Prediction of presence of histopathologic abnormalities

Abstract

The present invention relates to the application of data toxicological genomics methods to detect and / or predict histopathological abnormalities in a human or animal subject based on clinical pathological data. It has been observed that reliable results can be obtained without any image data. A computer-implemented method of predicting the presence of histopathologic abnormalities in an organ of a human or animal subject based on clinical pathological data includes: receiving clinical pathological data obtained from the human or animal subject; applying an analytical model to the clinical pathological data, the analytical model configured to output a result indicative of a likelihood of the presence of a histopathologic abnormality in the organ of the human or animal subject; and outputting the result.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method for identifying histopathological abnormalities in human or animal subjects. Corresponding methods and systems are also provided. Background Art

[0002] Drug development includes safety assessment of test compounds in animals to determine their safety for humans. Preclinical toxicity studies include in vivo, laboratory, molecular, and autopsy assessments in animals such as rodents, dogs, and nonhuman primates. Toxicogenomic studies are toxicology studies in which gene expression in certain organs is correlated with standard toxicology endpoints such as clinical pathology and histopathology, such as the study in Uehara et al. (2010).

[0003] As in human pathology, digital and computational techniques are increasingly used in toxicologic pathology, see Abels et al. (2019) (Note: Abels, Esther, et al. (2019) "Computational pathology definitions, best practices, and recommendations for regulatory guidance: a white paper from the Digital Pathology Association." The Journal of pathology 249.3 (2019): 286-294.), Turner et al. (2020) (Note: Turner, Oliver C., et al. "Society of toxicologic pathology digital pathology and image analysis special interest group article*: Opinion on the application of artificial intelligence and machine learning to digital toxicologic pathology." Toxicologic pathology 48.2 (2020): 277-294.), Turner et al., (2021) (Note: Turner, Oliver C., et al. "Mini Review: The Last Mile—Opportunities and Challenges for Machine Learning in Digital Toxicologic Pathology." Toxicologic pathology 49.4(2021): 714-719), and Mehrvar et al. (2021) (Note: Mehrvar S, Himmel LE, Babburi P, Goldberg AL, Guffroy M, Janardhan K, Krempley AL, Bawa B. Deep learning approaches and applications in toxicologic histopathology: Current status and future perspectives.J Pathol Inform [online series] 2021 [cited 2021 Nov 16];12:42. Available from: https: / / www.jpathinformatics.org / text.asp?2021 / 12 / 1 / 42 / 329733).

[0004] These new technologies will have a huge impact on the timeline of histopathology assessment. In addition, it will help improve the quality of histopathology data. Improve the quality, reproducibility, and rigor of histopathology data; uniformly set thresholds for morphological changes in control tissues; and shorten timelines. Computational pathology is not limited to detecting lesions or morphological patterns of lesions, but involves the integration, complex analysis, and interpretation of a wide range of assays for disease diagnosis, treatment, and prognosis. Summary of the invention

[0005] The present invention relates to the application of data toxicogenomics methods to detect and / or predict histopathological abnormalities in human or animal subjects based on clinical pathology data. Crucially, it has been observed that reliable predictions can be obtained without any image data. Therefore, a first aspect of the present invention provides a computer-implemented method for predicting the presence of a histopathological abnormality in an organ of a human or animal subject based on clinical pathology data, the computer-implemented method comprising: receiving clinical pathology data obtained from the human or animal subject; applying an analysis model to the clinical pathology data, the analysis model being configured to output a result indicating the likelihood of the presence of a histopathological abnormality in an organ of the human or animal subject; and outputting the result.

[0006] The term "predict" may refer to determining at least one numerical or categorical value indicative of the presence of a histopathological abnormality.

[0007] The term "subject" as used herein typically relates to mammals, but may also refer to animals of other classes. The subject may be suffering from or should be suspected of having a disease, i.e., it may have shown some or all of the negative symptoms associated with the disease. In the present application, the organ of the human or animal subject is preferably the liver. However, the computer-implemented method is equally applicable to other organs, such as the kidney. There are various types of histopathological abnormalities, and the computer-implemented method of the first aspect of the present invention can be used to detect their presence. However, in preferred cases, the histopathological abnormality is a lesion.

[0008] As previously mentioned, it has been observed that reliable predictions can be made in the absence of image data. Thus, by using the computer-implemented method of the present invention, the presence of histopathological abnormalities in an organ of a human or animal subject can be reliably predicted without the need for invasive or expensive procedures required to obtain histopathology slide images. For the sake of completeness, it is prudent to state that in a preferred embodiment of the present invention, clinical pathology data does not include image data. Here, "image data" refers to data, such as electronic data, representing an image of the region of the organ in question. The image can be a photograph of a histopathology slide or can be obtained using a range of well-known medical imaging techniques.

