Urine Culture Quantification Dual Panel Quantification of each half into 5 buckets using rules 4 Urine Culture Early Growth Detection Earliest growth detected on panel provides notification to user 2.1 Urine Culture Batch Negative Release Batch release of no growth panels (in USA) 5 Urine Presumptive ID Directional CHROMagar based ID of required taxon

[0107] In this specimen-based example, a series of 5 different modules are involved with one specimen type. The application of validation against a specific specimen type is one way of packaging functionality, however, the index approach will also allow other options. For example, a surveillance application would allow for initiation by specific target organisms (MRSA, Streptococcus, Shigella); a quantification application could be packaged by media type (number of samples on blood agar, independent of specimen), etc. However, in this sense, the value of certain applications for certain specimens can be minimal (i.e. number of samples on non-selective media with sputum, given high normal flora levels). Note that applications can be limited by region with specific rules and specific functionality limited to the geographic region where the specimen in the evaluation was obtained. Table 1 identifies the functionality specific to the clinical requirements for the United States (US) and Europe (EU).

[0108] Thus, possible applications can be separated into two high-level buckets. The first bucket set can be considered screening and critical identification applications. Such applications are typically targeted at specific organisms on CHROMagar, or for high value pathogens on e.g. blood agar. Each of these applications is discrete and can be prioritized for development and initiation independently with minimal impact on other applications or specimen types. Similarly, Kirby-Bauer zone applications are typically independent of Next Gen Apps with associated algorithms that can be independently prioritized. In addition, as new CHROMagar (i.e. Vancomycin Resistant Enterococci (VRE)) becomes available, appropriate specimens can be used to populate the data lake and added to this list. If the target isolate is relatively rare, the data lake can be supplemented with artificial (spiked specimen) samples. Additional example screening and critical ID applications are illustrated in the table below.

[0109] Table 2: Application Classification, Construction, Functionality and Output

[0110]

[0111]

[0112]

[0113] From Table 2, it can be observed that applications can be very specific and the output can depend on the specimen classification (i.e. specimen type, US region or EU region, etc.). Table 2 also illustrates the type of data used to train the application at a high level.

[0114] The second bucket set of applications uses more generic algorithms and can be prioritized and grouped to launch by several criteria. An overview of the instances of these applications is provided in Table 3 below. Essentially, each cell in Table 3 represents an application. Cells that share the same number in the table below are suitable capabilities for specimen types that are reasonably packaged together in a common module sharing the cell number. With this model, there are 8 additional launch packs / modules.

[0115] Table 3

[0116]

[0117]

[0118] An example imaging module can be considered with reference to the following table:

[0119] Table 4: List of example modules

[0120]

[0121]

[0122] EUCAST is the European Committee on Antimicrobial Susceptibility Testing. In one example of a process integrated with one or more applications for specimen evaluation and process control, specimens are inoculated on spread media using BD Kiestra TM InoqulA is inoculated on spread media. The specimens are streaked on the media using a predetermined pattern of the part of the metadata tracked via the barcode. The streaked specimens are transported to BD Kiestra TM ReadA compact system where the specimens are incubated and imaged at a time determined by the application. The images taken at the specified time by ReadA are analyzed by the application to determine if the specimens on the plate are pure specimens. Further checks of the present specification are performed based on the determined results. The images are evaluated to identify the coordinates of the selected colonies. The application can transport those coordinates to the equipment (or technical specialist). The application can issue instructions to transport the specimen to the equipment that will pick the colonies. The application can further coordinate or control the picking of the colonies and transport the picked colonies to another platform that performs the ID of the pathogen. In one example, the ID is performed by MALDI. As described elsewhere herein, the samples are evaluated by MALDI by placing the picked samples in a suspension and inoculating a MALDI plate with the suspension. The application can also coordinate or control the transfer of the colony suspension to BD Kiestra TM InoqulA. Here, the suspension can be inoculated on another type of media (e.g., Mueller Hinton) using a “scatter pattern” and then moved to an AST testing equipment where predetermined antibiotic disks (e.g., BD BBLTM Sensi-Discs TM ) placed on the culture. The plate carrying the inoculated specimen and antibiotic discs is then transported to the ReadA compact system under the coordination and control of the application. ReadA acquires images and provides those images to the application, which transports the results from the application to the expert system for analysis of the resulting antibiotic disc zones and interpretation of the results. The expert system then transports the analysis results to the clinical laboratory staff.

