Endometrial cancer nursing treatment response prediction method based on multi-dimensional pathological assessment

Abstract

The invention discloses an endometrial cancer nursing treatment response prediction method based on multi-dimensional pathology assessment, and relates to the technical field of intelligent diagnosis prediction, and the method comprises the steps: collecting and standardly managing patient information data, and constructing a dynamically updated and quality-controllable multi-modal clinical pathology database. Performing multi-scale processing on the image and immune features, extracting tumor structures and immune cell distribution, and constructing a structured immune feature database; and multi-modal data are fused, an intelligent evaluation model is established, nursing treatment response is predicted, and dynamic optimization of the model is realized through clinical feedback. According to the invention, a standardized and multi-modal clinical pathology database is constructed, image, immune and molecular features are fused, multi-scale spatial information is accurately extracted, and a high-precision intelligent prediction model is established. And by introducing a dynamic feedback mechanism, the model can be adaptively optimized, and the accuracy and clinical practicability of nursing treatment response prediction are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent diagnostic prediction, and specifically to a method for predicting the response of endometrial cancer conservation treatment based on multi-dimensional pathological evaluation. Background Art

[0002] In recent years, with the continuous development of medical imaging digitization, molecular typing technology, and artificial intelligence algorithms, the application of multi-modal pathological information in tumor-assisted diagnosis and treatment efficacy prediction has been gradually deepened. Endometrial cancer, as a malignant tumor with a relatively high incidence in the female reproductive system, its early identification and individualized treatment strategies have always been research hotspots. Traditional evaluation methods mainly rely on single clinical or histopathological indicators, and it is difficult to comprehensively reflect the tumor biological heterogeneity of patients. To improve the accuracy of efficacy prediction in conservation treatment (especially for young women), researchers have gradually tried to integrate multi-source information such as radiomics, immunohistochemistry, and molecular subtypes, promoting the formation of a fusion prediction model to provide a new path for realizing precision medicine.

[0003] Although some current studies have explored the application of multi-modal information in the diagnosis and treatment of endometrial cancer, there are still significant deficiencies in many aspects in the actual clinical prediction scenario. First, existing multi-modal fusion methods are mostly limited to simple information splicing, lacking a unified data standard and structured system, resulting in serious inconsistencies in the spatial resolution, time scale, and feature representation methods of different modal data, making it difficult to achieve effective linkage and collaborative modeling. Second, the processing of pathological image information mostly focuses on coarse-grained levels such as gland identification and cell counting, and fails to fully explore the potential relationship between the spatial distribution of immune cells and tumor tissue structure at different scales. Especially in the modeling of the immune microenvironment, there are problems such as incomplete expression and weak spatial structure analysis, making it difficult to support individualized treatment decisions. Third, as an important stratification variable, most current studies only incorporate molecular subtypes as label variables into the model, without conducting in-depth analysis in combination with the interaction patterns between them and image, clinical, and immune features. More critically, most existing prediction models are statically constructed and do not introduce a clinical feedback mechanism to optimize and adaptively update the prediction model in real time, restricting the long-term stability and generalization ability of the model. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: existing methods for predicting the response of endometrial cancer conservation treatment have problems such as lack of standardized and structured expression of multi-modal data, weak collaborative analysis ability of image features and immune features, ineffective integration of molecular subtype information with other modalities, and how to achieve a high-precision individualized response prediction model with dynamic update ability.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: A method for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation, including collecting the patient's clinical basic information, tissue pathological image data, immune microenvironment indicators, and molecular subtype information to construct a multi-modal clinicopathological database, standardizing the input and management of the multi-modal clinicopathological database information, and establishing a dynamic update and quality control mechanism; performing multi-scale analysis and quantification processing on the image data and immune characteristics in the multi-modal clinicopathological database, through an image preprocessing process, performing color correction, background removal, and image enhancement on the tissue section images, using multi-scale spatial segmentation, identifying the tumor area at the macroscopic scale, extracting glandular structures and cell-dense areas at the tissue scale, identifying nuclear morphology and cell density information at the cell scale, extracting the distribution characteristics, infiltration density, aggregation degree, and spatial interaction relationship of immune cells, expressing the spatial feature pattern of the immune microenvironment in a structured manner, and constructing an immune feature database; fusing the information of the multi-modal clinicopathological database, constructing a multi-modal intelligent evaluation model for predicting the response of endometrial cancer conservation therapy based on the immune feature database, by setting a multi-source data fusion mechanism, comprehensively analyzing the correlation between clinicopathological features, tissue image structures, immune infiltration status, and molecular subtype patterns, identifying the characteristic variables affecting the conservation therapy effect, constructing a treatment response prediction index, and dynamically optimizing and updating the parameters of the multi-modal intelligent evaluation model in combination with the treatment feedback data of real patients.