[0009] Alternatively, in a preferred embodiment of the present invention, the analytical model can be run only on clinical pathology data. In the context of the present application, "clinical pathology data" is used to refer to data that can be obtained non-invasively, and generally relates to the presence and amount of one or more analytes in the body fluid of a human or animal subject. Body fluid can include a sample of a tissue / organ of the subject and / or a sample of a product produced by the tissue / organ of the subject. The product produced by the tissue / organ of the subject can be, for example, a secretory product (e.g., glandular secretion, milk, colostrum, tears, saliva, sweat, cerumen, mucus), sputum, semen, vaginal fluid / cervical fluid, blood (plasma, serum), cerebrospinal fluid (CSF), excretion products, feces or urine, skin or hair.

[0010] Clinical pathology data can include the measurement of the concentration of one or more analytes in the body fluid of a human or animal subject. The analyte can include one or more biomarkers. Herein, the term "biomarker" refers to a biomolecule found in blood, other body fluids or tissues, which serves as a sign of a normal or abnormal process or condition or disease. In some cases, clinical pathology data can include the measurement of the concentration of one or more analytes in one or more body fluids. For example, a first subset of clinical pathology data can include the measurement of the concentration of one or more analytes in a first body fluid, and a second subset of clinical pathology data can include the measurement of the concentration of one or more analytes in a second body fluid. Of course, this can be extended to a general variety of body fluids.

[0011] The one or more biomarkers may include one or more of the following: liver injury biomarkers, muscle injury biomarkers, and kidney injury biomarkers. Herein, the term "injury biomarker" is used to refer to those biomarkers that are present in higher concentrations when there is an abnormality or other injury in the organ or tissue in question.

[0012] The analyte may also contain one or more electrolytes.

[0013] Clinical pathology data may include a ratio of the concentration of a first analyte to the concentration of a second analyte (or vice versa). Clinical pathology may include a plurality of such ratios. Specifically, clinical pathology data may include a ratio of the concentration of albumin in a bodily fluid of a human or animal subject to the concentration of globulin in a bodily fluid of a human or animal subject. Alternatively, and for completeness, clinical pathology data may include a ratio of the concentration of globulin in a bodily fluid of a human or animal subject to the concentration of albumin in a bodily fluid of a human or animal subject. This ratio is typically used to help a clinician identify the cause of a change in protein levels in a user's bodily fluid.

[0014] Liver injury biomarkers may include bilirubin; aspartate aminotransferase (Note: This may also be called "glutamate oxaloacetate aminotransferase."); γ-glutamyltransferase (Note: This may also be called "γ-glutamyltranspeptidase."); alanine aminotransferase (Note: This may also be called "glutamate pyruvate aminotransferase."); and one or more of lactate dehydrogenase. In preferred cases, liver injury biomarkers may include bilirubin; aspartate aminotransferase; γ-glutamyltransferase; alanine aminotransferase; and lactate dehydrogenase. Liver injury biomarkers may further include one or more of the following: albumin; alkaline phosphatase (Note: This may also be called "serum alkaline phosphatase" or "plasma alkaline phosphatase."); cholesterol; globulin; glucose; protein; and triglycerides.

[0015] In preferred cases, the electrolyte may include potassium. The electrolyte may further include one or more of: calcium; chloride; phosphate; and sodium.

[0016] In a preferred embodiment, the muscle damage biomarker may include creatinine kinase. High levels of creatinine kinase in the blood generally indicate recent muscle damage.

[0017] Kidney injury biomarkers may include one or more of the following: creatinine and urea nitrogen. These biomarkers are commonly used to study kidney function.

[0018] In a particularly preferred embodiment of the first aspect of the invention, the clinical pathology data comprises: the ratio of the albumin concentration in the body fluid of the human or animal subject to the globulin concentration in the body fluid of the human or animal subject, and the concentration of each of the following in the body fluid of the human or animal subject: bilirubin; aspartate aminotransferase; gamma glutamyl transferase; alanine aminotransferase; lactate dehydrogenase; potassium; and creatinine kinase.

[0019] After discussing the nature of clinical pathology data in detail, we now list more information about analysis models. When applied to clinical pathology data, the analysis model is configured to output the result of the possibility of the existence of histopathological abnormalities in the organs of human or animal subjects. Herein, the term "analysis model" may refer to a mathematical model configured to predict at least one target variable for at least one state variable. The term "target variable" may refer to a clinical value to be predicted. The target variable value to be predicted may depend on the disease or condition to be predicted by its presence or state. The target variable may be a numerical variable or a categorical variable. For example, the target variable may be a categorical variable, and may be "positive" in the presence of a disease, or may be "negative" in the absence of a disease. The term "state variable" as used herein may refer to an input variable that can be filled in a prediction model, such as data obtained by a physical examination and / or self-examination of a subject. The state variable may be determined in at least one active test and / or at least one passive monitoring.