[0123] While the application has been described herein with reference to particular embodiments thereof, it is to be understood that the application is not limited to the details described herein and numerous modifications and other arrangements can be devised by those skilled in the art which fall within the scope of the application as defined by the appended claims.

Claims

1. A system for microbial growth detection, colony counting, and / or identification, comprising: a database system comprising: (a) digital images of microbial specimens; (b) quantified values determined for the microbial specimens of the digital images; and (c) identifications of organisms determined to be significant to the microbial specimens of the digital images; one or more processor-readable media having processor control instructions defining a discrete set of application modules, the discrete set of application modules comprising: a growth detector configured to process a digital image of a growth medium from a set of imaging points in the digital image and generate a growth indicator comprising a probability value representing a probability of microbial growth occurring in the growth medium; a purity detector configured to generate a categorization of the digital image of a growth medium from at least one predetermined purity level, wherein the predetermined purity level comprises a discrete set of purity levels, the discrete set of purity levels comprising a pure level, a dominant level, and a mixed population level; a growth quantifier configured to process the digital image of the growth medium from the growth detector, the growth quantifier configured to generate a growth level quantification from the digital image, wherein the growth quantifier is configured to generate the growth level quantification as one or more of a slight growth probability, a moderate growth probability, and a robust growth probability; and a presumptive identifier configured to process the digital image from the growth quantifier, the presumptive identifier configured to generate a name indicator for a set of microbial specimens of the digital image based on training using digital images of the database system, wherein the presumptive identifier is configured to generate the name indicator by generating a probability for each of the name indicator.

2. The system of claim 1, wherein the growth level quantification comprises a set of probabilities ranging from 0 to 1, wherein the sum of the set of probabilities is 1.

3. The system of claim 1, wherein the presumptive identifier is configured to generate the name indicator as a list ranked by the generated probabilities of the name indicator, or configured to provide the list for each of a plurality of detected colony locations in the growth medium of the digital image.

4. The system of claim 1, wherein each probability in the discrete set of purity level probabilities ranges from 0 to 1, and wherein the sum of the probabilities of the discrete set of purity levels is equal to 1.

5. The system of any one of claims 3-4, wherein the purity detector generates at least one of a pure level characterization when a single organism causes detectable growth or a dominant level characterization when a single organism causes a predetermined percentage range of detectable growth.

6. The system of claim 5, wherein the predetermined percentage range is 90% to 99% of detected growth.

7. The system of any one of claims 5-6, wherein the purity detector generates a mixed flora level characterization when a single organism causes less than the predetermined percentage range of detectable growth.

8. The system of any one of claims 1-7, wherein the discrete set of application modules further comprises a key organism identifier configured to generate a probability indicating a likelihood that the digital image of a growth medium contains a colony of an input species request or a set of probabilities indicating a likelihood that the digital image of a growth medium contains a set of colonies of an input species request based on training using digital images of the database system.

9. The system of claim 8, wherein the key organism identifier is configured to access a set of rules trained using digital images of the database system, the set of rules configured for classifying the digital image of a growth medium inoculated with a specimen with respect to an input species request.

10. The system of claim 9, wherein the key organism identifier is configured to access a plurality of sets of rules trained using digital images of the database system, wherein each set of rules in the plurality of sets of rules is configured for classifying the digital image of a growth medium with respect to one species of the set of input species requests.

11. The system of any one of claims 1-10, wherein the discrete set of application modules further comprises a volume quantifier configured to generate a probability indicating a likelihood that the digital image of a growth medium contains a volume quantification of growth volume in a set of volume ranges, wherein the volume quantifier generates a probability value from 0 to 1 for each range in the set of volume ranges, wherein the sum of the probability values equals 1.

