Method and apparatus for predicting therapeutic response to HER2-targeted therapeutic agents

Abstract

The present invention relates to a method or apparatus for predicting the therapeutic response to HER2-targeted therapies, and according to one embodiment of the method, it is possible to more accurately predict the therapeutic response of cancer patients to HER2-targeted therapies and determine the treatment method.

Description

Technical Field

[0001] Relates to a method and apparatus for predicting a therapeutic response to a HER2-targeted therapeutic agent.

Background Art

[0002] Cancer therapeutic agents can be classified into first-generation chemotherapy agents that attack not only cancer cells but also normal cells, second-generation targeted therapeutic agents that selectively attack only cancer cells, and immuno-oncology therapeutic agents that activate the immune system so that lymphocytes surrounding cancer tissue selectively attack tumor cells.

[0003] In particular, human epidermal growth factor receptor 2 (HER2), a transmembrane tyrosine kinase receptor, has been shown to be involved in various types of cancer, so cancer therapeutic agents targeting HER2 have been developed in various ways. Among them, the development of trastuzumab, a humanized anti-HER2 monoclonal antibody (mAb), has greatly improved the clinical outcomes for HER2-positive breast cancer, providing a basis for further development of HER2-targeting approaches, including tyrosine-kinase inhibitors, novel monoclonal antibodies, and antibody-drug conjugates such as trastuzumab deruxtecan.

[0004] However, cancer drugs can cause side effects and are not always effective for cancer patients. Furthermore, treatments such as chemotherapy or targeted therapy can cause physical and / or mental distress to cancer patients. Therefore, before administering treatments such as immunotherapy or targeted therapy to a cancer patient, it is important to determine whether such treatments are effective for the patient, i.e., whether they can remove all or at least part of the tumor in the cancer patient.

[0005] On the other hand, current HER2-targeted therapy uses HER2 as a biomarker, employing the degree of HER2 protein expression in tumor tissue as the primary criterion for predicting a cancer patient's response to HER2-targeted drugs. Specifically, HER2-targeted therapy involves collecting tissue from patients before treatment, staining it using immunohistochemistry (IHC), and having a pathologist manually calculate the HER2 protein expression level in the stained areas. Patients with an expression level above a certain threshold are then administered HER2-targeted drugs. However, a single IHC-stained whole slide image can contain 1,000 to 1,000,000 tumor cells, and the process of quantifying cells with different staining intensities depending on HER2 protein expression levels is labor-intensive and time-consuming. Therefore, objectively, accurately, and rapidly quantifying HER2 expression without the help of a computing system is not easy.

[0006] The present invention provides a method for predicting the therapeutic response to HER2-targeted therapies using a computing device, namely an AI-based IHC analyzer, and for improving the prediction of a patient's therapeutic response to HER2-targeted therapies based on the patient's HER2 expression level and / or tumor microenvironment (TME) factors using the computing device. [Overview of the project] [Problems that the invention aims to solve]

[0007] The objective is to provide a method and apparatus for predicting the therapeutic response to HER2-targeted therapeutic drugs. Furthermore, it is also to provide a computer-readable recording medium storing a program for executing the method on a computer. The technical problems to be solved are not limited to those described above, and other technical problems may exist. [Means for solving the problem]

[0008] A method for predicting a therapeutic response to a HER2-targeted therapy according to one embodiment includes the steps of: identifying the HER2 staining intensity of tumor cells (TCs) in a first pathology slide image of a patient using a machine learning model; determining the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity based on the identified results; and predicting the patient's therapeutic response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity.

[0009] A method for providing information to predict a therapeutic response to a HER2-targeted therapy in another aspect includes the steps of: using a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in a first pathology slide image of a patient; determining the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity based on the identified results; and predicting the patient's therapeutic response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity.

[0010] Another embodiment of a computer-readable recording medium includes a recording medium that stores a program for causing a computer to perform the above-described method.

[0011] A computing system in yet another embodiment includes at least one memory and at least one processor connected to the memory and configured to execute at least one computer-readable program contained in the memory,

[0012] The at least one processor is configured to use a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in a patient's first pathology slide image by executing the at least one program, to determine the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity based on the identified results, and to predict the patient's therapeutic response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity.

[0013] Another embodiment provides a method for treating cancer, comprising the steps of: using a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in a first pathological slide image of a patient; determining the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity based on the identified results; predicting the patient's response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity; and, if it is determined that the patient will respond to the HER2-targeted therapy based on the treatment response prediction results, administering the HER2-targeted therapy to the patient. [Effects of the Invention]

[0014] By using a method or apparatus for predicting the therapeutic response to a HER2-targeted therapy according to one embodiment, it is possible to more accurately predict the therapeutic response of cancer patients to a HER2-targeted therapy and determine the appropriate treatment method. [Brief explanation of the drawing]

[0015] [Figure 1]Figure 1 illustrates an example of a computing system for generating analysis results of pathological images according to one embodiment. [Figure 2] Figure 2 is a diagram illustrating an example of a computing system for predicting therapeutic response to a HER2-targeted therapeutic agent according to one embodiment. [Figure 3] Figure 3 illustrates an example of a processor according to one embodiment that analyzes pathological images. [Figure 4] Figure 4 is a flowchart illustrating an example of a method for predicting the therapeutic response to a HER2-targeted drug according to one embodiment. [Figure 5] Figure 5 shows the results of confirming the HER2 status of colorectal cancer patients using HER2 IHC scores and HER2 gene amplification rates on patients' pathology slides. Figure 5A shows the results of evaluating HER2 IHC scores in the Prescreening cohort (N=144) and the TRIUMPH cohort (n=30) using an AI-powered HER2 quantification continuous score (QCS) analyzer. Figure 5B shows the results of analyzing HER2 amplification rates in the TRIUMPH study via HER2 IHC (immunohistochemistry) and FISH (fluorescence in situ hybridization) results or ctDNA analysis (circulating tumor DNA analysis) results. HER2 overexpression or amplification is indicated as HER2-positive (HER2+). [Figure 6]Figure 6 is a graph analyzing the HER2 status of patients enrolled in the TRIUMPH trial using HER2 IHC, FISH, ctDNA analysis results, and the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop, %) determined by the AI ​​model (the x-axis shows the HER2 IHC (immunohistochemistry) score evaluated by pathologists, and the y-axis shows the HER2 / CEP17 ratio determined by FISH (fluorescence in situ hybridization)). Figure 6A shows "HER2+ (positive), HER2- (negative), or unexamined case" determined by HER2 ctDNA analysis as colored dots, and Figure 6B shows "the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop, %)" determined by the AI ​​model as colored dots. [Figure 7] Figure 7 is a graph showing the results of evaluating the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop) in pathology slides (WSI) from the Prescreening cohort and the TRIUMPH cohort using an AI-powered HER2 QCS analyzer. [Figure 8] Figure 8 shows photographs of representative clinical cases from the TRIUMPH cohort with low (18.1%) AI-H3 prop and high (97.4%) AI-H3 prop in pathology slides with a HER2 IHC 3+ score. Figure 8A shows a photograph of a representative clinical case with low (18.1%) AI-H3 prop in pathology slides with a HER2 IHC 3+ score from the TRIUMPH cohort, and Figure 8B shows a photograph of a representative clinical case with high (97.4%) AI-H3 prop in pathology slides with a HER2 IHC 3+ score from the TRIUMPH cohort. [Figure 9]Figure 9 is a graph showing the percentages of tumor cells (TC) with HER2 staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) in slides classified by an AI-powered HER2 QCS analyzer in the Prescreening cohort and the TRIUMPH cohort according to HER2 IHC scores. [Figure 10] Figure 10 is a confirmation of the coincidence rate of HER2 immunohistochemistry (IHC) scores of pathological slides evaluated by an AI-powered HER2 QCS analyzer and pathologists for patients in the TRIUMPH cohort. [Figure 11] Figure 11 is a graph showing the clinical outcomes of the total HER2+, Path HER2+, ctDNA HER2+, Path HER2 IHC 3+, AI-H3 prop≧10% and AI-H3 prop≧50% groups. [Figure 12] Figure 12 is a waterfall plot showing the tumor regression percentages from baseline for patients in the TRIUMPH trial who were administered pertuzumab and Trastuzumab. [Figure 13] Figure 13 is a graph showing the correlation between the percentage of tumor cells with HER2 IHC 3+ (H3) staining intensity measured by an AI-powered HER2 analyzer and the change in tumor regression in patients in the TRIUMPH trial who were administered pertuzumab and Trastuzumab. [Figure 14]FIG. 14 is a graph showing the proportion of tumor cells (AI-H0 prop, AI-H1 prop, AI-H2 prop or AI-H3 prop) having each HER2 staining intensity (HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2) or HER2 IHC 3+ (H3)) in the TRIUMPH study. FIG. 14A is a graph showing the distribution of the proportion of tumor cells (AI-H0 prop, AI-H1 prop, AI-H2 prop and AI-H3 prop) having HER2 staining intensity of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2) or HER2 IHC 3+ (H3) by classifying the TRIUMPH cohort slides by HER2 IHC score, and FIG. 14B is a graph showing the proportion of tumor cells (AI-H0 prop, AI-H1 prop, AI-H2 prop and AI-H3 prop) having HER2 staining intensity of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2) and HER2 IHC 3+ (H3) for each case in the TRIUMPH cohort. [Figure 15] FIG. 15 is a graph showing the overall response rate (ORR) and Kaplan-Meier curves of progression-free survival (PFS) and overall survival (OS) according to whether AI-H3 prop is less than 50% (<50%) or 50% or more (≧50%). [Figure 16] FIG. 16 is a graph showing a comparative analysis of AI analysis tumor microenvironment factors (lymphocyte (LC) density, macrophage (MP) density, fibroblast (FB) density and endothelial cell (EC) density) between responders and non-responders in TRIUMPH patients. [Figure 17]Figure 17 is a graph comparing tumor microenvironment factors (TME factors) in the TRIUMPH cohort between the group with less than 50% AI-H3 prop (AI-H3 prop < 50%) and the group with 50% or more AI-H3 prop (AI-H3 prop ≥ 50%). [Figure 18] Figure 18 is a graph comparing TME factors quantified by an AI-based TME analyzer in the TRIUMPH cohort group with AI-H3 prop ≥ 50% (AI-H3 prop ≥ 50%) between responders (patients who achieved complete response (CR) or partial response (PR) according to RECIST V1.1 criteria) and non-responders (patients who achieved stable disease (SD) or progression (PD)). [Figure 19] Figure 19 is a graph showing Kaplan-Meier curves representing the quantitative percentage of tumor cells exhibiting AI-based HER2 IHC 3+(H3) staining intensity (AI-H3 prop) and progression-free survival (PFS) based on AI-based tumor microenvironment factors. Figure 19A shows the Kaplan-Meier curve for AI-H3 prop and lymphocyte density in the tumor stromal region (LC-CS), Figure 19B shows the Kaplan-Meier curve for AI-H3 prop and macrophage density in the tumor stromal region (MP-CS), Figure 19C shows the Kaplan-Meier curve for AI-H3 prop and fibroblast density in the tumor stromal region (FB-CS), and Figure 19D shows the Kaplan-Meier curve for AI-H3 prop and endothelial cell density in the tumor stromal region (EC-CS). [Figure 20]Figure 20 shows Kaplan-Meier curves representing the quantitative percentage of tumor cells exhibiting AI-based HER2 IHC 3+(H3) staining intensity (AI-H3 prop) and overall survival (OS) based on AI-based tumor microenvironment factors. Figure 20A shows the Kaplan-Meier curve for AI-H3 prop and lymphocyte density in the tumor stromal region (LC-CS), Figure 20B shows the Kaplan-Meier curve for AI-H3 prop and macrophage density in the tumor stromal region (MP-CS), Figure 20C shows the Kaplan-Meier curve for AI-H3 prop and fibroblast density in the tumor stromal region (FB-CS), and Figure 20D shows the Kaplan-Meier curve for AI-H3 prop and endothelial cell density in the tumor stromal region (EC-CS). [Figure 21] Figure 21 is a graph showing the hazard ratios (HR) for lymphocyte density (LC-CS), fibroblast density (FB-CS), macrophage density (MP), and endothelial cell density (EC-CS) in the cancer stromal region (CS) for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%). Figure 21A is a graph showing the hazard ratio (HR) for PFS for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%), and Figure 21B is a graph showing the hazard ratio (HR) for OS for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%). [Figure 22]Figure 22 is a graph showing the hazard ratios (HR) for lymphocyte density (LC-CA), fibroblast density (FB-CA), macrophage density (MP), and endothelial cell density (EC-CS) in cancer areas (CA) for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%). Figure 22A is a graph showing the hazard ratio (HR) for PFS for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%), and Figure 22B is a graph showing the hazard ratio (HR) for OS for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%). [Figure 23] Figure 23 shows the Kaplan-Meier curves illustrating the progression-free survival (PFS) and overall survival (OS) rates for the "LC-CS, MP-CS, and FB-CS combinations." [Figure 24] Figure 24 shows representative clinical cases of the TME-high or TME-low group within the TRIUMPH cohort, where 50% or more of tumor cells exhibit HER2 IHC 3+(H3) staining intensity (AI-H3 prop ≥ 50%). Figure 24A shows representative clinical cases of PD treatment response in the "AI-H3 prop ≥ 50% group with TME-high," and Figure 24B shows representative clinical cases of CR treatment response in the "AI-H3 prop ≥ 50% group with TME-low." [Figure 25] Figure 25 is a nomogram for estimating 12-month and 24-month overall survival (OS) in patients in whom the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is ≥50% (AI-H3 prop ≥50%). [Figure 26] Figure 26 is a graph showing the correlation between factors analyzed in HER2 IHC WSI and H&E WSI via AI-based HER2 QCS analyzer and AI-based TME analyzer. [Modes for carrying out the invention]

[0016] The terminology used in the embodiments has been selected to the greatest extent possible from currently widely used and common terms, although this may change depending on the intent of the articulators, case law, the emergence of new technologies, etc. In certain cases, the applicant may have arbitrarily selected terms, in which case their meaning will be described in detail in the relevant explanatory section. Therefore, the terminology used in the specification should not be merely names of terms, but should be defined based on the meaning of the term and its context throughout the specification.