[0020] The target variable may be a numerical variable, such as at least one value and / or scale value. In this case, the state variable may include clinical pathology data, and the target variable may include an indication of the likelihood of the presence of a histopathological abnormality in an organ of a human or animal subject.

[0021] The analysis model can be a regression model or a classification model. In the context of the present application, the term "regression model" can be used to refer to an analysis model whose output is a numerical value within a certain range. For example, in the present case, the output of such a regression model can be a numerical value corresponding to the possibility or probability of the presence of a histopathological abnormality in an organ of a human or animal subject. In the context of the present application, the term "classification model" can be used to refer to an analysis model whose output is a binary classification or score indicating the presence or absence of a histopathological abnormality in an organ of a human or animal subject.

[0022] Specifically, the analysis model can be a machine learning model that is trained to output a result based on an input including clinical pathology data obtained from a human or animal subject, the result indicating the likelihood of the presence of a histopathological abnormality in the organ of the human or animal subject. The machine learning model can be a regression model or a classification model, as previously defined. The machine learning model is preferably trained using supervised learning. Therefore, in a preferred embodiment, the machine learning model is a random forest model or a gradient boosting model. Herein, when we refer to a machine learning model as a "random forest model", we mean that the machine learning model is trained using a random forest algorithm. Similarly, when we refer to a machine learning model as a "gradient boosting model", we mean that the machine learning model is trained using a gradient boosting algorithm.

[0023] The random forest algorithm is a supervised learning algorithm that combines the outputs of multiple decision trees to obtain a single result. When the random forest algorithm is used for regression, the output can include the mean or average prediction of multiple individual decision trees. When random forest is used for classification, the output can include the class selected by the majority of individual decision trees. You can use methods such as Breiman (2001) (Note: Breiman, L. Random Forests. Machine Learning 45, 5–32 (2001). https: / / doi.org / 10.1023 / A:1010933404324), Ooka et al. (2021) (Note: Ooka T, JohnoH, Nakamoto K, et al. Random forest approach for determining risk prediction and predictive factors of type 2 diabetes: large-scale health check-up data in Japan. BMJ Nutrition, Prevention & Health 2021;bmjnph-2020-000200. doi:10.1136 / bmjnph-2020-000200) and Christodoulous et al. (2022) (Note: “Random forest classification algorithm for medical industry data” Christodoulos Vlachas,Lazaros Damianos, Nikolaos Gousetis, Ioannis Mouratidis, DimitriosKelepouris, Konstantinos-Filippos Kollias, Nikolaos Asimopoulos and George FFragulis; SHS Web Conf., 139 (2022) 03008;DOI: https: / / doi.org / 10.1051 / shsconf / 202213903008). Other examples of random forest algorithms are also applicable.

[0024] The gradient boosting algorithm is a supervised learning algorithm that generates a predictive model in the form of an ensemble of weak predictive models (such as decision trees). Gradient boosting tree models are constructed in a step-wise manner like other boosting methods, but it generalizes other methods by allowing the optimization of arbitrary differentiable loss functions. Examples of gradient boosting algorithms that can be used include XGBoost (Note: https: / / arxiv.org / abs / 1603.02754; https: / / www.ncbi.nlm.nih.gov / pmc / articles / PMC6511546 / ).

[0025] The analysis model is preferably configured to output a histopathology score indicating the likelihood of the presence of a histopathological abnormality in an organ of a human or animal subject. This is true regardless of whether the analysis model is a regression model or a classification model. In a preferred case, the histopathology score is a binary score (e.g., "1" or "0", but it should be understood that any binary score can be used). Here, a "binary score" is a score that can take only two values. Preferably, one of these values ​​corresponds to a prediction of the presence of a histopathological abnormality, while the other value corresponds to a prediction of the absence of a histopathological abnormality.

[0026] The second aspect of the present invention provides a computer-implemented method for generating a machine learning model, the machine learning model being configured to output a result indicating the likelihood of the presence of a histopathological abnormality in an organ of a human or animal subject, the computer-implemented method comprising: receiving training data, for each of a plurality of human or animal subjects, the training data comprising: clinical pathology data and a histopathology score indicating whether a histopathological abnormality is present in the organ of the human or animal subject; and training the machine learning model using the received training data. Unless clearly incompatible, the optional features described above with respect to the first aspect of the present invention are equally applicable to the second aspect of the present invention. Specifically, training the machine learning algorithm may include using a random forest algorithm or a gradient boosting algorithm. In a preferred case, the analysis model of the first aspect of the present invention is a machine learning model generated according to the computer-implemented method of the second aspect of the present invention.