12. The system of any one of claims 1-11, wherein the database system is coupled to a network to receive data from one or more clinical laboratories containing imaging systems used to generate digital images of microbial specimens on growth media, wherein the database system further comprises de-identified patient demographic data.

13. A method for processing a biological specimen, comprising: obtaining a biological specimen; combining the biological specimen with a nutrient medium on a plate; incubating the biological specimen; obtaining a digital image of the incubated biological specimen; classifying the digital image according to analysis criteria selected from the group consisting of specimen source information, clinical specimen guidelines, processing materials, and processing conditions, the specimen source information including geographic information about a source of the biological specimen and a type of the biological specimen, the processing materials including a type of nutrient medium; obtaining data from historical digital images of incubated biological specimens from nutrient media that share at least one of the analysis criteria assigned to the digital image, wherein the historical digital image data is classified by at least one of specimen type, organism taxon, or medium type; and outputting instructions to a user for further processing of the biological specimen using the historical digital image data.

14. The method of claim 13, further comprising analyzing the digital image data and determining from the analyzed data whether the digital image reflects microbial growth, and outputting an indication of no microbial growth if the digital image does not exhibit microbial growth.

15. The method of claim 14, wherein in response to determining an indication of microbial growth, determining whether the biological sample is a sterile specimen and identifying one or more coordinates of a microbial colony in the digital image that is the basis for the indication of microbial growth, wherein in response to determining that the biological sample is a sterile specimen, indicating that the specimen is a high value positive and sending instructions to further process the specimen.

16. The method of claim 15, wherein in response to determining that the specimen is not sterile, comparing the historical digital image data to the image data to identify a particular predetermined species of microbe in the digital image of the incubated biological sample.

17. The method of claim 15, further comprising comparing the historical digital image data to the digital image of the incubated biological sample and determining an amount of microbial growth, wherein the comparing and determining steps are performed in a module in communication with an imaging device that obtained the digital image of the incubated biological sample.

18. The method of claim 15, further comprising determining whether the microbial growth is a pure colony, a dominant colony, or a complex colony by transmitting the digital image of the incubated biological sample to a module that determines a growth level as a vector of three probabilities.

19. The method of claim 18, wherein in response to determining that the colony is a pure colony, the method further comprises: reporting by a quadrant quantification module that the plate is pure; and determining by the quadrant quantification module whether the growth exceeds a predetermined threshold growth.

20. The method of claim 19, wherein if the quadrant quantification module determines that the growth exceeds the predetermined threshold growth, the method further comprises: identifying that the biological sample is a high value positive; alerting a user of the high value positive; identifying the coordinates of the high value positive; transmitting the coordinates of the colony to a picking device that will pick the colony from the biological sample; and transferring the biological sample to the picking device and picking the colony from the biological sample.

21. The method of claim 19, wherein if the quadrant quantification module determines that the growth does not exceed the predetermined threshold, a comparison of an image of the colony to the historical digital image data provides a presumptive identification of the colony, and the presumptive identification is reported to a user.

22. The method of claim 18, wherein in response to determining that the colony is a dominant colony, a comparison of an image of the colony to the historical digital image data provides a presumptive identification of the colony, and the presumptive identification is reported to a user.

23. The method of claim 22, the method further comprising: determining by a quadrant quantification module whether the growth exceeds a predetermined threshold growth, and if the quadrant quantification module determines that the growth exceeds the predetermined threshold growth, the method further comprises: identifying the sample as a high value positive; alerting a user of the high value positive; identifying the coordinates of the high value positive; transferring the coordinates of the colony to a picking device that will pick the colony from the biological sample; transferring the biological sample to the picking device and picking the colony from the biological sample; and alerting a user that further review is needed.

24. The method of claim 18, wherein in response to determining that the colony is a complex colony, the method further comprises: reporting, by a quadrant quantification module, that the plate is complex; and determining, by the quadrant quantification module, whether the growth exceeds a predetermined threshold growth.

Images