[0017] Throughout the specification, when a part "includes" a component, this means, unless otherwise stated, that it may include other components rather than excluding them. Furthermore, terms such as "~unit," "~module," and "~part" as used herein mean a unit that performs at least one function or operation, which may be implemented in hardware or software, or in a combination of hardware and software.

[0018] According to one embodiment of this specification, a “module” or “part” may be implemented by a processor and memory. “Processor” should be broadly interpreted to include general-purpose processors, central processing units (CPUs), GPUs (Graphics Processing Units), microprocessors, digital signal processors (DSPs), controllers, microcontrollers, state machines, and the like. In some environments, “processor” may refer to on-demand semiconductors (ASICs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), and the like.

[0019] In this specification, “each of the A's” or “each of the A's” may refer to each of all the components included in the A's, or to each of some of the components included in the A's. For example, “each of the tumor cells” may refer to each of all the tumor cells included in the A's, or to each of some of the tumor cells included in the A's. Similarly, “each of the immune cells” may refer to each of all the immune cells included in the A's, or to each of some of the immune cells included in the A's.

[0020] In this specification, "similarity" can encompass all meanings of being identical or similar. For example, two pieces of information being similar can mean that the two pieces of information are identical or similar to each other.

[0021] In this specification, “instruction” may refer to a set of instructions grouped on a functional basis, which are components of a computer program and are executed by a processor.

[0022] The embodiments will be described in detail below with reference to the attached drawings. However, the embodiments can be realized in a variety of different forms and are not limited to the examples described herein.

[0023] Figure 1 illustrates an example of a computing system 10 that generates analysis results of pathological images 20 according to one embodiment.

[0024] Referring to Figure 1, the computing system 10 can receive pathological images 20 and generate analysis results 30 of the pathological images 20. Here, the analysis results 30 and / or medical information generated based on the analysis results 30 can be used to predict the therapeutic response to HER2-targeted therapies in cancer patients.

[0025] In Figure 1, the computing system 10 is shown as a single computing device, but is not limited to this; the computing system 10 may be configured to distribute information and / or data through multiple computing devices. Furthermore, although Figure 1 does not show a storage system that can communicate with the computing system 10, the computing system 10 may be configured to connect to or communicate with one or more storage systems. The computing system 10 may be any computing device used to generate the analysis results 30 of pathological images 20. Here, the computing device can refer to any type of device with computing capabilities, such as a laptop, desktop, laptop, server, cloud system, etc., but is not limited to these.

[0026] A memory system configured to communicate with the computing system 10 may be a device or cloud system for storing and managing various data related to pathological image analysis tasks. To efficiently manage the data, the memory system can use a database to store and manage the various data. Here, the various data may include any data related to pathological image analysis. For example, the various data may include histological information (histological components) regarding the type, location, and state of cells, tissues, and / or structures contained in the pathological image 20. The various data may further include clinical factors such as the patient's age, menopausal status, clinical T stage (Clinical_T), BIRADS, number of tumors, tumor size, node enlargement, biopsy_ER, biopsy_PR, biopsy_HER2, pCR_final, pathology type, and homologous recombination deficiency (HRD).

[0027] The computing system 10 can receive pathological images 20 obtained from the human tissue of a patient who is a target for predicting the therapeutic response to a HER2-targeted therapy drug. Such pathological images 20 can be received via a communicable storage medium (e.g., a hospital system, a local / cloud storage system, etc.). The computing system 10 can analyze the received pathological images 20 to generate analysis results 30 of the pathological images 20. Here, the pathological images 20 may include histological information (histological components) about at least one patch contained in the image.

[0028] As used herein, the term "patch" can refer to a small area within a pathological image. For example, a patch may include a region corresponding to a semantic object extracted by segmenting a pathological image. Alternatively, a patch may refer to a combination of pixels associated with histological information generated by analyzing a pathological image.

[0029] The term “histological components” as used herein may include characteristics or information relating to cells, tissues, and / or structures within the human body contained in pathological images. Here, cell characteristics may include cytologic features such as nucleus, cytoplasm, and cell membrane. The histological components may refer to histological information relating to at least one patch contained in a pathological image inferred through a machine learning model. The term “annotation” means the task of tagging histological information to a data sample or the tagged information (i.e., annotation) itself. The term “annotation” may be used interchangeably with terms such as tagging and labeling in the art.

[0030] According to one embodiment, the computing system 10 can extract histological information, which is a characteristic of cells, tissues, and / or structures within the human body of a target patient, by analyzing the pathological image 20. Specifically, the computing system 10 can extract histological information about at least one patch contained in the pathological image 20 by analyzing (e.g., inference) the pathological image 20 using a machine learning model. For example, the histological information may include, but is not limited to, information about the cells within the patch (e.g., tumor cells, lymphocytes, macrophages, dendritic cells, fibroblasts, endothelial cells, etc.) (e.g., the number of specific cells, information about the tissue in which the specific cells are located). Here, tumor cells can refer to cells that continue to overgrow, disregarding the cell growth cycle, and malignant tumor cells that penetrate (invade) surrounding tissues and spread and grow (metastasize) to distant tissues can be called cancer cells.

[0031] According to one embodiment, the computing system 10 can detect the expression of biomarkers in cells contained in a pathological image 20, or extract information about biomarker expression from the pathological image 20. For example, the pathological image 20 may include an image stained using an immunohistochemical (IHC) method. Furthermore, the computing system 10 can extract information about the expression levels of biomarkers of interest from the stained pathological image 20. For example, the computing system 10 can identify the HER2 staining intensity from the pathological image 20. The information processing system 10 can calculate a quantitative continuous score (QCS) from the pathological image 20 that indicates the percentage of tumor cells (TCs) showing a specific HER2 staining intensity. Furthermore, the computing system 10 can use the calculated quantitative continuous score (QCS) to predict the therapeutic response of cancer patients to HER2-targeted therapies. For example, the computing system 10 can predict the therapeutic response of cancer patients to HER2-targeted therapies based on the QCS indicating the percentage of tumor cells showing HER2 IHC 3+ staining intensity.

[0032] Another example of histological information extracted from the patch by the computing system 10 may include information about the tissue within the patch (e.g., cancer area, cancer epithelial, cancer stroma, normal epithelial, normal stroma, necrosis, fat, background, etc.). Another example of histological information extracted from the patch by the computing system 10 may include information about tumor microenvironment factors (TME factors). Examples of tumor microenvironment factors (TME factors) may include, but are not limited to, lymphocyte (LC) density, macrophage (MP) density, fibroblast (FB) density, endothelial cell (EC) density, etc. The histological information extracted from the patch is not limited to the examples above and may include any quantifiable histological information from the patch, such as cellular instability, cell cycle, and biological function.

[0033] As described above, the analysis results 30 generated by the computing system 10 and / or the medical information generated / output based on the analysis results 30 can be used to predict and / or output the outcome of the treatment response to HER2-targeted therapies. Furthermore, by utilizing patient clinical information related to pathology slides received from an accessible external system as additional input data, it is possible to infer and / or output the predicted treatment response to HER2-targeted therapies for cancer patients.

[0034] The following describes an example of how the computing system 10 analyzes pathological images, with reference to Figures 2 to 4.

[0035] Figure 2 is a diagram illustrating an example of a computing system for predicting therapeutic response to a HER2-targeted therapeutic agent according to one embodiment.

[0036] Referring to Figure 2, the computing system 200 includes a processor 210 and memory 220. For convenience of explanation, only components relevant to the present invention are shown in Figure 2. Therefore, in addition to the components shown in Figure 2, other general-purpose components may be further included in the computing system 200. For example, the computing system may include, but is not limited to, at least one of a server device and a cloud device. As another example, the computing system may consist of one or more server devices. As yet another example, the computing system may consist of one or more cloud devices. As yet another example, the system may consist of a server device and a cloud device configured and operating together. Furthermore, it will be apparent to those ordinary skill in the art related to the present invention that the processor 210 and memory 220 shown in Figure 2 can be implemented as independent devices.

[0037] The processor 210 can process computer program instructions by performing basic arithmetic, logic, and input / output operations. Instructions may be provided from memory 220 or an external device (e.g., server 20). Furthermore, the processor 210 can provide overall control over the operation of other components included in the computing system 200.

[0038] The processor 210 may be implemented as an array of multiple logic gates, or as a combination of a general-purpose microprocessor and memory storing a program that can be executed by this microprocessor. For example, the processor 210 may include a general-purpose processor, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a controller, a microcontroller, a state machine, and the like. In some environments, the processor 210 may include an on-demand semiconductor (ASIC), a programmable logic device (PLD), a field-programmable gate array (FPGA), and the like. For example, the processor 210 may refer to a combination of processing units such as a combination of a digital signal processor (DSP) and a microprocessor, a combination of multiple microprocessors, a combination of one or more microprocessors coupled with a digital signal processor (DSP) core, or any other combination of such configurations.

[0039] The processor 210 analyzes pathological images. The analysis of pathological images includes the process of dividing or detecting specific tissues and / or cells in various tissues and cells expressed on the pathology slide and then converting them into information useful for making medical decisions. Here, the process of converting into information useful for making medical decisions can mean extracting features that can be used to classify the disease pathology of a patient, diagnose cancer, establish a cancer treatment plan, prescribe anticancer drugs, or predict the likelihood of developing cancer, and in particular, it may include predicting the treatment response to HER2-targeted therapies in cancer patients.

[0040] For example, the processor 210 can identify tumor cells from pathological images using at least one machine learning model and identify the HER2 staining intensity of each tumor cell. For example, the processor 210 can obtain information on HER2 protein expression in whole slide images (WSI) stained by immunohistochemical staining (IHC) and quantify HER2 expression based on the obtained information. More specifically, the processor 210 can classify the degree of HER2 expression (HER2 state) into several steps according to the degree of HER2 staining in tumor cells. Here, the multiple steps may be the four most commonly used steps (0, 1+, 2+, 3+), but are not limited to these, and can be classified into fewer or more steps. Specifically, the processor can classify cells identified from pathological images into five categories: HER2 IHC 0 (H0) tumor cells, HER2 IHC 1+ (H1) tumor cells, HER2 IHC 2+ (H2) tumor cells, HER2 IHC 3+ (H3) tumor cells, and other cells (OT). H0 indicates tumor cells (TCs) without cell membrane staining, H1 indicates tumor cells (TCs) with partial cell membrane staining of faint or almost inconspicuous intensity, H2 indicates tumor cells (TCs) with partial lateral or periphery cell membrane staining of weak / moderate intensity, and H3 indicates tumor cells (TCs) with complete lateral or periphery cell membrane staining of strong intensity. OT indicates cells other than tumor cells.

[0041] Furthermore, the processor 210 can calculate a quantitative continuous score (QCS) indicating the proportion of tumor cells at each HER2 staining intensity, based on the results of identifying the degree of HER2 expression (HER2 status) of each tumor cell based on HER2 staining intensity using at least one machine learning model. The quantitative continuous score (QCS) calculated by the processor 210 can be said to represent the proportion of tumor cells having each HER2 staining intensity (HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), HER2 IHC 3+ (H3)).

[0042] Furthermore, the processor 210 can calculate the percentage of tumor cells with the aforementioned HER2 staining intensity and classify pathology slides obtained from the patient's human tissue as HER2 IHC 3+ positive (if AI-H3 prop > 10%), HER2 IHC 2+ equivocal (if AI-H3 prop < 10% or AI-H2 prop > 10%), HER2 IHC 1+ negative (if AI-H1 prop > 10%), and HER2 IHC 0 negative (if AI-H1 prop ≤ 10% or if only H0 cells are present). Here, AI-H3 prop represents the percentage (%) of tumor cells showing HER2 IHC 3+(H3) staining intensity as determined by processor 210 in pathology slides obtained from the patient's tissue, AI-H2 prop represents the percentage (%) of tumor cells showing HER2 IHC 2+(H2) staining intensity as determined by processor 210 in pathology slides obtained from the patient's tissue, and AI-H1 prop represents the percentage (%) of tumor cells showing HER2 IHC 1+(H1) staining intensity as determined by processor 210 in pathology slides obtained from the patient's tissue.

[0043] The internationally agreed-upon diagnostic criteria for the aforementioned HER2 status can be interpreted as the guidelines for interpreting HER2 IHC trial results presented by the SCO (Standards and Guidelines Committee) and the CAP (College of American Pathologists). Here, the SCO / CAP guidelines are guidelines for the accurate evaluation and reporting of HER2-positive tumors used in pathological examination, and represent guidance for accurately interpreting the HER2 status of tumors through HER2 IHC (Immunohistochemistry) evaluation and HER2 FISH (Fluorescence In Situ Hybridization) evaluation.

[0044] As a result of the analysis, if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity among all tumor cells is above a predetermined value, the pathology slide images may be classified as cases in which the patient responds to HER2-targeted therapy, or cases in which cancer patients have good drug sensitivity and treatment prognosis to HER2-targeted therapy.

[0045] On the other hand, the processor 210 can identify tumor microenvironment factors (TME factors) from pathological images using at least one machine learning model. For example, the processor 210 can analyze whole slide images (WSI) stained with H&E to identify various objects such as cancerous areas, cancerous stroma areas, tumor cells, and immune cells, and identify tumor microenvironment factors (TME factors). These tumor microenvironment factors (TME factors) may be measured in cancerous areas (CA) or cancerous stroma areas (CS) and may include lymphocyte (LC) density, macrophage (MP) density, fibroblast (FB) density, endothelial cell (EC) density, etc.

[0046] Furthermore, as a result of the analysis, if one or more of the values ​​of lymphocyte (LC) density, macrophage (MP) density, fibroblast (FB) density, and endothelial cell (EC) density are below a predetermined value, the processor 210 may output a classification of the pathology slide image as a case in which the patient responds to HER2-targeted therapy, or a case in which there is good drug sensitivity and treatment prognosis to HER2-targeted therapy.