[0007] As a preferred embodiment of the method for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation according to the present invention, wherein: the image preprocessing process includes image analysis, standardizing the image color, unifying the resolution, and removing background noise.

[0008] As a preferred embodiment of the method for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation according to the present invention, wherein: the multi-scale spatial segmentation includes three levels: Identifying the overall tumor area at the macroscopic level; Identifying glandular and tissue structures at the tissue scale; Identifying cell-level morphological features and nuclear density at the cell scale.

[0009] As a preferred embodiment of the method for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation according to the present invention, wherein: the spatial feature pattern of the immune microenvironment includes the density value of immune cells, spatial distribution pattern, co-localization relationship of cell types, and infiltration difference between the edge and central regions.

[0010] As a preferred embodiment of the method for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation according to the present invention, wherein: the multi-modal intelligent evaluation model includes using feature contribution analysis, feature screening and combination methods to identify the associated variables among clinical information, image information, immune indexes and molecular subtype information.

[0011] As a preferred embodiment of the method for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation according to the present invention, wherein: the molecular subtype pattern is obtained by sequencing technology or immunohistochemistry method, including collecting molecular features, performing standardized format conversion and structured management on the molecular features, and matching and fusing the modal data at the patient level; The molecular features include POLE gene mutation status, mismatch repair protein status and p53 expression pattern.

[0012] As a preferred embodiment of the method for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation according to the present invention, wherein: the dynamic optimization and parameter update of the multi-modal intelligent evaluation model by combining real patient treatment feedback data includes performing prospective verification through an independent clinical validation cohort after constructing the multi-modal intelligent evaluation model, and continuously adjusting the structure and parameters of the multi-modal intelligent evaluation model according to the actual treatment effect feedback.

[0013] Another object of the present invention is to provide a system for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation, which can construct a multi-modal intelligent evaluation model for predicting the response of endometrial cancer conservation therapy based on an immune feature database, and solves the problem that the current method for predicting the response of endometrial cancer conservation therapy contains a high-precision individualized response prediction model that cannot achieve dynamic update ability.

[0014] As a preferred embodiment of the system for predicting the response of endometrial cancer conservation therapy based on multi-dimensional pathological evaluation according to the present invention, wherein: it includes a multi-modal pathological data standardization construction and feature refinement extraction module, an intelligent fusion prediction module for molecular subtype information and pathological features, and a multi-modal fusion intelligent prediction model construction and clinical verification optimization module; the multi-modal pathological data standardization construction and feature refinement extraction module is used to standardize the collection and structured processing of the clinical pathological information, tissue pathological images, immune microenvironment features and molecular subtype data of endometrial cancer patients to form a multi-modal clinical pathological database; the intelligent fusion prediction module for molecular subtype information and pathological features is used to determine the comprehensive role and contribution degree of pathological features and molecular subtypes in predicting the treatment response, and determine the prediction value of the fusion features for the patient's treatment response; the multi-modal fusion intelligent prediction model construction and clinical verification optimization module is used to construct a multi-modal intelligent evaluation model, and optimize it through the verification of an independent clinical cohort and actual feedback.

[0015] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for predicting the response of endometrial cancer conservation treatment based on multi-dimensional pathological evaluation.

[0016] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of a method for predicting the response of endometrial cancer conservation treatment based on multi-dimensional pathological evaluation.