[0027] In some cases, the training data may include multiple subsets of training data, each set originating from a different source. In these cases, it may be necessary to standardize the training data to reduce the variability between subsets of training data from different sources. Data standardization can also reduce the variability within a given subset of training data. Therefore, prior to the step of training the machine learning model, the computer-implemented method may further include normalizing the received training data. A variety of normalization techniques may be employed to achieve this purpose, including logarithmic normalization, batch normalization, z-score normalization, and the use of a location scale model.

[0028] As is well known, the effectiveness of a machine learning model may depend on the quality of the training data used to train the machine learning model. Therefore, a computer-implemented method for generating a machine learning model configured to output a result indicating the likelihood of the presence of a histopathological abnormality in an organ of a human or animal subject may include: receiving training data, the training data including, for each of a plurality of human or animal subjects: clinical pathology data and a histopathology score indicating the presence or absence of a histopathological abnormality in an organ of the human or animal subject; using the training data in a first manner to train a first machine learning model; using the training data in a second manner to train a second machine learning model; and selecting the first trained machine learning model or the second trained machine learning model based on a performance indicator.

[0029] In some cases, the first approach and the second approach may refer to using different subsets of training data to train corresponding machine learning models. Specifically, the training data may include a first subset of training data and a second subset of training data. Then, the first subset of training data may be used to train the first machine learning model, and the second subset of training data may be used to train the second machine learning model. The first subset of training data and the second subset of training data may not overlap, that is, there may be no individual subjects represented in both the first subset of training data and the second subset of training data. Alternatively, the first subset of training data may partially overlap with the second subset of training data. However, the first subset of training data should not be the same as the second subset of training data.

[0030] In other cases, the first approach and the second approach may refer to using different training algorithms to train respective machine learning models. Specifically, the first machine learning model may be trained using the first training algorithm, and the second machine learning model may be trained using the second training algorithm. For example, the first training algorithm may be a random forest algorithm, and the second training algorithm may be a gradient boosting algorithm. Alternatively, both the first training algorithm and the second training algorithm may be random forest training algorithms (although different), or both the first training algorithm and the second training algorithm may be gradient boosting algorithms (again, although different).

[0031] This can be further generalized: the computer implementation can include training each of the multiple machine learning models in a respective manner. Then, the final step can include selecting one of the multiple trained machine learning models based on the performance metric. "Each respective manner" can refer to, for example, a specific subset of training data and a combination of training algorithms. In this way, when there are multiple training data available, the subset of training data that results in the best performance can be identified. Similarly, when there are multiple training algorithms available for a given machine learning model, this enables a user to select the training algorithm with the best performance.

[0032] We now discuss performance metrics. There are a variety of metrics that can be used to evaluate the performance of machine learning algorithms (https: / / scikit-learn.org / stable / auto_examples / miscellaneous / plot_roc_curve_visualization_api.html). In the preferred case, the performance metric is the area under the receiver operating characteristic curve (often abbreviated as AUC or AUC-ROC), which quantifies the ability of the classification model to accurately perform classification. Alternative performance metrics that can be used include F1 score, precision, recall, and classification accuracy.

[0033] In order to obtain the performance indicator, it is usually necessary to apply the trained machine learning model to the test data. In a preferred case, there is no overlap between the test data and the training data used to train the model to avoid training bias. Therefore, the computer-implemented method may further include determining a first value of the performance indicator of the first trained machine learning model; and determining a second value of the performance indicator of the second trained machine learning model; and selecting the first trained machine learning model or the second trained machine learning model that is associated with a better score. Herein, "better" is used to refer to a more favorable score, that is, a score indicating better performance. In many cases, this will be a higher score, but in some cases, a lower score may indicate better performance. In each case, determining the corresponding value of the performance indicator may include applying the trained machine learning model to the test data, and determining the value of the performance indicator based on the output of the machine learning model. Preferably, each trained machine learning model is applied to the same test data to ensure consistency between tests.

[0034] The foregoing aspects of the invention relate to computer-implemented methods. Another aspect of the invention may provide a system for predicting the presence of a histopathological abnormality in an organ of a human or animal subject based on clinical pathology data, the system comprising a computer-implemented method configured to perform any of the foregoing aspects of the invention. The optional features set forth with reference to the foregoing aspects of the invention also apply to this system aspect of the invention, unless clearly incompatible.

[0035] Another aspect of the invention may provide a computer program comprising instructions which, when executed by a computer or its processor, causes it to perform the steps of the computer-implemented method of the aforementioned aspect of the invention. The optional features set forth with reference to the aforementioned aspect of the invention also apply equally to this aspect of the invention unless clearly incompatible. Yet another aspect of the invention provides a computer-readable storage medium having the computer program of the aforementioned invention stored thereon.