[0047] Memory 220 may include non-temporary, readable storage media. For example, memory 220 may include permanent mass storage devices such as RAM (random access memory), ROM (read-only memory), disk drives, SSDs (solid-state drives), and flash memory. Alternatively, permanent mass storage devices such as ROM, SSDs, flash memory, and disk drives may be independent persistent storage devices separate from memory. Memory 220 may also store an operating system (OS) and at least one program code.

[0048] These software components can be loaded from a computer-readable storage medium separate from memory 220. Such a separate computer-readable storage medium may be a storage medium that can be directly connected to the computing system 200 and may include, for example, computer-readable storage media such as floppy drives, disks, tapes, DVD / CD-ROM drives, and memory cards.

[0049] On the other hand, although not shown in Figure 2, the computing system 200 may further include a display device. Alternatively, the computing system 200 may be connected to an independent display device by wired or wireless means, and data may be sent and received from each other. For example, pathological images, pathological slide images, analysis information of pathological slide images, medical information, and additional information based on medical information can be provided to the user via the display device.

[0050] Figure 3 illustrates an example of a processor analyzing pathological images according to one embodiment. The following describes an example of processor 210 analyzing pathological images with reference to Figure 3. The examples described later with reference to Figure 3 can all be applied to the processor 210 for tumor cell identification, identification of HER2 staining intensity of tumor cells, calculation of a quantitative continuous score (QCS), determination of the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity, and deriving tumor microenvironment factors (TME factors).

[0051] Referring to Figure 3, the processor 210 can analyze the pathological image 310 using a machine learning model 320. For example, the processor 210 can derive analysis results of the pathological image 310, including quantitative continuous scores 330. Specifically, the analysis results of the pathological image 310 can be derived as the percentage of each tumor cell having each HER2 staining intensity (HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), HER2 IHC 3+ (H3)).

[0052] For example, the processor 210 can use a machine learning model 320 to output the detection results in the form of layers showing the tissue in the pathological image 310. In this case, the machine learning model 320 can be trained to detect regions in the pathological image 310 that correspond to the tissue in the reference pathological slide images, using training data that includes multiple reference pathology slide images and multiple reference label information.

[0053] The processor 210 can perform classification on multiple tissues represented in the pathological image 310. Specifically, the processor 210 can classify the pathological image 310 into at least one of the following: cancer area, cancer stroma area, necrosis area, and background area. Alternatively, the processor 210 can classify the pathological image 310 into at least one of invasive cancer area and carcinoma in situ.

[0054] However, the examples of the processor 210 classifying at least some of the regions represented in the pathological image 310 are not limited to those described above. In other words, the processor 210 can classify at least one region represented in the pathological image 310 into multiple categories according to various criteria, not limited to the six types of regions described above (cancer region, cancerous stromal region, necrotic region, background region, invasive cancer region, and carcinoma in situ). At least one region represented in the pathological image 310 can be classified into multiple categories according to pre-set criteria or user-set criteria.

[0055] The processor 210 can then analyze the pathological image 310 and perform classification on the multiple cells represented in the pathological image 310.

[0056] First, the processor 210 analyzes the pathological image 310 to detect cells from the pathological image 310 and can output the detection results in the form of a layer showing the cells.

[0057] The processor 210 can use a machine learning model 320 to output detection results in the form of layers showing cells in the pathological image 310. In this case, the machine learning model 320 can be trained to detect the location and type of cells in the reference pathological slide images within the pathological image 310 using training data that includes multiple reference pathological slide images and multiple reference label information.

[0058] The processor 210 can then perform classification on multiple cells represented in the pathological image 310. For example, the processor 210 can classify multiple cells expressed in the pathological image 310 into at least one of the following: tumor cells, lymphocytes, macrophages, fibroblasts, endothelial cells, and other cells.

[0059] However, the examples of how the processor 210 classifies cells represented in the pathological image 310 are not limited to those described above. In other words, the processor 210 can classify cells represented in the pathological image 310 into multiple categories according to various criteria, not limited to the six types of cells mentioned above (i.e., tumor cells, lymphocytes, macrophages, fibroblasts, endothelial cells, and other cells). Cells represented in the pathological image 310 can be grouped into multiple categories according to pre-set criteria or user-defined criteria.

[0060] Following the process described above, the processor 210 can analyze the pathological image 310 to generate information about at least one subject and / or biomarker expression information. For example, the biomarker expression information may be, but is not limited to, the HER2 staining intensity or quantitative continuous score (QCS) 330 of tumor cells.

[0061] Here, machine learning model 320 refers to a statistical learning algorithm or a structure that executes such an algorithm, which is based on the structure of a biological neural network.

[0062] For example, machine learning model 320 can demonstrate a model that possesses problem-solving ability by having nodes, which are artificial neurons that form a network through synaptic connections, similar to a biological neural network, repeatedly adjust the weights of the synapses and learn to reduce the error between the correct output corresponding to a particular input and the inferred output. For example, machine learning model 320 can include any probabilistic model, neural network model, etc., used in artificial intelligence learning methods such as deep learning.

[0063] For example, the machine learning model 320 can be implemented using a multilayer perceptron (MLP) consisting of multilayer nodes and connections between them. The machine learning model 320 according to this embodiment can be implemented using one of various artificial neural network model structures, including MLPs. For example, the machine learning model 320 can consist of an input layer that receives an input signal or data from the outside, an output layer that outputs an output signal or data corresponding to the input data, and at least one hidden layer located between the input layer and the output layer that receives a signal from the input layer, extracts characteristics, and transmits them to the output layer. The output layer receives a signal or data from the hidden layer and outputs it to the outside.

[0064] Therefore, the machine learning model 320 can learn to receive one or more pathological images 310 and extract information about one or more objects (e.g., cells, tissues, structures, etc.) contained in the pathological images 310 and / or biomarker expression information.

[0065] Figure 4 is a flowchart illustrating an example of a method for analyzing pathological images according to one embodiment. The method shown in Figure 4 consists of steps processed chronologically by the computing systems 10, 200, or processor 210 shown in Figures 1 and 2. Therefore, even if the details are omitted below, the above-mentioned details regarding the computing systems 10, 200, or processor 210 shown in Figures 1 and 2 can also be applied to the method shown in Figure 4.

[0066] Furthermore, the method described later can be created using a computer-executable program and implemented on a general-purpose digital computer that runs the program using a computer-readable recording medium. Also, the data structure used in the method described above can be recorded on a computer-readable recording medium through various means.

[0067] The recording medium readable by the aforementioned computer includes recording media such as magnetic recording media (e.g., ROM, RAM, USB, floppy disk, hard disk, etc.) and optical recording media (e.g., CD-ROM, DVD, etc.).

[0068] Referring to Figure 4, in step 410, the processor 210 uses a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in the patient's first pathology slide image. The first pathology slide image can mean an image scanned from a pathology slide that has been fixed and stained through a series of chemical processing steps for microscopic observation of tissues or other materials taken from the human body. The pathology image can refer to a whole slide image (WSI) containing a high-resolution image of the entire slide and may include, but preferably, an immunohistochemical (IHC) stained slide or an immunohistochemical (IHC) stained slide. The first pathology slide image can also refer to a portion of the high-resolution whole slide image, for example, one or more patches. Furthermore, the pathology image can refer to a digital image obtained by scanning the pathology slide using a digital scanner and may contain information about cells, tissues, and / or structures within the human body.

[0069] In one embodiment, the step of identifying the HER2 staining intensity of tumor cells (TCs) in a patient's first pathology slide image using the machine learning model can mean that the processor 210 uses at least one machine learning model to obtain information related to HER2 protein expression in the whole slide image (WSI) stained by immunohistochemical staining (IHC), and quantifies HER2 expression based on the obtained information. Specifically, the information regarding HER2 protein expression may be obtained by the processor 210 using at least one machine learning model to classify tumor cells from the first pathology slide image into HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3). Here, H0 indicates tumor cells (TCs) without cell membrane staining, H1 indicates tumor cells (TCs) with partial cell membrane staining at a faint or almost inconspicuous intensity, and H2 indicates tumor cells (TCs) with partial lateral or peripheral cell membrane staining at a weak / moderate intensity. H3 indicates tumor cells (TCs) with strong intensity and complete lateral or periphery-based cell membrane staining.

[0070] In step 420, based on the identification of tumor cells with HER2 staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) in the first pathology slide image, the percentage of tumor cells showing HER2 IHC 3+ (H3) staining intensity is determined. Specifically, the percentage of tumor cells showing HER2 IHC 3+ (H3) staining intensity in the first pathology slide image can be derived as a quantitative continuous score (QCS).

[0071] In step 430, the patient's response to the HER2-targeted therapy is predicted based on the percentage of tumor cells exhibiting the HER2 IHC 3+(H3) staining intensity. Specifically, the step of predicting the patient's response to the HER2-targeted therapy may involve determining that the patient will respond to the HER2-targeted therapy if the percentage of tumor cells exhibiting the HER2 IHC 3+(H3) staining intensity is 10% or more. For example, if the processor 210 determines, using at least one machine learning model, that the percentage of tumor cells (TCs) with HER2 staining intensity of HER2 IHC 3+(H3) in the patient's first pathology slide image is 10%, the result of classifying the pathology slide image as a case with good drug sensitivity and treatment prognosis to the patient's HER2-targeted therapy may be output.

[0072] In one embodiment, if the processor 210 determines, using at least one machine learning model, that the percentage of tumor cells in the patient exhibiting HER2 IHC 3+(H3) staining intensity is 50% or more, it can be determined that the patient will respond to a HER2-targeted therapy.

[0073] The method for predicting the therapeutic response to the aforementioned HER2-targeted therapy will be described in more detail below.

[0074] According to one embodiment, a method for predicting a therapeutic response to a HER2-targeted therapy is provided, comprising the steps of: using a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in a first pathology slide image of a patient; determining the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity based on the identified results; and predicting the patient's therapeutic response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity.

[0075] In one embodiment, a method for predicting the therapeutic response to the HER2-targeted therapy drug may include the steps of: using a machine learning model to calculate the percentage of tumor cells (TCs) having each HER2 staining intensity of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) from pathology slide images; deriving the percentage of tumor cells showing the HER2 IHC 3+ (H3) staining intensity as a quantitative continuous score (QCS); and predicting the patient's therapeutic response to the HER2-targeted therapy drug based on the percentage of tumor cells showing the HER2 IHC 3+ (H3) staining intensity derived from the quantitative continuous score (QCS).

[0076] In this specification, the term "pathological image" refers to an image obtained by scanning a pathological slide that has been fixed and stained through a series of chemical processes for microscopic observation of tissues or other materials removed from the human body.

[0077] The term "pathological image" may refer to a whole slide image (WSI) including a high-resolution image of the entire slide, and may include images relating to at least one of H&E (Hematoxylin & Eosin) stained slides, immunohistochemistry (IHC) stained slides, or pathological slides stained by various other staining methods. The term "pathological image" may refer to a portion of a high-resolution whole slide image, for example, one or more patches. Furthermore, the term "pathological image" may refer to a digital image obtained by scanning a pathological slide using a digital scanner, and may include information about cells, tissues, and / or structures within the human body. In this specification, "pathological image" may be used interchangeably with "pathological image," "pathological slide image," "tissue slide image," "pathological slide image," "whole slide image (WSI)," etc. Furthermore, in this specification, "pathological image" may also refer to "at least a portion of the area included in the pathological image."

[0078] The term "HER2 (Human epidermal growth factor receptor type 2)" as used herein refers to a protein involved in regulating cell proliferation, which, when overexpressed or amplified, can cause cells to develop into tumors or cancer.

[0079] In this specification, the term "tumor cell (TC)" can refer to cells that continue to proliferate excessively, disregarding the cell growth cycle. In particular, malignant tumor cells that invade surrounding tissues and spread and grow (metastasize) to distant tissues can be called cancer cells.

[0080] As used herein, the term "HER2 IHC score" refers to the evaluation of the expression level of HER2 receptor protein in tumor cells within pathological slides obtained from the patient's human tissue using the IHC (Immunohistochemistry) staining method, and the categorization of the pathological slides into HER2 IHC 0, HER2 IHC 1+, HER2 IHC 2+, and HER2 IHC 3+ scores according to the degree of staining of the tumor cells within the pathological slides, in accordance with internationally agreed diagnostic criteria.

[0081] Specifically, if the number of tumor cells with unstained cell membranes exceeds 90% of all tumor cells, the HER2 IHC score on the pathology slide may be determined to be 0. Conversely, if the number of tumor cells with complete and intensely stained cell membranes exceeds 10% of all tumor cells, the HER2 IHC score on the pathology slide may be determined to be 3.

[0082] If the number of tumor cells with complete and intense membrane staining is less than 10% of the total tumor cells, and the number of tumor cells with complete and intense membrane staining or weak-moderate membrane staining is 10% or more of the total tumor cells, the HER2 IHC score on the pathology slide may be determined to be 2.

[0083] If the number of tumor cells with complete and intense membrane staining is less than 10% of all tumor cells, and the number of tumor cells with complete and intense membrane staining or weak-moderate membrane staining is less than 10% of all tumor cells, and the number of tumor cells with complete and intense membrane staining or weak-moderate or light membrane staining is 10% or more of all tumor cells, then the HER2 IHC score on the pathology slide may be determined to be 1.

[0084] If the HER2 IHC score on the pathology slide is not determined to be 1, 2, or 3, all of the remaining cases may be determined to have a HER2 IHC score of 0.

[0085] The aforementioned internationally agreed-upon diagnostic criteria can refer to the guidelines for interpreting HER2 IHC test results presented by the SCO (Standards and Guidelines Committee) and the CAP (College of American Pathologists). Here, the SCO / CAP guidelines are guidelines for the accurate evaluation and reporting of HER2-positive tumors used in pathological examination, and refer to guidance for accurately interpreting the HER2 status of tumors through HER2 IHC (Immunohistochemistry) evaluation and HER2 FISH (Fluorescence In Situ Hybridization) evaluation. In accordance with the SCO / CAP guidelines, the HER2 status in tumor tissue obtained from the patient's human tissue can be determined.