[0017] Advantages of the present invention: The method for predicting the response of endometrial cancer conservation treatment based on multi-dimensional pathological evaluation provided by the present invention significantly improves the accuracy and clinical practicability of predicting the response of endometrial cancer conservation treatment by constructing a multi-modal data standardization processing and deep fusion mechanism. Through the standardized collection and structured management of clinical pathological information, tissue images, immunohistochemistry, and molecular subtype information, high-quality integration of multi-source heterogeneous data is achieved, providing a stable data foundation for subsequent modeling; combined with multi-scale image processing and spatial segmentation strategies, the structural features of tumors and their immune microenvironments at different scales are finely extracted, ensuring the spatial expression integrity of pathological information; an immune feature database is constructed using the spatial distribution, density, and aggregation patterns of immune cells, significantly enhancing the model's ability to model the "tumor-immune" relationship; an innovative non-linear fusion function and multi-modal attention mechanism are introduced to achieve complex interactive expression between pathological and molecular features, effectively capturing the high-order coupling relationship of treatment response influencing factors; finally, by constructing a fusion prediction index and an adaptive dynamic optimization mechanism, continuous learning and feedback iteration of the model under real clinical data are realized, improving the generalization performance and interpretability of the model. Overall, the present invention not only breaks through the technical bottlenecks of existing models in feature fusion, prediction dimension, and update mechanism, but also provides a quantifiable, traceable, and interpretable intelligent auxiliary tool for clinical implementation of individualized treatment decisions. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is the overall flowchart of a method for predicting the response of endometrial cancer conservation treatment based on multi-dimensional pathological evaluation provided by the first embodiment of the present invention. Detailed Embodiments

[0020] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0021] Example 1, referring to Figure 1 , which is an embodiment of the present invention, provides a method for predicting the response of endometrial cancer conservation treatment based on multi-dimensional pathological evaluation, including: S1: Collect the clinical basic information, histopathological image data, immune microenvironment indicators, and molecular subtype information of patients to construct a multi-modal clinical pathology database, standardize the input and management of the information in the multi-modal clinical pathology database, and establish a dynamic update and quality control mechanism.

[0022] Furthermore, collect the basic clinical pathology information related to patients, formulate unified data recording standards and electronic data forms, and standardize the input, management, and establish a unique patient ID identification system.

[0023] It should be noted that the preparation and digital scanning process of histopathological sections are uniformly standardized. A high-resolution digital pathology scanner is used to scan the pathological tissue sections of all enrolled patients to obtain high-quality digital pathology images, form a unified HE staining, standardize the image size and resolution, and based on image feature automatic detection technology, accurately identify pathological features such as lesion areas, stromal areas, glandular structures, and tumor cell nucleus morphology, forming a refined histopathological image feature database.

[0024] Furthermore, strictly in accordance with the immunohistochemical staining method and index collection standard, select and collect key immune marker indicators, including the expression levels of CD3, CD8, CD20, CD68, and PD-L1. Use a standardized automatic image analysis software system to extract information such as the spatial distribution pattern, infiltration degree, cell density, cell co-localization, and microenvironment type of immune cells, quantify and structurally express the immune microenvironment features, and incorporate them into a unified immune microenvironment feature database system.

[0025] It should be noted that according to the molecular subtype analysis, design the genomic sequencing and typing process of molecular subtypes. Use next-generation sequencing technology NGS or immunohistochemistry methods to obtain key molecular subtype information such as POLE mutation status, mismatch repair status, and p53 abnormality, establish a molecular subtype data format, and accurately match the data with the patient ID to achieve seamless docking of molecular subtype data and modal data.

[0026] Furthermore, through a unified unique patient ID, the above-mentioned clinicopathological data, histopathological image data, immune microenvironment data, and molecular subtype data are matched and integrated to establish a patient-centered multimodal pathology database platform, ensuring that the data structures of each modality are unified, interoperable, queryable, and traceable.