[0036] The present invention includes any combination of described aspects and preferred features unless such a combination is expressly impermissible or explicitly avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Embodiments of the present invention will now be described with reference to the accompanying drawings, in which:

[0038] FIG. 1 is a schematic diagram of a system that may be used to implement the present invention.

[0039] FIG2 is a flow chart illustrating the generation of a machine learning model that can be used to predict the presence or absence of a histopathological abnormality.

[0040] FIG3 is a flow chart illustrating the use of a machine learning model to predict the presence or absence of a histopathological abnormality. Detailed description with drawings

[0041] Aspects and embodiments of the present invention will now be discussed with reference to the accompanying drawings. Other aspects and embodiments will be apparent to those skilled in the art. All documents mentioned in this text are incorporated herein by reference.

[0042] FIG. 1 shows a system 1 that can be used to perform a computer-implemented method according to the present invention. The system 1 includes a data acquisition unit 100, an analysis system 200, and a display module 300. The data acquisition unit 100, the analysis system 200, and the display module 300 are all interconnected via a network 400. The network 400 can be a wired network (such as a LAN or WAN), or a wireless network (such as a Wi-Fi network, the Internet, or a cellular network). In some cases, the data acquisition unit 100, the analysis system 200, and the display module 300 can be connected via multiple networks 400 (not shown). For example, the acquisition module 100 and the analysis system 200 can be connected via a first network, the data acquisition unit 100 and the display module 300 can be connected via a second network, and the analysis system 200 and the display module 300 can be connected via a third network. Other combinations are contemplated. Alternatively, some subsets of the data acquisition unit 100, the analysis system 200, and the display module 300 can be integrated with each other. For example, the data acquisition unit 100, the analysis system 200, and the display module 300 may all be integrated into a single system, such as a smartphone, a desktop computer, a laptop computer, or a tablet computer. In some cases, the data acquisition unit 100 and the display module 300 may be client-facing modules or user-facing modules (i.e., they may be accessed by an end user or client of the system), while the analysis system 200 may be located remotely, such as on a server (such as a back-end server). Alternatively, the analysis system 200 may be located on the cloud, such that the processing performed occurs external to the user device, on a server (or the like) with higher computing power. Various other arrangements are contemplated.

[0043] In the context of the present application, the data acquisition unit 100 is a unit that may include hardware and software components that is suitable for obtaining clinical pathology data from a human or animal subject. For example, a clinician or other scientist may obtain a body fluid sample from the human or animal subject in question and, for example, use dedicated hardware to generate clinical pathology data using the obtained sample. Of course, in the context of the present invention, the clinical pathology data is preferably electronic data that contains information related to the concentration of various analytes in the body fluid of the human or animal subject in question, as outlined elsewhere in this patent application.

[0044] The analysis system 200 is where most of the analysis that is core to the present invention is performed. The analysis system 200 includes a processor 204 and a memory 208. The processor 204 includes a plurality of modules. In the context of the present application, a module can be implemented in hardware or software and is adapted or configured to perform a specific function. For example, it can be used to refer to a physical element within a processor, or it can refer to, for example, a code segment including instructions, which, when executed by the processor 204, causes it to perform the function in question. Specifically, the processor 204 includes a pre-processing module 2041, a training module 2042, a testing module 2044, a selection module 2046, and an analysis module 2048. The respective functions of each of these modules will be discussed later.

[0045] The memory 206 of the analysis system 200 may include a persistent memory and a temporary memory. In the specific embodiment shown in FIG1 , the memory 206 stores: a gradient boosting algorithm 2062, a random forest algorithm 2064, training data 2066 (including two subsets 20662, 20664), and test data 2068. The memory 206 also includes a cache 2070, which may store, for example, clinical pathology data 20702 (explained in more detail later) while processing is ongoing.

[0046] Having described the structure of system 1, we now explain its operation with reference to Figures 2 and 3. Figures 2 and 3 are high-level flow charts illustrating, respectively, a computer-implemented method for generating a machine learning model and a computer-implemented method for using a machine learning model to predict the presence of a histopathological abnormality in an organ of a human or animal subject based on clinical pathology data.