[0086] For example, if the HER2 IHC score of the pathology slide determined according to the SCO / CAP guidelines is 0, it is considered HER2-negative; if the HER2 IHC score of the pathology slide is 3 (HER2 IHC 3+), it is considered HER2-positive; and if the HER2 IHC score of the pathology slide is 1+ (HER2 IHC 1+) or 2+ (HER2 IHC 2+), the HER2 FISH (Fluorescence In Situ Hybridization) evaluation can be further advanced to analyze whether it is HER2-negative or HER2-positive. The term "quantification continuous score (QCS)" as used herein refers to a quantitative analysis of HER2 expression based on information obtained from whole slide images (WSI) stained using immunohistochemical staining (IHC). Specifically, the quantitative continuous score can be defined as a value derived from classifying the HER2 protein expression levels of each tumor cell in pathological slide images obtained from the patient's human tissue using an AI model, based on the HER2 staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3). Based on the results of identifying tumor cells with each of the aforementioned HER2 staining intensities, the percentage of tumor cells exhibiting HER2 staining intensities of HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) can be derived.

[0087] For example, when classifying cells in pathology slide images obtained from a patient's tissue using an AI model into five categories—"tumor cells with staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), HER2 IHC 3+ (H3), and other cells (OT)"—a quantitative continuous score can mean a continuous value representing the percentage of tumor cells with each staining intensity. Tumor cells with HER2 IHC 0 (H0) staining intensity refer to tumor cells without cell membrane staining, tumor cells with HER2 IHC 1+ (H1) staining intensity refer to tumor cells with partial cell membrane staining at a weak or almost inconspicuous intensity, tumor cells with HER2 IHC 2+ (H2) staining intensity refer to tumor cells with partial lateral or periphery cell membrane staining at a weak / moderate intensity, and tumor cells with HER2 IHC 3+ (H3) staining intensity refer to tumor cells that are completely stained at a strong intensity on the lateral or periphery of the cell membrane. In one embodiment, the HER2 staining intensity of tumor cells (TCs) may include HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3).

[0088] In this specification, the term "HER2 IHC 3+(H3) staining intensity" refers to the degree to which the HER2 expression level in each tumor cell within a pathology slide is classified into IHC staining intensities (HER2 IHC 0(H0), HER2 IHC 1+(H1), HER2 IHC 2+(H2), HER2 IHC 3+(H3)) based on analysis of pathology slide images obtained from patient tissue using an AI-based HER2 analyzer. Specifically, the HER2 IHC 3+(H3) staining intensity indicates the degree to which the cell membrane of the tumor cell is completely stained with strong intensity. For example, the HER2 IHC 3+(H3) staining intensity can mean "complete and intense staining" as defined in the SCO / CAP guidelines.

[0089] The term "treatment response in cancer patients" as used herein may include pathological complete response, response to immunotherapy, etc. Here, "pathological complete response (pCR)" may mean, but is not limited to, the absence of invasive cancer in human tissue as a result of anticancer treatment. For example, pathological complete response may mean a state in which all or at least some of the tumor cells present in human tissue are removed as a result of anticancer treatment.

[0090] In this specification, the term "biomarker" refers to a marker that can be measured objectively, such as a normal or pathological state or the degree of response to a drug.

[0091] In one embodiment, the HER2 IHC score of a pathology slide image obtained from a patient's tissue can be determined based on the percentage of tumor cells (TCs) with a specific HER2 staining intensity in the slide image, and it can be predicted whether the patient will respond to a HER2-targeted therapy drug.

[0092] For example, if the percentage of cells with HER2 IHC 3+(H3) staining intensity in a pathology slide image obtained from a patient's tissue using the AI-powered HER2 analyzer exceeds 10% (AI-H3 prop > 10%), the HER2 IHC score of the pathology slide can be determined as HER2 IHC 3+. If the percentage of cells with HER2 IHC 3+(H3) staining intensity in a pathology slide image obtained from a patient's tissue is less than 10%, or if the percentage of cells with HER2 IHC 2+(H2) staining intensity exceeds 10% (AI-H3 prop < 10% or AI-H2 prop > 10%), the HER2 IHC score of the pathology slide image can be determined as HER2 IHC 2+ equivocal. Furthermore, if the percentage of cells with HER2 IHC 1+(H1) staining intensity in the pathology slide image obtained from the patient's tissue exceeds 10% (AI-H1 prop > 10%), the HER2 IHC score of the pathology slide image can be determined as HER2 IHC 1+ negative. If the percentage of cells with HER2 IHC 1+(H1) staining intensity in the pathology slide image obtained from the patient's tissue is 10% or less (AI-H1 prop ≤ 10%) or if only H0 cells are present, the HER2 IHC score of the pathology slide image can be determined as HER2 IHC 0 negative.

[0093] In this specification, the term "immunohistochemistry (IHC) staining" refers to a staining method that utilizes the principle of reacting an antibody of interest onto a tissue or cell sample in order to observe, under a light microscope, the presence or absence of a protein (antigen) present in the nucleus, cytoplasm, or cell membrane. Antigen-antibody reaction products cannot be directly observed under a microscope, so a method is used in which a marker is applied and then the marker is colored. As a color developer, red-tinged AEC (3-amino-9-ethylcarbazole) or brown-tinged DAB (3,3′-diaminobenzidine) can be used. Furthermore, in order to accurately identify the site of protein expression, counterstaining with hematoxylin can be performed after treatment with the color developer.

[0094] In one embodiment, at least one step included in the method may be performed by a machine learning model.

[0095] In this specification, the term “machine learning model” or “machine learning model” may mean a structure of a computer algorithm that learns from data to discover patterns, predict, or make decisions. The machine learning model can generally perform training on training data and then perform prediction or classification on new data. For example, the machine learning model may include any model used to infer an answer to a given input.

[0096] According to one embodiment, the machine learning model may include an artificial neural network model comprising an input layer, a plurality of occlusion layers, and an output layer, where each layer may include one or more nodes. For example, the machine learning model may be trained to infer histological information about a pathological image and / or at least one patch contained in the pathological image. In this case, the machine learning model can be trained using histological information generated by an annotation task. As another example, the machine learning model may be trained to infer the responsiveness of a cancer patient to treatment based on interaction scores, at least one characteristic of cells, tissues, or structures in the pathological image, and / or clinical information about the patient. Furthermore, the machine learning model may include weights associated with a plurality of nodes contained in the machine learning model. The weights may include any parameters associated with the machine learning model.

[0097] In this specification, machine learning models or machine learning models may refer to artificial neural network models, and artificial neural network models may refer to machine learning models or machine learning models. The machine learning models described herein may be models trained using various learning methods. For example, a variety of learning methods such as supervised learning, unsupervised learning, self-supervised learning, and reinforcement learning may be used in this disclosure, but are not limited to these.

[0098] As used herein, the term “learning” may refer to any process of modifying the weights included in a machine learning model using at least one patch, interaction score, histological information, and / or clinical information. According to one embodiment, learning may refer to the process of modifying or updating the weights associated with a machine learning model through one or more forward propagations and backward propagations using at least one patch and histological information.

[0099] In one embodiment, the HER2-targeted therapeutic agent may include one or more drugs selected from the group consisting of pertuzumab, trastuzumab, trastuzumab emtansine, lapatinib, neratinib, apatinib, tucatinib, and pyrotinib.

[0100] As used herein, the term "HER2-targeted therapeutic agent" may mean one that targets the overexpression of the HER2 protein in HER2-positive tumors. One such HER2-targeted therapeutic agent may, but is not limited to, a monoclonal antibody that specifically binds to the HER2 protein present on the surface of tumor cells. Furthermore, HER2-targeted therapeutic agents can be used to control or shrink tumors by inhibiting the growth and metastasis of HER2-positive tumors and by inducing tumor cell death.

[0101] In this specification, the term "pertuzumab" refers to a monoclonal antibody that belongs to the category of targeted anticancer drugs and targets the HER2 / neu receptor on cancer cells. When used in combination with trastuzumab, it is said to effectively inhibit HER2 signaling and slow tumor progression.

[0102] In this specification, the term "trastuzumab" refers to a targeted anticancer agent that acts by targeting the HER2 / neu receptor, which is overexpressed in cancer cells. Trastuzumab binds to the HER2 receptor, inhibiting tumor cell replication and slowing tumor progression.

[0103] In one embodiment, the step of predicting the therapeutic response of the cancer patient to a HER2-targeted therapy drug may include determining that the patient has good drug sensitivity and treatment prognosis to the HER2-targeted therapy drug, and that the cancer patient will respond to the HER2-targeted therapy drug, if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is 10% or more.

[0104] In one embodiment, the step of predicting the therapeutic response of the cancer patient to a HER2-targeted therapy drug may include determining that the patient will respond to the HER2-targeted therapy drug if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is 50% or more, and that the patient will have good drug sensitivity and therapeutic prognosis to the HER2-targeted therapy drug.

[0105] In this specification, the term “cancer” means a physiological condition in an animal that is typically characterized by abnormal or uncontrolled cell growth. Such cancer may be associated with, for example, metastasis, interference with normally functioning surrounding cells, release of cytokines or other secretory products at abnormal levels, suppression or increase of inflammatory or immunological responses, neoplasia, premalignant, malignancy, or invasion of surrounding or distant tissues or organs, such as lymph node invasion.

[0106] In one embodiment, pathological slide images can be obtained from tissue slides taken from solid tumor tissue. Specifically, the solid tumor can be one or more selected from the group consisting of lung cancer, skin cancer, stomach cancer, gastrointestinal cancer, intestinal cancer, colorectal cancer, colon cancer, pancreatic cancer, liver cancer, thyroid cancer, uterine cancer, cervical cancer, ovarian cancer, testicular cancer, prostate cancer, breast cancer, and oral cancer, but is not limited to these. For example, the colorectal cancer may be metastatic, or it may be in a state of HER2 overexpression or gene amplification, but is not limited to these.

[0107] In one embodiment, the solid tumor may be in a state of HER2 overexpression or amplification. In one embodiment, the step of determining whether the cancer tissue from which the tissue slide was taken is in a state of HER2 overexpression or amplification may further be included.

[0108] In this specification, the terms "overexpression" or "amplification" can refer to the phenomenon in which the number of gene replicas increases, leading to increased expression of genes related to the HER2 protein.

[0109] In one embodiment, the method for predicting the therapeutic response to the HER2-targeted therapeutic drug may further include the steps of deriving tumor microenvironment factors (TME factors) from pathological images and predicting the therapeutic response to the HER2-targeted therapeutic drug using the tumor microenvironment factors (TME factors).

[0110] As used herein, the term "tumor microenvironment" can refer to the environment consisting of tissues and cells surrounding a tumor. For example, it may include, but is not limited to, the tumor cells themselves, as well as surrounding blood vessels, extracellular matrix (surrounding tissue), immune cells, inflammatory mediators, and other related elements.

[0111] The term "tumor microenvironment factors (TME factors)" as used herein may mean various cells, signaling molecules, and other environmental factors that are expressed or act in the tumor microenvironment. For example, the tumor microenvironment factors (TME factors) may be one or more selected from the group consisting of lymphocyte (LC) density, macrophage (MP) density, fibroblast (FB) density, and endothelial cell (EC) density.

[0112] In one embodiment, the tumor microenvironment factors (TME factors) may be measured in the cancer area (CA) or the cancer stroma area (CS).

[0113] In this specification, the term "cancer area (CA)" can refer to an area where tumor cells exhibiting invasion are clustered.

[0114] As used herein, the term "cancre stroma area (CS)" can refer to the area surrounding a tumor where tumor-related matrix changes are observed, such as the formation of fibrous tissue (desmoplasia) or the aggregation of lymphoid cells. For example, the cancre stroma area (CS) may include, but is not limited to, fibrous tissue, blood vessels, immune cells, inflammatory cells, etc.

[0115] In one embodiment, the step of predicting the therapeutic response to the HER2-targeted therapy drug may include determining that the cancer patient will respond to the HER2-targeted therapy drug if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is 50% or more, and if one or more tumor microenvironment factors (TME factors) measured in the cancer stroma area (CS) are less than the following values. - Lymphocyte (LC) density: 766.3 - Macrophage (MP) density: 26.4 - Fibroblast (FB) density: 1790.8 -Endothelial cell (EC) density: 88.3

[0116] As used herein, the term "good prognosis" can mean a high survival rate for cancer patients in response to treatment and follow-up after cancer diagnosis. Specifically, it can mean the absence or reduction of invasive cancer within the tissue due to anti-cancer treatment, and a low or no likelihood of tumor recurrence.

[0117] Another embodiment provides a method for treating cancer, comprising the steps of: using a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in a first pathological slide image of a patient; determining the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity based on the identified results; predicting the patient's response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity; and, if it is determined that the patient will respond to the HER2-targeted therapy based on the treatment response prediction results, administering the HER2-targeted therapy to the patient.

[0118] The invention will be explained in more detail by the following examples. However, these examples are for illustrative purposes only, and the scope of the present invention is not limited to these examples. [Examples]

[0119] Example 1. Cohort composition

[0120] The Prescreening cohort and the TRIUMPH cohort of this study were constructed through the HER2-Screening study and the GOZILA study.

[0121] In the HER2 screening study, HER2 IHC slides were stained using PATHWAY HER2 / neu(4B5) rabbit monoclonal primary antibody (Ventana Medical Systems, Tucson, AZ, US), and FISH analysis was performed using the PathVysion HER-2 DNA probe kit (Abbott Laboratories, IL, US).

[0122] The GOZILA study was conducted through patients with metastatic gastrointestinal cancer (mCRC) who showed disease progression during chemotherapy in a HER2-screening study. Plasma NGS analysis was performed using Guardant360 at Guardant Health in the GOZILA study.

[0123] The prescreening cohort for this study was formed by selecting cases from the aforementioned HER2-Screening study and the GOZILA study where both HER2 IHC slides and H&E slides were available.