[0047] FIG. 2 illustrates a computer-implemented method for generating a machine learning model that can be used to predict the presence of histopathological abnormalities in an organ of a human or animal subject based on clinical pathology data. In a first step, the analysis system 200 receives training data 2066. The training data 2066 can be received from an appropriate source, such as from one or more external databases (not shown, but non-limiting examples are given in the experimental results section). Then, in step S101, the preprocessing module 2041 of the processor 204 preprocesses the training data. This can include data normalization, as outlined elsewhere in this application, in order to reduce variability between and within data sets. The overall goal of the training scheme shown in FIG. 2 is to train multiple machine learning models in a variety of different ways, evaluate the performance of each trained machine learning model, and select the trained machine learning model with the best performance. Therefore, in step S102, the (first) machine learning model is trained in a first way using the training module 2042. This may mean several things: such as, the machine learning model may be trained using a first type of training algorithm (e.g., gradient boosting algorithm 2062 or random forest algorithm 2064), or the machine learning model may be trained using a first subset of data (e.g., subset 20662 or subset 20664). After the first machine learning model has been trained, in step S104, it is determined whether the training module 2042 has trained all machine learning models. Here, "all" machine learning models refer to training machine learning models in each of the different ways of training them. This may cover, for example, training machine learning models using all combinations of subsets 20662, 20664 of training data 2066 and training algorithms (such as gradient boosting algorithm 2062 and random forest algorithm 2064). If it is determined in step S104 that not all machine learning models have been trained, the computer-implemented method returns to step S102 and trains another machine learning model in a different way than the first machine learning model. This iterative process continues until it is determined in step S104 that all machine learning models have been trained. The computer-implemented method then proceeds to step S106, where the testing module 2044 evaluates the performance of each trained machine learning model. Preferably, the evaluation of the testing module 2044 involves the determination or calculation of performance indicators, as outlined elsewhere in this application. The testing module 2044 can be configured to apply each trained machine learning model to the test data 2068, and determine or calculate the performance indicator based on the results of applying the trained machine learning model to the test data 2068. The test data 2068 can be a subset of the training data 2062, which is not used to train the machine learning model (to avoid training bias).In the example of FIG. 2 , performance evaluation is performed after all machine learning models have been trained. However, it is equally feasible to perform performance evaluation on a given machine learning model (immediately) after it is generated. Alternatively, the performance evaluation process and the training process can be performed in parallel. Other arrangements are contemplated. After the testing module 2044 has evaluated the performance of all trained machine learning models in step S106, the computer-implemented method moves to step S108, where the selection module 2046 selects the trained machine learning model with the best performance. This is preferably done based on the value of the performance indicator. The selected machine learning model 2072 can then be stored in the memory 206 of the analysis system 200.

[0048] FIG. 2 relates to the generation of a machine learning model. FIG. 3 illustrates a computer-implemented method according to a first aspect of the invention, wherein the presence or absence of a histopathological abnormality is predicted. In a first step S200, clinical pathology data 20702 is received. For example, the processor 204 of the analysis system 200 may receive the clinical pathology data 20702 from the data acquisition unit 100. For processing, the clinical pathology data 20702 may be stored in a cache 2070 of the memory 206 of the analysis system 200. Then, in step S202, the analysis model 2048 may retrieve the trained machine learning model 2072 from the memory 206 of the analysis system and apply it to the clinical pathology data 20702. As discussed, the trained machine learning model 2072 is preferably configured to output a binary histopathology score indicating the presence or absence of a histopathological abnormality based on the clinical pathology data 20702. Then, in step S204, a result is output. This may include processor 204 transmitting the results to display module 300 where they may then be displayed, for example, to a clinician or other client.

[0049] Experimental Results

[0050] We now list the results of studies demonstrating the effectiveness of the computer-implemented method according to the present invention. These experiments focus on detecting histopathological lesions in rat livers in toxicity studies. The data in the experimental studies were from rats for which clinical pathological data and histopathological assessments of the liver were available. Data from intermediate bleeding points not associated with histopathology and data from recovered animals were excluded.

[0051] Three data sets (i.e. three sets of training data) are used for analysis:

[0052] (i) TG-GATE: An online open-source dataset for rat toxicogenomics studies, including whole slide images (20x) and associated clinical pathology data and histopathology results including classical test items with well-characterized liver toxicity. The final dataset for liver cases consists of 3,894 individual animal entries. Only male rats are present in this dataset.

[0053] (ii) Roche Toxicogenomics Program: Dataset of rat studies of the Toxicogenomics Program conducted between 1999 and 2003. The test items used in these studies are classical hepatotoxicants or nephrotoxicants. Hepatotoxicants cause toxicity via the following mechanisms: direct action, steatosis, cholestasis, or immune-mediated. Some Roche compounds that were withdrawn due to clinical hepatotoxicity are also included. The dataset consists of 1,768 entries, 1,705 males and 63 females.

[0054] (iii) Roche Toxicity Studies: 600 or similar SEND (Standardized Exchange of Nonclinical Data) studies completed in the last 20 years. This dataset spans over 20 years and contains 43,878 entries extracted from over 600 studies. From this larger dataset, a smaller subset used in another project (called (iv)) was extracted and tested separately. This extracted dataset consists of 832 entries extracted from 34 studies.

[0055] As we explain later, different combinations of the above datasets are also tested to evaluate the impact on the AUC of the machine learning models generated using different datasets.

[0056] All data sets were visually inspected using the data analysis software Spotfire (TIBCO).

[0057] Earlier in this application, we mentioned using data normalization to reduce variability. In these experiments, a variety of normalization techniques were tested to minimize variability within and between datasets (batch effects), including: log normalization, batch normalization, z-score normalization, and position-scale models.

[0058] We now consider the location scale model reference range. Due to the diversity of datasets in terms of time range and sources, normalized reference values ​​were collected from the literature. Among the various location scale models, the inventors chose the Chang-Stein model (note: https: / / www.lexjansen.com / phuse / 2019 / dh / DH05.pdf). The idea of ​​the Chuang-Stein model is to normalize all values ​​related to a selected set of reference ranges:

[0059]

[0060] In this article:

[0061] value is the number to be normalized.

[0062] ULN and LLN are the upper and lower values ​​in the data set for the variable in question, respectively.

[0063] ULN std and LLN std are, respectively, the upper and lower values ​​from a standard reference list for the variable in question.

[0064] Reference values ​​were collected from the literature. Due to the higher prevalence of male subjects than female subjects in the dataset, male reference values ​​were considered in this specific context. The same reference ranges were applied to all examined datasets to obtain comparable data.

[0065] For all datasets, missing values ​​were replaced by randomly generated values.

[0066] Histopathology scores were calculated in two ways depending on the dataset type.

[0067] For the Roche Toxicogenomics Program studies, a senior toxicology pathologist manually reviewed the histologic diagnosis and severity to assign a final score. As a general approach, controls were set as normal (0) unless a specific finding of concern was detected (e.g., liver necrosis). Treated animals were evaluated based on the histologic score, with scores of 1 out of 5 or lower set as normal (0). Samples with a histologic score of 2 out of 5 or higher were set as pathologic (1). An exception to this rule was hepatic glycogen storage disease, which was always considered normal.

[0068] For the Roche toxicity studies and TG-GATE, a Python script analyzed the severity of the findings reported by the pathologists and generated an automated score. The script was set up to consider findings labeled “not significant,” “present,” or 1 out of 5 in severity as normal (denoted as 0 in the table). Findings with higher severity were set as pathological (1). A junior toxicology pathologist manually reviewed the automated scores to be similar to those given by the senior toxicology pathologist for toxicogenomics.

[0069] Then process the data using:

[0070] 1. Random forest and gradient boosting implementation in Orange (Note: https: / / orangedatamining.com / ).

[0071] 2. Implement random forest in Python using scikit-learn (Note: https: / / scikit-learn.org / stable / modules / generated / sklearn.ensemble.RandomForestClassifier.html).

[0072] To avoid exposing the algorithm in question to the test information in advance, each dataset was split by study ID. The histopathology score was set as the target value.

[0073] The Orange settings are as follows:

[0074]

[0075] The setup for the Python implementation is as follows:

[0076]

[0077] High variability within and between datasets was observed. Of note, there was high variability between TG-GATE and toxicogenomics, whereas variability was lower between TG-GATE and Roche general studies and between the Roche toxicogenomics program and Roche toxicity studies. Variability within datasets was particularly high in the toxicogenomics studies.

[0078] After normalization, the data distributions of the three data sets are similar, as shown in Table 1, Table 2 and Table 3 attached to this patent application.

[0079] Among all types of data normalization, position-scale model is an effective technique for normalizing clinical pathology data.

[0080] The results for each dataset acquired individually processed in Orange are as follows, where the datasets have the same labels as assigned above.

[0081]

[0082] CA = Classification Accuracy

[0083] F1 = The harmonic mean of precision and recall.

[0084] Precision: https: / / en.wikipedia.org / wiki / Precision_and_recall

[0085] When the Roche Toxicogenomics Project (ii) dataset was combined with other datasets and processed in Orange, the results were as follows:

[0086]

[0087] When we combine dataset (iv) with TG-GATE (i), the results for Orange are as follows:

[0088]

[0089] The Roche toxicity dataset (iii) was combined with TG-GATE (i) and processed in Orange:

[0090]

[0091] Dataset (iv) was not included in the test because it is a subset of the Roche toxicity dataset.

[0092] Results of 10-fold cross validation random forest:

[0093]

[0094] A high variability in predictivity was observed between datasets, which was independent of the number of samples. Based on our preliminary results, the Roche toxicity data performed better when used in combination with TG-GATE. The combination of the Roche toxicity and TG-GATE datasets had an AUC of 0.82, suggesting that clinical pathology can be used for liver histopathology lesion detection.