[0124] Furthermore, the TRIUMPH cohort was comprised of patients identified as HER2-positive by tissue and / or ctDNA analysis in the HER2-Screening and GOZILA studies. The TRIUMPH trial was a multicenter phase 2 clinical trial conducted in Japan, and information on participants' place of origin, race, ethnicity, and socioeconomic status was not collected as part of this study. Detailed characteristics of the TRIUMPH cohort patients are shown in Table 1.

[0125] [Table 1]

[0126] The research protocol was approved through research ethics reviews at all participating institutions, including National Cancer Center Hospital East, Aichi Cancer Center Hospital, National Cancer Center Hospital, National Hospital Organization Kyushu Cancer Center, Hokkaido University Hospital, National Hospital Organization Shikoku Cancer Center, and National Hospital Organization Osaka National Hospital. The study was conducted in strict compliance with the protocol, the Ministerial Ordinance on Good Clinical Practice for Drugs, and the Declaration of Helsinki.

[0127] Example 2. Development of an AI-powered pathology slide image analyzer.

[0128] 2-1. Development of an AI-powered HER2 analyzer

[0129] According to one embodiment, an AI-based HER2 analyzer can be used to analyze the HER2 quantitative continuous score (QCS). The AI-based HER2 analyzer according to one embodiment is based on a DeepLabV3+ convolutional neural network architecture and can be developed using a ResNet-34 backbone network.

[0130] The aforementioned AI-powered HER2 analyzer can consist of two deep learning (DL) based AI models: a cell detection model and a tissue segmentation model.

[0131] Specifically, the cell detection model was obtained from 1,259 HER2 [Ventana anti-HER2 / neu(4B5) (Ventana Medical Systems, Tucson, AZ, US)] IHC-stained WSI cells from breast cancer cases, resulting in an 8.13 × 10⁶ cell count. 5 The model was trained using a dataset containing tumor cells (TCs). The cell detection model can classify cells into five categories: HER2 IHC 0 (H0) tumor cells, HER2 IHC 1+ (H1) tumor cells, HER2 IHC 2+ (H2) tumor cells, HER2 IHC 3+ (H3) tumor cells, and other cells (OT). H0 represents tumor cells (TCs) with no cell membrane staining, H1 represents tumor cells (TCs) with partial cell membrane staining at a faint or barely noticeable intensity, H2 represents tumor cells (TCs) with partial lateral or periphery cell membrane staining at a weak / moderate intensity, and H3 represents tumor cells (TCs) with complete lateral or periphery cell membrane staining at a strong intensity.

[0132] The aforementioned tissue segmentation model was derived from 1,214 PD-L1 22C3 [PD-L1 22C3 pharmDx IHC (Agilent TecHnologies Inc., Santa Clara, CA, US)] IHC-stained WSI (including 197 cases of colorectal cancer (CRC)) samples, yielding 6.42 × 10⁶ samples. 2 mm 2The model was trained using a dataset containing cancer area (CA), and then trained using various datasets consisting of gallbladder cancer, bladder cancer, breast cancer, colorectal cancer, esophageal cancer, liver cancer, lung cancer, pancreatic cancer, prostate cancer, and gastric cancer.

[0133] The tissue segmentation model classifies the WSI into cancer area (CA) and background (BG: the portion excluding CA). The performance of the tissue segmentation model was evaluated using the Intersection Over Union (IoU) metric. The evaluation was performed using a tuning dataset and an internal test dataset, and the tissue segmentation model achieved IoU values ​​of 65.16 and 63.48 for CA and BG, respectively. In the case of the tissue detection model, the entire WSI was analyzed pixel by pixel, and pixels corresponding to tumor areas were displayed.

[0134] By integrating information from the cell detection model and the tissue segmentation model, only tumor cells (TCs) located within the cancer area (CA) were considered for HER2 evaluation. The proportion of tumor cells (TCs) exhibiting each staining intensity within the cancer area (CA) was calculated using Equation 1 below.

[0135]

number

[0136] Using the aforementioned AI-powered HER2 analyzer, HER2 status was classified according to internationally agreed diagnostic criteria as follows: HER2 IHC 3+ positive (AI-H3 prop > 10%), HER2 IHC 2+ equivocal (AI-H3 prop < 10% or AI-H2 prop > 10%), HER2 IHC 1+ negative (AI-H1 prop > 10%), and HER2 IHC 0 negative (AI-H1 prop ≤ 10% or presence of only H0 cells).

[0137] 2-2. Development of an AI-powered tumor microenvironment analyzer

[0138] According to one embodiment, an AI-based tumor microenvironment analyzer can be used for tumor microenvironment (TME) analysis.

[0139] An AI-based tumor microenvironment analyzer according to one embodiment may be based on the DeepLabV3+ architecture. The AI-based TME analyzer may consist of two deep learning (DL)-based AI models: a cell detection model and a tissue segmentation model.

[0140] The aforementioned AI-based tumor microenvironment (TME) analyzer measured 17,524 H&E-stained WSIs (tissue area 14.6 × 10⁶) annotated by pathologists. 9 μm 2 TCs2.2×10 6The model was trained using the dataset. Furthermore, to further improve its capabilities, additional data on tumor cells (TCs), lymphocytes (LCs), macrophages (MPs), fibroblasts (FBs), endothelial cells (ECs), and other cells (OTs) from various cancers, including colorectal cancer (CRC), were updated.

[0141] Specifically, the tissue segmentation model divided the WSI region into cancer area (CA), cancer stroma area (CS), and background (BG). Information from the cell detection model was then integrated to calculate the cell density of each region for each cell type.

[0142] The tumor cell (TC) density in the cancerous region (CA) and the cancerous stromal region (CS) was defined as TC-CA and TC-CS, respectively (cell counts / mm³). 2 Lymphocytes, macrophages, fibroblasts, and endothelial cells were also defined using the same method (LC-CA, LC-CS, MP-CA, MP-CS, FB-CA, FB-CS, EC-CA, and EC-CS).

[0143] Example 3: HER2 status analysis of colorectal cancer patients using AI-based HER2 QCS analysis

[0144] 3-1. AI-based HER2 QCS analysis

[0145] To analyze the HER2 status of colorectal cancer patients, we used two AI-based HER2 analyzers to determine the quantitative proportion of tumor cells (TCs) with each HER2 staining intensity (HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), HER2 IHC 3+ (H3)) (AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop).

[0146] Specifically, the HER2 IHC score (Negative, 1+, 2+, or 3+) of pathology slides from colorectal cancer patients was determined using the AI-powered HER2 analyzer of Example 2-1 in the Prescreening cohort (N=144) and the TRIUMPH cohort (N=30), according to internationally agreed diagnostic criteria.

[0147] The proportions of HER2 amplification were confirmed in the TRIUMPH study by HER2 immunohistochemistry (IHC) and FISH (Fluorescence In Situ Hybridization) or circulating tumor DNA (ctDNA) analysis.

[0148] On the other hand, for each case in the Prescreening cohort (N=144) and the TRIUMPH cohort (N=30), tumor cells (TCs) with HER2 staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and IHC 3+ (H3) were separated using AI-powered HER2 QCS analysis, and the proportion of tumor cells with each of these HER2 staining intensities (AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop) was derived.

[0149] Specifically, using an AI-based HER2 analyzer, tumor cells were classified into tumor cells and other cells (OT) based on their respective HER2 staining intensities: HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3). Tumor cells without cell membrane staining were classified as tumor cells with the aforementioned HER2 IHC 0 (H0) staining intensity, while tumor cells with faint or almost imperceptible partial cell membrane staining were classified as tumor cells with the HER2 IHC 1+ (H1) staining intensity. Tumor cells with weak / moderate partial lateral or periphery cell membrane staining were classified as tumor cells with the HER2 IHC 2+ (H2) staining intensity, and tumor cells with strong, complete staining of the side or periphery of the cell membrane were classified as tumor cells with the HER2 IHC 3+ (H3) staining intensity. Based on the classification results, the proportion of tumor cells with each HER2 staining intensity (AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop) in the pathology slide images was derived. The results of analyzing the HER2 status in the Prescreening cohort (N=144) and the TRIUMPH cohort using an AI-based HER2 analyzer are shown in Figure 5.

[0150] Figure 5 shows the results of confirming the HER2 status of colorectal cancer patients using the HER2 IHC score and HER2 gene amplification rate (Proportions of HER2 amplification) from the patients' pathology slides.

[0151] Figure 5A shows the results of evaluating HER2 IHC scores in the Prescreening cohort (N=144) and the TRIUMPH cohort (n=30) using an AI-powered HER2 QCS analyzer. Figure 5B shows the results of analyzing HER2 amplification rates in the TRIUMPH study via HER2 IHC (immunohistochemistry) and FISH (fluorescence in situ hybridization) results or ctDNA analysis (circulating tumor DNA analysis) results. HER2 overexpression or amplification is indicated as HER2-positive (HER2+).

[0152] As shown in Figure 5A, in the Prescreening cohort (N=144), HER2 IHC scores were analyzed as follows: 86 cases (59.7%) were HER2 IHC negative, 7 cases (4.9%) were HER2 IHC 1+, 10 cases (6.9%) were HER2 IHC 2+, and 41 cases (28.5%) were HER2 IHC 3+. In the TRIUMPH cohort (n=30), HER2 IHC scores were analyzed as follows: 3 cases (10%) were HER2 IHC negative, 2 cases (6.7%) were HER2 IHC 1+, 2 cases (6.7%) were HER2 IHC 2+, and 23 cases (76.7%) were HER2 IHC 3+.

[0153] As shown in Figure 5B, analysis of HER2 IHC scores and FISH results in the TRIUMPH cohort revealed amplification in 27 cases (90%), with only 3 cases (10%) being negative (HER2). Similarly, ctDNA analysis showed HER2+ (positive) in 25 cases (83.3%), HER2- (negative) in 4 cases (13.3%), and 1 case (3.3%) was not tested, showing a similar trend to that observed in the HER2 IHC scores and FISH analysis results.

[0154] The results shown in Figures 5A and 5B confirm that the HER2 IHC score assessment in the TRIUMPH cohort (n=30) using the AI-based HER2 QCS analyzer showed a similar trend to the HER2 IHC score and FISH analysis results in the TRIUMPH cohort, which consisted of patients confirmed to be HER2-positive through tissue and / or ctDNA analysis in the HER2-Screening and GOZILA studies. This means that the results analyzed using the AI-based HER2 QCS analyzer are consistent with the actual HER2 status of patients.

[0155] 3-2. HER2 status analysis of colorectal cancer patients

[0156] The HER2 status of colorectal cancer patients enrolled in the TRIUMPH trial was examined based on HER2 IHC, FISH, and ctDNA analysis results, and compared with the AI-based HER2 QCS analysis results from Example 3-1.

[0157] Figure 6 is a graph analyzing the HER2 status of patients enrolled in the TRIUMPH trial using HER2 IHC, FISH, ctDNA analysis results, and the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop, %) determined by the AI ​​model (the x-axis shows the HER2 IHC (immunohistochemistry) score evaluated by pathologists, and the y-axis shows the HER2 / CEP17 ratio determined by FISH (fluorescence in situ hybridization)).

[0158] Figure 6A shows the "HER2+ (positive), HER2- (negative), or unexamined case" determined by HER2 ctDNA analysis, indicated by colored dots, while Figure 6B shows the "percentage of tumor cells showing HER2 IHC 3+ (H3) staining intensity (AI-H3 prop, %)" determined by the AI ​​model, indicated by colored dots.

[0159] As shown in Figure 6, the majority of patients analyzed as HER2-positive (HER2+) through HER2 IHC, FISH, and ctDNA analysis in the TRIUMPH study were found to have a percentage of tumor cells with HER2 IHC 3+(H3) staining intensity (AI-H3 prop) of 50% or more, as determined by AI-based HER2 QCS analysis.

[0160] On the other hand, the proportion of tumor cells with HER2 IHC 3+(H3) staining intensity (AI-H3 prop) was evaluated in the Prescreening cohort and the TRIUMPH cohort through AI-based HER2 QCS analysis, and the AI-H3 prop is shown for each HER2 IHC score in each pathology slide.

[0161] Figure 7 is a graph showing the results of evaluating the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop) in pathology slides (WSI) from the Prescreening cohort and the TRIUMPH cohort using an AI-powered HER2 QCS analyzer.

[0162] As shown in Figure 7, the median AI-H3 prop levels in the prescreening cohort for HER2 IHC-negative, HER2 IHC 1+, and HER2 IHC 2+ cases were 0%, with ranges of 0.00–0.39% (minimum–maximum), 0.00–4.87%, and 0.00–9.43%, respectively. On the other hand, a high median of 75.8% (minimum–maximum 11.9–99.0%) was observed in HER2 IHC 3+ cases. In the TRIUMPH cohort, we confirmed that a similar pattern to the prescreening cohort was observed, with median AI-H3 prop percentages for HER2 IHC-negative, HER2 IHC 1+, HER2 IHC 2+, and HER2 IHC 3+ cases being 0% (minimum-maximum 0.00-0.0003%), 0% (0.00-0.00%), 0.005% (0.00-0.009%), and 73.9% (11.9-99.0%), respectively.

[0163] On the other hand, we confirmed that the percentage of tumor cells with HER2 IHC 3+ (H3) staining intensity (AI-H3 prop) in HER2 IHC 3+ cases showed considerable variability. Specifically, 18.1% had low AI-H3 prop, while 97.4% had high AI-H3 prop. Cases corresponding to the ranges of low and high AI-H3 prop assessed by the AI-based HER2 QCS analyzer were confirmed by representative clinical cases from the TRIUMPH study.

[0164] Figure 8 shows photographs of representative clinical cases from the TRIUMPH cohort with low (18.1%) and high (97.4%) AI-H3 prop in pathology slides with HER2 IHC 3+ scores.