[0095] General Statement

[0096] The features disclosed in the foregoing description, or in the following claims, in terms of the manner of expressing or implementing the disclosed functions in their specific forms, or in terms of the methods or processes for obtaining the disclosed results, may be appropriately used alone in their various forms, or in any combination to implement the present invention.

[0097] Although the present invention has been described in conjunction with the above exemplary embodiments, many equivalent modifications and variations will be apparent to those skilled in the art when this disclosure is given. Therefore, the above exemplary embodiments of the present invention are considered to be illustrative rather than restrictive. Various changes may be made to the described embodiments without departing from the spirit and scope of the present invention.

[0098] For the avoidance of any doubt, any theoretical explanations provided herein are intended to improve the reader's understanding. The inventors do not wish to be bound by any of these theoretical explanations.

[0099] Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

[0100] Throughout the specification, including the following claims, unless the context requires otherwise, the words "comprise" and "include" and variations such as "comprises and comprising" and "including", will be understood to imply the inclusion of stated integers or steps or groups of integers or steps but not the exclusion of any other integers or steps or groups of integers or steps.

[0101] It must be noted that, as used in this specification and the appended claims, the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" one particular value and / or to "about" another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values ​​are expressed as approximations, by use of the antecedent "about", it will be understood that the particular value forms another embodiment. The term "about" in relation to a numerical value is optional and means, for example, + / - 10%.

[0102] Appendix – Table 1: TG-GATE Summary

[0103]

[0104] Appendix – Table 2: Summary of Roche’s Toxicogenomics Program

[0105]

[0106] Appendix – Table 3: Summary of Roche Toxicity Studies

[0107]

Claims

1. A computer-implemented method for predicting the presence of a histopathological abnormality in an organ of a human or animal subject based on clinical pathology data, the computer-implemented method comprising: receiving clinical pathology data obtained from said human or animal subject; applying an analytical model to the clinical pathology data, the analytical model being configured to output a result indicative of a likelihood of the presence of a histopathological abnormality in the organ of the human or animal subject; and The result is output.

2. The computer-implemented method of claim 1, wherein: The clinical pathology data comprises a ratio of the albumin concentration in a body fluid of the human or animal subject to the globulin concentration in the body fluid of the human or animal subject.

3. The computer-implemented method of claim 1 or claim 2, wherein: The clinical pathology data comprises one or more concentration measurements of one or more corresponding liver injury biomarkers in a bodily fluid of the human or animal subject.

4. The computer-implemented method of claim 3, wherein: The one or more liver injury biomarkers include: bilirubin; aspartate aminotransferase; gamma glutamyl transferase; alanine aminotransferase; and lactate dehydrogenase.

5. The computer-implemented method of any one of claims 1 to 4, wherein: The clinical pathology data comprises a measurement of creatinine kinase concentration in a bodily fluid of the human or animal subject.

6. A computer-implemented method according to any one of claims 1 to 5, wherein: The clinical pathology data comprises a measurement of potassium concentration in a body fluid of the human or animal subject.

7. A computer-implemented method according to any one of claims 1 to 6, wherein: The clinical pathology data does not include image data.

8. The computer-implemented method of any one of claims 1 to 7, wherein: The analytical model is a machine learning model that is trained to output a result based on an input comprising clinical pathology data obtained from the human or animal subject, the result indicating the likelihood of the presence of a histopathological abnormality in the organ of the human or animal subject.

9. The computer-implemented method of claim 8, wherein: The machine learning model is a random forest algorithm or a gradient boosting algorithm.

10. The computer-implemented method of any one of claims 1 to 9, wherein: The analytical model is configured to output a histopathology score indicative of a likelihood of the presence of a histopathological abnormality in the organ of the human or animal subject.

11. The computer-implemented method of claim 10, wherein: The histopathology score is a binary score.

12. The computer-implemented method of any one of claims 1 to 11, wherein: The organ of the human or animal subject is the liver.

13. The computer-implemented method of any one of claims 1 to 12, wherein: The histopathological abnormality is a lesion.

14. A computer-implemented method of generating a machine learning model configured to output a result indicative of a likelihood of the presence of a histopathological abnormality in an organ of a human or animal subject, the computer-implemented method comprising: receiving training data, the training data comprising, for each of a plurality of human or animal subjects: clinical pathology data and a histopathology score indicating the presence or absence of a histopathological abnormality in an organ of the human or animal subject; and The machine learning model is trained using the received training data.

15. The computer-implemented method of any one of claims 1 to 13, wherein the analytical model is a machine learning model trained according to the computer-implemented method of claim 14.

Images