[0165] Figure 8A is a photograph showing a representative clinical case with low (18.1%) AI-H3 prop in a pathology slide of a HER2 IHC 3+ score in the TRIUMPH cohort, and Figure 8B is a photograph showing a representative clinical case with high (97.4%) AI-H3 prop in a pathology slide of a HER2 IHC 3+ score in the TRIUMPH cohort.

[0166] Figures 8A and 8B show clinical cases classified as having a HER2 IHC 3+ score by an AI-based HER2 QCS analyzer. The left image is the original WSI, and the right image is the WSI processed by the AI ​​analyzer. (Tumor cells (TCs) are color-coded according to HER2 staining intensity (blue: HER2-negative (HER2 IHC 0(H0)) TC, green: HER2 1+ (HER2 IHC 1+(H1)) TC, yellow: HER2 2+ (HER2 IHC 2+(H2)) TC, red: HER2 IHC 3+(H3)) TC), with green shaded areas indicating cancerous areas.)

[0167] As shown in Figures 7 and 8, we have confirmed that the percentage of tumor cells with HER2 IHC 3+ (H3) staining intensity (AI-H3 propr) in pathology slides with a HER2 IHC 3+ score exhibits considerable variability, and these results have been confirmed in actual clinical outcomes. This means that even when a patient's pathology slide is determined to have a HER2 IHC 3+ score, the percentage of tumor cells with HER2 IHC 3+ (H3) staining intensity within the said pathology slide is distributed in a varied manner.

[0168] More specifically, Figure 9 shows the results of evaluating the proportion of tumor cells (TCs) with staining intensities other than HER2 IHC 3+(H3) (HER2 IHC 0(H0), HER2 IHC 1+(H1), HER2 IHC 2+(H2)) in the Prescreening cohort and the TRIUMPH cohort, as classified by an AI-powered HER2 QCS analyzer.

[0169] Figure 9 is a graph showing the percentage of tumor cells (TCs) with staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) in slides classified by HER2 IHC score using an AI-powered HER2 QCS analyzer in the Prescreening cohort and the TRIUMPH cohort. In Figure 9, the x-axis represents the HER2 IHC score of the pathology slides determined by an AI-based HER2 analyzer, and the percentage of tumor cells exhibiting each HER2 staining intensity (HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), or HER2 IHC 3+ (H3)) within each group with each HER2 IHC score (negative, 1+, 2+, 3+ are indicated by AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop) is shown in color-coded box plots.

[0170] As shown in Figure 9, the proportion of tumor cells with HER2 IHC 3+(H3) staining intensity (AI-H3 prop) was clearly higher in the pathology slides with the HER2 IHC 3+ score.

[0171] Example 4. Confirmation of agreement between AI-based HER2 analyzer and HER2 IHC score evaluated by pathologists.

[0172] To confirm the agreement between the HER2 IHC scores of the AI-based HER2 analyzer and the HER2 IHC scores of pathology slide images obtained from patient tissue, as evaluated by pathologists, the degree of agreement was calculated.

[0173] Specifically, the x-axis showed the HER2 IHC scores of the pathology slides evaluated by pathologists, and the y-axis showed the HER2 IHC scores of the pathology slides determined by an AI-based HER2 analyzer. The degree of agreement was then calculated. The agreement was calculated horizontally and shown with a blue gradient, and the results are shown in Figure 10.

[0174] Figure 10 shows the concordance rate between HER2 immunohistochemistry (IHC) scores from an AI-based HER2 QCS analyzer and pathology slides evaluated by pathologists in patients from the TRIUMPH cohort.

[0175] As shown in Figure 10, the agreement between the HER2 IHC scores of pathology slides determined by the AI-based HER2 QCS analyzer and the HER2 IHC scores of pathology slides evaluated by pathologists was confirmed to be 86.7% (95% confidence interval [95% CI] 69.3–96.2%, Cohen's kappa coefficient 0.663). In particular, the accuracy was 100% for pathology slides with a HER2 IHC 3+ score.

[0176] Example 5. Clinical outcome analysis of HER2-targeted therapeutics using AI-powered HER2 QCS analysis.

[0177] 5-1. Clinical outcome analysis based on tumor response rate

[0178] We analyzed the clinical outcomes of patients in the TRIUMPH cohort (N=30) who received pertuzumab and trastuzumab (dual HER2-targeted therapy). Specifically, we examined the objective response rate (ORR) to confirm the efficacy of pertuzumab and trastuzumab administration.

[0179] The tumor response rate was evaluated based on the Response Evaluation Criteria in Solid Tumors (RECIST version 1.1). While the response for each case was determined through both on-site investigators and an independent central review, this study utilized only the evaluations performed by on-site investigators.

[0180] Tumor response rates were compared to determine if there were differences in objective response rates (ORR) among the following subgroups. In this embodiment, the ORR refers to the percentage of patients who achieved a complete response (CR) or partial response (PR) confirmed by follow-up examinations at least four weeks after the initial response.

[0181] [Table 2]

[0182] The clinical results (CR, complete response; PR, partial response; SD, stable; PD, progression) for analyzing the objective response rate (ORR) were confirmed through the best overall response (BOR) for each subgroup shown in Table 2, and the results are shown in Figure 11.

[0183] Figure 11 is a graph showing the clinical outcomes for the total HER2+, Path HER2+, ctDNA HER2+, Path HER2 IHC 3+, AI-H3 prop ≥ 10%, and AI-H3 prop ≥ 50% groups.

[0184] As shown in Figure 11, the overall response rate (ORR) for TRIUMPH cohort patients (total HER2+) treated with pertuzumab and trastuzumab was 26.7% (8 / 30). This is consistent with the ORRs of 29.6% (8 / 27) and 28.0% (7 / 25), respectively, observed when pathologists assessed HER2+ (path HER2+) by HER2 IHC, FISH, and ctDNA.

[0185] Furthermore, the agreement between the group classified as HER2 IHC 3+ by pathologists and the group determined to have AI-H3 prop ≥ 10% was 100%, and both groups showed the same ORR (Overall Assessment Rate) of 34.7% (8 / 23) out of 23 cases. In particular, it was confirmed that setting the cutoff to AI-H3 prop ≥ 50% increased the ORR to 42.1% (8 / 19).

[0186] 5-2. Clinical outcome analysis using tumor regression percentages

[0187] We analyzed the clinical outcomes of patients in the TRIUMPH cohort (N=30) who received pertuzumab and trastuzumab (dual HER2-targeted therapy).

[0188] Specifically, tumor regression percentages were examined to confirm the effects of pertuzumab and trastuzumab administration, and the results are shown in Figure 12.

[0189] Figure 12 is a waterfall plot showing the tumor regression percentages from baseline in TRIUMPH trial patients treated with pertuzumab and trastuzumab.

[0190] As shown in Figure 12, among the cases showing a decrease in tumor size from baseline, 77.3% (17 / 22) had AI-H3 prop ≥ 50%, 9.1% (2 / 22) had AI-H3 prop in the range of 10-49%, and 13.6% (3 / 22) had AI-H3 prop < 10%.

[0191] This means that among HER2-positive (overexpression or amplification) patients treated with pertuzumab and trastuzumab, those with AI-H3 prop ≥ 10% are likely to have a good prognosis with pertuzumab and trastuzumab (86.4%), and in particular, those with AI-H3 prop ≥ 50% are highly likely to have a good prognosis with pertuzumab and trastuzumab (a high rate of tumor size reduction from baseline) (77.3%).

[0192] On the other hand, the correlation between the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop), measured by an AI-based HER2 analyzer, and tumor regression changes in TRIUMPH trial patients treated with pertuzumab and trastuzumab was analyzed using Spearman's correlation analysis, and the results are shown in Figure 13.

[0193] Figure 13 is a graph showing the correlation between the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity, as measured by an AI-based HER2 analyzer, and tumor regression changes in TRIUMPH trial patients treated with pertuzumab and trastuzumab. In Figure 13, the y-axis represents tumor regression percentages relative to baseline in TRIUMPH patients treated with pertuzumab and trastuzumab, and the x-axis represents AI-H3 prop (%). The points in each color indicate the combined results of the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity (AI-H3 prop, %) as measured by an AI-based HER2 analyzer and the HER2 IHC score (HER2 Path) as assessed by pathologists.

[0194] As shown in Figure 13, Spearman's correlation analysis revealed a Spearman's rho (ρ) of -0.32 (p=0.089), confirming a borderline negative correlation between tumor regression changes and AI-H3 prop.

[0195] This means that if the AI-H3 prop level is 50% or higher, there is a high probability of good drug sensitivity and treatment prognosis to pertuzumab and trastuzumab.

[0196] Furthermore, using an AI-based HER2 analyzer, we classified TRIUMPH cohort cases by HER2 IHC score for all slide images (WSI). For each case, we measured the percentage of tumor cells showing HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) staining intensity levels (denoted as AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop, respectively), and the results are shown in Figure 14.

[0197] Figure 14 is a graph showing the percentage of tumor cells with each HER2 staining intensity (HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), or HER2 IHC 3+ (H3)) in the TRIUMPH study (AI-H0 prop, AI-H1 prop, AI-H2 prop, or AI-H3 prop).

[0198] Figure 14A is a graph showing the distribution of tumor cells with HER2 staining intensity of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), or HER2 IHC 3+ (H3) (AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop) in the TRIUMPH cohort slides, categorized by HER2 IHC score. Figure 14B is a graph showing the distribution of tumor cells with HER2 staining intensity of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) (AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop) for each TRIUMPH cohort case.

[0199] As shown in Figure 14, the proportion of tumor cells with HER2 staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) differed depending on the HER2 IHC score of the TRIUMPH cohort slides. It was also confirmed that the proportion of tumor cells with HER2 staining intensities of HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), and HER2 IHC 3+ (H3) (AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop) differed among different TRIUMPH cohort cases.

[0200] The results shown in Figures 12 to 14 confirm that AI-H3 prop differs among patients, and that a higher percentage of tumor cells (AI-H3 prop) with HER2 IHC 3+ (H3)HER2 staining intensity were associated with patients whose HER2 IHC score on their slides was HER2 IHC 3+.

[0201] 5-3. Clinical outcome analysis by progression-free survival (PFS) and overall survival (OS)

[0202] We analyzed the clinical outcomes of patients in the TRIUMPH cohort (N=30) who received pertuzumab and trastuzumab (dual HER2-targeted therapy).

[0203] Specifically, the results of AI-powered HER2 QCS analysis were compared with progression-free survival (PFS) and overall survival (OS) based on the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop < 50% and AI-H3 prop ≥ 50%) using Kaplan-Meier curves and a Cox proportional hazards model. The results are shown in Figure 15.

[0204] Figure 15 is a graph showing the overall response rate (ORR) and Kaplan-Meier curves for progression-free survival (PFS) and overall survival (OS) depending on whether the AI-H3 prop content is less than 50% (<50%) or greater than 50% (≧50%). As shown in Figure 15, progression-free survival (PFS) was 1.4 months (95% confidence interval, 1.3-not reached) in the group where the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity, as determined by AI-powered HER2 QCS analysis, was less than 50% (AI-H3 prop < 50%), and was extended to 4.4 months in the group where AI-H3 prop was ≥ 50% (AI-H3 prop ≥ 50%) (95% confidence interval, 4.0-12.0), with a hazard ratio (HR) of 0.12 (95% confidence interval, 0.04-0.38; p < 0.001). This means that the group with AI-H3 prop ≥ 50% (AI-H3 prop ≥ 50%) showed significantly better progression-free survival (PFS). Similarly, overall survival (OS) was 4.1 months (95% confidence interval, 2.4-not achieved) in the group with less than 50% AI-H3 prop (AI-H3 prop < 50%), and increased to 16.5 months (95% confidence interval, 11.6-not achieved) in the group with 50% or more AI-H3 prop (AI-H3 prop ≥ 50%). This means that the group with 50% or more AI-H3 prop (AI-H3 prop ≥ 50%) showed significantly better overall survival (OS).

[0205] Therefore, we confirmed that in the group where the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity determined by AI-powered HER2 QCS analysis was ≥50% (AI-H3 prop ≥50%), there was a high probability of better treatment response, drug sensitivity, and treatment prognosis to pertuzumab and trastuzumab.

[0206] Example 6. Clinical outcome analysis of HER2-targeted therapy using AI-powered tumor microenvironment analysis (TME analysis)

[0207] 6-1. AI-based analysis of tumor microenvironment factors in responders and non-responders

[0208] To understand the trends in tumor microenvironment factors (TME factors), responders (patients who achieved complete or partial response) and non-responders (patients who did not achieve complete or partial response) were compared across the entire TRIUMPH cohort. The TME factors analyzed were lymphocyte density (LC-CA, LC-CS), fibroblast density (FB-CA and FB-CS), macrophage density (MP-CA, MP-CS), and endothelial cell density (EC-CA and EC-CS) analyzed within the cancer area (CA) and cancer stroma area (CS). The p-values ​​for each factor comparison are shown at the top of the box plots and were calculated using Student's t-test or Wilcoxon test.

[0209] Specifically, the best overall response (BOR) was classified into complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD) according to the RECIST V1.1 criteria. Patients who achieved a complete response (CR) or partial response (PR) were classified as responders, while patients who achieved stable disease (SD) or progressive disease (PR) were classified as non-responders. A comparative analysis of tumor microenvironment factors (TME factors) was performed on responders and non-responders classified according to the above criteria, and the results are shown in Figure 16.

[0210] Figure 16 is a graph showing a comparative analysis of AI-analyzed tumor microenvironment factors (lymphocyte (LC) density, macrophage (MP) density, fibroblast (FB) density, and endothelial cell (EC) density) between responders and non-responders in TRIUMPH patients.

[0211] As shown in Figure 16, respondents found a tendency for increased lymphocyte (LC), macrophage (MP), fibroblast (FB), and endothelial cell (EC) densities within the cancer area (CA) (referred to as LC-CA, MP-CA, FB-CA, and EC-CA, respectively).

[0212] On the other hand, LC, MP, and FB densities in the cancer stroma area (CS) were generally lower in responders (referred to as LC-CS, MP-CS, and FB-CS, respectively), but these results were not statistically significant. However, EC density in the cancer stroma area (CS) (EC-CS) was found to be considerably lower in responders compared to non-responders, which is a significant result (median values ​​were 56.8 [95% confidence interval, 21.0-81.8] vs. 89.3 [95% confidence interval, 58.4-162.9], p=0.041, respectively).

[0213] This means that EC-CS can be used to select patients who are likely to show high drug sensitivity and a good treatment outcome to pertuzumab and trastuzumab.

[0214] 6-2. AI-based tumor microenvironment factor analysis using AI-H3 prop

[0215] We compared tumor microenvironment factors (TME factors) analyzed by AI in two groups: one with less than 50% AI-H3 prop (AI-H3 prop < 50%) and another with 50% or more AI-H3 prop (AI-H3 prop ≥ 50%), as determined by an AI-powered HER2 QCS analyzer.

[0216] Specifically, the p-values ​​for the comparison of each factor are shown at the top of the box plot, and these were calculated using Student's t-test or Wilcoxon test. Lymphocyte density (LC-CA, LC-CS), fibroblast density (FB-CA and FB-CS), macrophage density (MP-CA, MP-CS), and endothelial cell density (EC-CA and EC-CS) were analyzed within the cancerous region (CA) and the cancerous stromal region (CS), and these results are shown in Figure 17.

[0217] Figure 17 is a graph comparing tumor microenvironment factors (TME factors) in the TRIUMPH cohort between the group with less than 50% AI-H3 prop (AI-H3 prop < 50%) and the group with 50% or more AI-H3 prop (AI-H3 prop ≥ 50%).

[0218] As shown in Figure 17, we confirmed that there were no significant differences in any of the tumor microenvironment factors (TME factors) between the group with less than 50% AI-H3 prop (AI-H3 prop < 50%) and the group with 50% or more AI-H3 prop (AI-H3 prop ≥ 50%).

[0219] This means that it may be difficult to select patients who can demonstrate high drug sensitivity and a good treatment outcome to pertuzumab and trastuzumab using only each TME factor.

[0220] 6-3. AI-based tumor microenvironment factor analysis in patients classified as responders and non-responders with AI-H3 prop levels of 50% or higher.

[0221] We analyzed the clinical outcomes of patients in the TRIUMPH cohort (N=30) who received pertuzumab and trastuzumab (dual HER2-targeted therapy).

[0222] Specifically, in the TRIUMPH cohort, tumor microenvironment factors (TME factors) quantified using an AI-based tumor microenvironment analyzer were compared between responders (patients who achieved complete response (CR) or partial response (PR) according to RECIST V1.1 criteria) and non-responders (patients who achieved stable disease (SD) or progressive disease (PD)).

[0223] The tumor microenvironment factors (TME factors) analyzed were lymphocyte density (LC-CA, LC-CS), fibroblast density (FB-CA, FB-CS), macrophage density (MP-CA, MP-CS), and endothelial cell density (EC-CA and EC-CS) in the cancerous region (CA) and cancerous stromal region (CS). The results of the analysis of TME factors between responders and non-responders are shown in Figure 18. The p-values ​​for the comparison of each TME factor are shown at the top of the box plot and were calculated using Student's t-test or Wilcoxon test.

[0224] Figure 18 is a graph comparing TME factors quantified by an AI-based TME analyzer in the TRIUMPH cohort group with AI-H3 prop ≥ 50% (AI-H3 prop ≥ 50%) between responders (patients who achieved complete response (CR) or partial response (PR) according to RECIST V1.1 criteria) and non-responders (patients who achieved stable disease (SD) or progression (PD)).

[0225] As shown in Figure 18, LC-CA, MP-CA, FB-CA, and EC-CA levels were similar in both the responder and non-responder groups, but the responder group showed a tendency towards lower levels of LC-CS, MP-CS, FB-CS, and EC-CS. However, this trend was statistically significant only for LC-CS and EC-CS.

[0226] Specifically, the median LC-CS between responders and non-responders were 297.1 (95% confidence interval, 235.6-797.9) and 1143.8 (95% confidence interval, 686.7-1473.1), respectively, with p=0.043. The median EC-CS was confirmed to be 56.8 (95% confidence interval, 21.0-81.8) and 143.9 (95% confidence interval, 87.9-159.6) between responders and non-responders, respectively.

[0227] This means that patients with AI-H3 prop ≥ 50% and low LC-CS and EC-CS factors are likely to show a high treatment response, drug sensitivity, and good treatment prognosis to pertuzumab and trastuzumab.

[0228] Example 7. Quantitative analysis of progression-free survival (PFS) or overall survival (OS) using AI-based HER2 IHC 3+(H3) staining intensity of tumor cells (AI-H3 prop) and AI-based tumor microenvironment factors (TME factors).

[0229] Using Kaplan-Meier curves, we analyzed the quantitative percentage of tumor cells exhibiting AI-based HER2 IHC 3+(H3) staining intensity (AI-H3 prop) and progression-free survival (PFS) or overall survival (OS) based on AI-based tumor microenvironment factors.

[0230] Specifically, 30 patients from the TRIUMPH cohort were divided into two groups: one with 50% or more tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop ≥ 50%) and another with less than 50% tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop < 50%). Progression-free survival (PFS) or overall survival (OS) was analyzed using Kaplan-Meier curves based on AI-analyzed tumor microenvironment factors (TME factors). Kaplan-Meier curves for groups combined with AI-H3 prop and one of the AI-analyzed tumor microenvironment factors (LC-CS, MP-CS, FB-CS, or EC-CS) are shown in Figures 19 and 20. In Figures 19 and 20, the cutoff for AI-based tumor microenvironment factors was determined as the median for the AI-H3 prop ≥ 50% group, as shown in the plot.

[0231] Figure 19 is a graph showing Kaplan-Meier curves representing the quantitative percentage of tumor cells exhibiting AI-based HER2 IHC 3+(H3) staining intensity (AI-H3 prop) and progression-free survival (PFS) based on AI-based tumor microenvironment factors.

[0232] Figure 19A shows the Kaplan-Meier curves for AI-H3 prop and lymphocyte density in the cancer stromal region (LC-CS), Figure 19B shows the Kaplan-Meier curves for AI-H3 prop and macrophage density in the cancer stromal region (MP-CS), Figure 19C shows the Kaplan-Meier curves for AI-H3 prop and fibroblast density in the cancer stromal region (FB-CS), and Figure 19D shows the Kaplan-Meier curves for AI-H3 prop and endothelial cell density in the cancer stromal region (EC-CS).

[0233] Figure 20 shows the quantitative percentage of tumor cells exhibiting AI-based HER2 IHC 3+(H3) staining intensity (AI-H3 prop) and overall survival (OS) based on AI-based tumor microenvironment factors, as shown by Kaplan-Meier curves.

[0234] Figure 20A shows the Kaplan-Meier curve for AI-H3 prop and lymphocyte density in the cancer stromal region (LC-CS), Figure 20B shows the Kaplan-Meier curve for AI-H3 prop and macrophage density in the cancer stromal region (MP-CS), Figure 20C shows the Kaplan-Meier curve for AI-H3 prop and fibroblast density in the cancer stromal region (FB-CS), and Figure 20D shows the Kaplan-Meier curve for AI-H3 prop and endothelial cell density in the cancer stromal region (EC-CS).

[0235] As shown in Figure 19, the median values ​​for LC-CS, MP-CS, FB-CS, and EC-CS were 766.3, 26.4, 1790.8, and 88.3, ​​respectively. When these medians were used as a baseline, patients with values ​​above the median tended to have worse PFS. However, this was not statistically significant except for FB-CS (HR 2.86, 95% CI 1.02-8.02, p=0.046). A similar pattern was observed for OS, as shown in Figure 20, and this was also not statistically significant.

[0236] Example 8. Clinical outcome analysis of HER2-targeted therapy with AI-based tumor microenvironment factors when the quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity is 50% or more (AI-H3 prop ≥ 50%).

[0237] 8-1. Hazard ratio (HR) analysis of tumor microenvironment factors based on each AI.

[0238] Based on the results of determining the median progression-free survival (mPFS), median overall survival (mOS), and 95% confidence intervals (95 CIs) using the Kaplan-Meier curves shown in Example 7, the hazard ratios (HR) and p-values ​​for the 95% confidence intervals were calculated for each AI-analyzed tumor microenvironment factor (TME factor) using Cox regression analysis, and the results are shown in Figures 21 and 22.

[0239] Figure 21 is a graph showing the hazard ratios (HR) for lymphocyte density (LC-CS), fibroblast density (FB-CS), macrophage density (MP-CS), and endothelial cell density (EC-CS) in the cancer stromal region (CS) for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%).

[0240] Figure 21A is a graph showing the hazard ratio (HR) for PFS in patients where the quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity is 50% or more (AI-H3 prop ≥ 50%), and Figure 21B is a graph showing the hazard ratio (HR) for OS in patients where the quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity is 50% or more (AI-H3 prop ≥ 50%).

[0241] As shown in Figure 21, in the patient group with AI-H3 prop ≥ 50% (N=19), a specific trend emerged when stratifying by the median value of tumor microenvironment factors (TME factors) determined by AI-based TME analysis. Specifically, the medians of LC-CS, MP-CS, FB-CS, and EC-CS were 766.3, 26.4, 1790.8, and 88.3, ​​respectively. When these medians were used as a cutoff, patients with values ​​above the median tended to have worse PFS. However, this trend was not statistically significant except for FB-CS (HR 2.86, 95% confidence interval, 1.02-8.02, p=0.046). Furthermore, a similar pattern was observed for OS, but this was not statistically significant.

[0242] Figure 22 is a graph showing the hazard ratios (HR) for lymphocyte density (LC-CA), fibroblast density (FB-CA), macrophage density (MP-CS), and endothelial cell density (EC-CS) in cancerous areas (CA) for patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%).

[0243] Figure 22A is a graph showing the hazard ratio (HR) for PFS in patients where the quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity is 50% or more (AI-H3 prop ≥ 50%), and Figure 22B is a graph showing the hazard ratio (HR) for OS in patients where the quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity is 50% or more (AI-H3 prop ≥ 50%).

[0244] As shown in Figure 22, in contrast to the hazard ratio (HR) measured in the cancer stromal region (CS), the patient group with values ​​above the median for LC-CA, MP-CA, FB-CA, and EC-CA tended to have a better PFS, although this was not statistically significant. However, for OS, the patient group with values ​​above the median tended to have a worse OS, although this was not statistically significant.

[0245] This means that patients with a quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity of 50% or more (AI-H3 prop ≥ 50%) and a median FB-CS value less than 1790.8 are likely to have good drug sensitivity and treatment prognosis to pertuzumab and trastuzumab.

[0246] 8-2. Progression-free survival (PFS) and overall survival (OS) analysis using AI-based combinations of tumor microenvironment factors.

[0247] In patients with HER2-amplified metastatic colorectal cancer (mCRC) who received pertuzumab and trastuzumab (dual HER2-targeted therapy), survival analysis was performed using a combination of the quantitative percentage of tumor cells showing AI-based HER2 IHC 3+(H3) staining intensity (AI-H3 prop) and two or three tumor microenvironment factors (TME factors) derived from TME analysis.

[0248] Specifically, survival outcomes were compared by dividing the patients into three groups: the "AI-H3 prop < 50% group," the "AI-H3 prop ≥ 50% group where all selected TME factors were above the median (TME-high)," and the "AI-H3 prop ≥ 50% group excluding those where all selected TME factors were above the median (TME-low)." Table 3 shows the results of the survival analysis for progression-free survival (PFS), and Table 4 shows the results of the survival analysis for overall survival (OS).

[0249] Table 3 shows the results of survival analysis against PFS, combining two or three tumor microenvironment factors (TME factors) derived from an AI-powered TME analyzer with the quantitative percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop) derived from AI-powered HER2 QCS analysis results. Kaplan-Meier curves were used to determine the median progression-free survival (mPFS) and 95% confidence interval (CI). Hazard ratios (HR) with 95% CI and p-values ​​were calculated using Cox regression analysis. The TME factor cutoff for the AI-H3 prop ≥ 50% group was set as the median.

[0250] Specifically, in the cancer stromal region (CS), the lymphocyte density (LC-CS), macrophage density (MP-CS), fibroblast density (FB-CS), and endothelial cell density (EC-CS) were 766.3, 26.4, 1790.8, and 88.3, ​​respectively. "N" indicates the number of patients in each group, "NR" indicates not reached, and "Ref." means the reference group.

[0251] [Table 3-1] [Table 3-2]

[0252] Table 4 shows the results of survival analysis against overall survival (OS) performed by combining two or three TME factors derived from an AI-powered TME analyzer with the quantitative percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity (AI-H3 prop) derived from AI-powered HER2 QCS analysis results.

[0253] Kaplan-Meier curves were used to determine the median overall survival (mOS) and 95% confidence interval (CI). Hazard ratios (HR) with 95% CI and p-values ​​were calculated using Cox regression analysis. The TME factor cutoff for the AI-H3 prop ≥ 50% group was set as the median. Specifically, in the cancer stromal region (CS), lymphocyte density (LC-CS), macrophage density (MP-CS), fibroblast density (FB-CS), and endothelial cell density (EC-CS) were 766.3, 26.4, 1790.8, and 88.3, ​​respectively. "N" indicates the number of patients in each group, "NR" indicates not reached, and "Ref." means the reference group.

[0254] [Table 4-1] [Table 4-2]

[0255] As shown in Tables 3 and 4, the "AI-H3 prop < 50% group" showed significantly worse PFS and OS compared to the "AI-H3 prop ≥ 50% group with TME-low." In particular, the highest HR for PFS was observed when "LC-CS and FB-CS were combined" or "LC-CS, FB-CS, and EC-CS were combined" (HR 32.27, 95% CI, 5.98-174.05, p<0.001). Furthermore, the highest HR for OS was observed when "LC-CS, MP-CS, and FB-CS were combined" (HR 51.97, 95% CI, 6.35-425.62, p<0.001).

[0256] Figure 23 shows the Kaplan-Meier curves for progression-free survival (PFS) and overall survival (OS) for the "LC-CS, MP-CS, and FB-CS combinations".

[0257] Figure 23 shows the Kaplan-Meier curves illustrating the progression-free survival (PFS) and overall survival (OS) rates for the "LC-CS, MP-CS, and FB-CS combinations."

[0258] The median progression-free survival (mPFS) was 1.4 months (95% CI, 1.3-not achieved) for the "AI-H3 prop < 50% group," 1.3 months (95% CI, 1.3-not achieved) for the "AI-H3 prop ≥ 50% group with TME-High," and 5.6 months (95% CI, 4.4-20.2) for the "AI-H3 prop ≥ 50% group with TME-low."

[0259] As shown in Figure 23, the hazard ratio (HR) for PFS in the "TME-high group with AI-H3 prop ≥ 50%" was not significantly different from that of the AI-H3 prop < 50% group (HR 0.62, 95% CI, 0.19-2.01, p=0.428). However, the HR for PFS in the "TME-low group with AI-H3 prop ≥ 50%" was significantly better compared to the AI-H3 prop < 50% group (HR 0.04, 95% CI, 0.01-0.19, p<0.001).

[0260] The median overall survival (mOS) was 4.1 months (95% CI, 2.4-not achieved) for the "AI-H3 prop < 50% group," 4.5 months (95% CI, 2.2-not achieved) for the "AI-H3 prop ≥ 50% group with TME-High," and 26.0 months (95% CI, 16.5-not achieved) for the "AI-H3 prop ≥ 50% group with TME-low." Similar to the results of the HR analysis for PFS mentioned above, the HR for OS in the "TME-High group within AI-H3 prop ≥ 50%" was not significantly different from that of the "AI-H3 prop < 50% group" (HR 1.08, 95% CI, 0.36-3.21, p=0.894). However, the hazard ratio (HR) for OS in the "TME-low group with AI-H3 prop ≥ 50%" was significantly lower at HR 0.02 (95% CI, 0.002-0.16, p<0.001). In particular, all cases achieving CR or PR were included in the "group with TME-low in AI-H3 prop ≥ 50%", resulting in an ORR of 57.1% (8 / 14).

[0261] Figure 24 shows representative clinical cases of the group with TME-high or TME-low within the TRIUMPH cohort, where the proportion of tumor cells showing HER2 IHC 3+(H3) staining intensity was 50% or more (AI-H3 prop ≥ 50%).

[0262] Figure 24A shows a representative clinical case demonstrating the PD treatment response of the "AI-H3 prop≥50% group with TME-high", and Figure 24B shows a representative clinical case demonstrating the CR treatment response of the "AI-H3 prop≥50% group with TME-low".

[0263] The left panel shows the original whole slide image (WSI), and the right panel shows the WSI processed by the AI analyzer. Each cell in the tumor microenvironment (TME) is indicated by a colored dot (yellow: lymphocyte [LC], green: macrophage [MP], orange: fibroblast [FB]). Also, the green shaded area indicates the dark area (CA), and the blue shaded area indicates the cancer stromal area (CS).

[0264] As shown in Figure 24, it was confirmed that the "AI-H3 prop≥50% group with TME-high" had poor drug sensitivity and treatment prognosis (PD) to pertuzumab and Trastuzumab even in the actual clinical results, and the "AI-H3 prop≥50% group with TME-low" had a good treatment prognosis (CR) even in the actual clinical results.

[0265] 8-3. Analysis of Overall Survival (OS) at 12 and 24 Months by Combining AI-Based Tumor Microenvironment Factors

[0266] A nomogram was created to estimate the overall survival (OS) at 12 and 24 months of patients with HER2-amplified metastatic colorectal cancer (mCRC) who received pertuzumab and Trastuzumab administration (dual HER2-targeted therapy) and had a quantitative ratio (AI-H3 prop) of tumor cells showing AI-based HER2 IHC 3+ (H3) staining intensity of 50% or more (AI-H3 prop≥50%).

[0267] Specifically, for each patient with AI-H3 prop ≥ 50%, the TME factors were placed on the corresponding axis, the scores assigned to each factor were summed, and the total score was applied to the survival probability scale in the lower row. The TME factors integrated into the nomogram were lymphocyte density (LC-CS), macrophage density (MP-CS), fibroblast density (FB-CS), and endothelial cell density (EC-CS) in the cancer stroma (CS).

[0268] Figure 25 is a nomogram for estimating 12-month and 24-month overall survival (OS) in patients in whom the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is ≥50% (AI-H3 prop ≥50%).

[0269] As shown in Figure 25, the highest probability of survival at 12 and 24 months was achieved when the lymphocyte density (LC-CS), macrophage density (MP-CS), fibroblast density (FB-CS), and endothelial cell density (EC-CS) were 780, 50, 1500, and 100, respectively.

[0270] This is consistent with the results observed in Example 8-2, the "AI-H3 prop ≥ 50% group with TME-low," which showed a low hazard ratio (HR) and high progression-free survival (PFS) and overall survival (OS).

[0271] Example 9. Correlation analysis using AI-based HER2 QCS analyzer and AI-based TME analyzer.

[0272] To analyze the relationships between factors derived via AI-based HER2 QCS analyzers and AI-based TME analyzers, a total of 144 pairs of HER2 IHC WSIs and H&E (Hematoxylin and Eosin) WSIs were analyzed in a prescreening cohort.

[0273] Specifically, the factors analyzed by HER2 IHC WSI using an AI-based HER2 QCS analyzer may be the quantitative proportion of tumor cells (TCs) with staining intensities separated according to HER2 expression levels (AI-H0 prop, AI-H1 prop, AI-H2 prop, and AI-H3 prop). Factors derived via an AI-based TME analyzer may be the density of each cell type in the cancer stromal region (CS) and cancer region (CA) (LC-CA, MP-CA, FB-CA, EC-CA / LC-CS, MP-CS, FB-CS, EC-CS).

[0274] Figure 26 shows the results of the correlations between factors analyzed in HER2 IHC WSI and H&E WSI via the AI-based HER2 QCS analyzer and AI-based TME analyzer mentioned above.

[0275] Figure 26 is a graph showing the correlation between factors analyzed in HER2 IHC WSI and H&E WSI via AI-based HER2 QCS analyzer and AI-based TME analyzer.

[0276] As shown in Figure 26, among the factors analyzed by HER2 IHC WSI, AI-H3 prop showed a positive correlation with AI-H2 prop (Spearman's ρ=0.17, p<0.001), and negative correlations with AI-H1 prop (ρ=-0.12, p<0.001) and AI-H0 prop (ρ=-0.81, p<0.001), respectively.

[0277] On the other hand, positive correlations were confirmed between the densities of each cell type in the cancer stromal region (CS) and cancer region (CA) among the TME factors analyzed by H&E WSI. In particular, LC-CS showed a positive correlation with LC-CA (ρ=0.24, p<0.001), and EC-CS showed a positive correlation with EC-CA (ρ=0.45, p<0.001).

[0278] Between factors analyzed by HER2 IHC WSI and factors analyzed by H&E WSI, AI-H3 prop showed a weak negative correlation with LC-CA, EC-CA, and EC-CS (ρ=-0.08, -0.09, and -0.14, p=0.026, 0.001, and 0.017, respectively).

[0279] Example 10. Quantification and Statistical Analysis

[0280] The F1 score and IoU index were used to evaluate the performance of cell and tissue models, respectively. Agreement or Cohen's kappa value was used to assess the difference between pathologist interpretations and AI analyzer interpretations. The chi-squared test or Fisher's exact test was used for group comparisons of categorical variables. Student's t-test or nonparametric Mann-Whitney U test was used for group comparisons of continuations variables.

[0281] Correlation analysis was performed using either the Pearson correlation coefficient or Spearman's rho (ρ). Survival analysis involved creating Kaplan-Meier curves, and hazard ratios (HRs) were calculated using the Cox proportional hazards model. Two-sided p-values ​​were calculated, with a threshold of p < 0.05 indicating statistical significance. All statistical analyses were performed using Python (version 3.7) and R software (version 4.2.3) (R Foundation for Statistical Computing, Vienna, Austria).

Claims

1. A method for predicting the therapeutic response to HER2-targeted therapies, The process involves using a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in the patient's first pathology slide image, and Based on the identified results, the step is to determine the percentage of tumor cells that show HER2 IHC 3+(H3) staining intensity, A method comprising the step of predicting the patient's therapeutic response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting the HER2 IHC 3+(H3) staining intensity.

2. The method according to claim 1, wherein the HER2 staining intensity of the tumor cells includes HER2 IHC 0 (H0), HER2 IHC 1+ (H1), HER2 IHC 2+ (H2), or HER2 IHC 3+ (H3).

3. The method according to claim 1, wherein the HER2-targeted therapeutic agent comprises one or more drugs selected from the group consisting of pertuzumab, trastuzumab, trastuzumab emtansine, lapatinib, neratinib, afatinib, tucatinib, and pyrotinib.

4. The method according to claim 1, wherein the step of predicting the patient's therapeutic response to a HER2-targeted therapy is to determine that the patient will respond to the HER2-targeted therapy if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is 10% or more.

5. The method according to claim 4, further comprising the step of determining that the patient will respond to a HER2-targeted therapy if the percentage of tumor cells exhibiting the HER2 IHC 3+(H3) staining intensity is 50% or more.

6. The method according to claim 1, wherein the first pathological slide image is a tissue slide taken from solid tumor tissue.

7. The method according to claim 6, further comprising the step of determining whether the cancerous tissue from which the tissue slide was taken is in a state of HER2 overexpression or gene amplification.

8. The method according to claim 6, wherein the solid tumor is one or more selected from the group consisting of lung cancer, skin cancer, stomach cancer, gastrointestinal cancer, intestinal cancer, colorectal cancer, colon cancer, pancreatic cancer, liver cancer, thyroid cancer, uterine cancer, cervical cancer, ovarian cancer, testicular cancer, prostate cancer, breast cancer, and oral cancer.

9. The method further includes the step of deriving tumor microenvironment factors (TME factors) from the second pathological slide image of the aforementioned patient, The step of predicting the patient’s treatment response to HER2-targeted therapy is: The method according to claim 1, comprising the step of predicting the therapeutic response to a HER2-targeted therapeutic agent using the percentage of tumor cells exhibiting the HER2 IHC 3+(H3) staining intensity and the tumor microenvironment factors (TME factors).

10. The method according to claim 9, wherein the tumor microenvironment factors (TME factors) are measured in the cancer area (CA) or cancer stroma area (CS).

11. The method according to claim 9, wherein the tumor microenvironment factors (TME factors) are one or more selected from the group consisting of lymphocyte (LC) density, macrophage (MP) density, fibroblast (FB) density, and endothelial cell (EC) density.

12. The method according to claim 9, wherein the predictive step includes determining that the patient will respond to a HER2-targeted therapy if the proportion of tumor cells showing HER2 IHC 3+(H3) staining intensity is 50% or more, and one or more tumor microenvironment factors (TME factors) measured in the cancer stroma area (CS) are less than the following values. - Lymphocyte (LC) density: 766.3 - Macrophage (MP) density: 26.4 - Fibroblast (FB) density: 1790.8 - Endothelial cell (EC) density: 88.3

13. A computer program stored on a computer-readable recording medium for performing the method according to claim 1 on a computer.

14. As a computing system, At least one memory, It includes at least one processor connected to the memory and configured to execute at least one computer-readable program contained in the memory, The computing system is configured such that the at least one processor, by executing the at least one program, uses a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in a first pathology slide image of a patient, determines the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity based on the identified results, and predicts the patient's therapeutic response to a HER2-targeted therapy based on the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity.

15. The computing system according to claim 14, wherein the processor is further configured to derive tumor microenvironment factors (TME factors) from a second pathological slide image of the patient, and to predict the therapeutic response to a HER2-targeted drug using the percentage of tumor cells exhibiting the HER2 IHC 3+(H3) staining intensity and the tumor microenvironment factors (TME factors).

16. The computing system according to claim 14, wherein the processor is configured to determine that the cancer patient will respond to a HER2-targeted therapy if the percentage of tumor cells exhibiting HER2 IHC 3+(H3) staining intensity is 50% or more.

17. The computing system according to claim 14, wherein the processor is configured to determine that a cancer patient will respond to a HER2-targeted therapeutic drug if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is 50% or more, and one or more tumor microenvironment factors (TME factors) measured in the cancer stroma area (CS) are less than the following values. - Lymphocyte (LC) density: 766.3 - Macrophage (MP) density: 26.4 - Fibroblast (FB) density: 1790.8 - Endothelial cell (EC) density: 88.3

18. The process involves using a machine learning model to identify the HER2 staining intensity of tumor cells (TCs) in the patient's first pathology slide image, and Based on the identified results, the step is to determine the percentage of tumor cells that show HER2 IHC 3+(H3) staining intensity, A step of predicting the patient's therapeutic response to a HER2-targeted drug based on the percentage of tumor cells exhibiting the HER2 IHC 3+(H3) staining intensity, A method for treating cancer, comprising the step of administering a HER2-targeted therapy drug to a patient if it is determined, based on the treatment response prediction results, that the patient will respond to the HER2-targeted therapy drug.

19. The method according to claim 18, wherein the step of predicting the patient's therapeutic response to a HER2-targeted therapy drug includes determining that the cancer patient will respond to the HER2-targeted therapy drug if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is 10% or more.

20. The method according to claim 18, wherein the step of predicting the patient's therapeutic response to a HER2-targeted therapy drug is to determine that the cancer patient will respond to the HER2-targeted therapy drug if the percentage of tumor cells showing HER2 IHC 3+(H3) staining intensity is 50% or more, and one or more tumor microenvironment factors (TME factors) measured in the cancer stroma area (CS) are less than the following values. - Lymphocyte (LC) density: 766.3 - Macrophage (MP) density: 26.4 - Fibroblast (FB) density: 1790.8 - Endothelial cell (EC) density: 88.3

Images