[0027] S2: Conduct multi-scale analysis and quantification processing on the image data and immune features in the multimodal clinicopathological database. Through the image preprocessing process, perform color correction, background removal, and image enhancement on tissue section images. Adopt multi-scale spatial segmentation to identify tumor regions at the macroscopic scale, extract glandular structures and cell-dense regions at the tissue scale, identify nuclear morphology and cell density information at the cellular scale, and extract the distribution characteristics, infiltration density, aggregation degree, and spatial interaction relationship of immune cells, expressing the spatial feature pattern of the immune microenvironment in a structured manner to construct an immune feature database.

[0028] Furthermore, adopt a unified standard pathological tissue preparation and digital scanning technology to obtain high-resolution pathological section images of endometrial cancer patients, and perform automated preprocessing on the original images: use a color standardization algorithm to unify the color patterns of digital pathological images obtained by different devices; automatically remove the image background area and impurity interference based on tissue characteristics to ensure the accuracy of subsequent immune microenvironment feature analysis; perform image enhancement and noise suppression processing on the preprocessed images to improve the image quality and the accuracy of feature extraction.

[0029] It should be noted that for the organizational structure features at different scales in pathological images, perform multi-scale spatial segmentation analysis: Automatically identify the overall tumor area, tumor stroma area, and normal tissue area in the low-scale (macroscopic) field of view; At the medium-scale (tissue level) field of view, further refine the specific structural features within the lesion area and stroma area, including glandular structures and cell-dense regions; At the high-scale (cellular level) field of view, accurately identify cell morphology, nuclear features, and cell density features to ensure the refinement and accuracy of subsequent immune cell quantitative analysis.

[0030] Furthermore, after completing tissue spatial segmentation, automatically extract the immune microenvironment features in each spatial region: Analyze the overall immune cell infiltration pattern of tumor tissues at the macroscopic scale and evaluate the overall density and distribution characteristics of immune cells; Automatically analyze the difference in immune cell density between the stroma area and tumor area at the tissue scale, and clarify the immune cell infiltration boundary and distribution pattern, including dense at the tumor edge and sparse in the center; Precisely analyze the spatial distribution and aggregation degree of specific immune cell types at the cellular scale, and automatically identify the spatial aggregation patterns and infiltration characteristics of different immune cell types.

[0031] It should be noted that further quantitative analysis is carried out on the automatically extracted multi-scale immune microenvironment features, specifically including quantitative analysis of the overall density index of immune cells, the spatial distribution pattern index of immune cells, the index of the distance and aggregation degree between immune cells, and the index of the spatial co-localization characteristics of different types of immune cells.

[0032] Among them, the spatial distribution pattern index of immune cells includes the marginal infiltration type and the scattered infiltration type.

[0033] Furthermore, collect and process the clinical and pathological basic information of patients, and comprehensively construct the immune microenvironment characteristic data of patients based on the results of multi-scale analysis of the immune microenvironment.

[0034] The clinical and pathological basic information includes: Patient age: Precisely record the age of the patient at the time of the first diagnosis; Body mass index (BMI): Collect height and weight data using a unified measurement standard and calculate the BMI; Medical history information: Record the patient's family history of tumors, history of endocrine diseases, previous history of malignant tumors, etc.; Reproductive history information: Detailed record of the patient's number of pregnancies, number of births, number of miscarriages, and history of contraceptive drug use; Treatment experience: The patient's previous treatment plan.

[0035] Based on a unified patient identifier, perform precise matching and fusion of clinical and pathological data and immune microenvironment characteristic data. Taking the patient as the core, integrate clinical and pathological characteristic data and immune microenvironment characteristic data to construct a multi-dimensional comprehensive feature matrix; perform unified preprocessing on the fusion data to ensure the quality of the fusion data and the accuracy of subsequent analysis. Construct a multi-modal data fusion structure to form a comprehensive clinical-immune feature integrated data system.

[0036] Among them, the immune microenvironment characteristics include the density value of immune cells, the spatial distribution pattern, the co-localization relationship of cell types, and the infiltration difference between the marginal and central regions.

[0037] The infiltration difference represents the immune infiltration difference between the marginal and central regions, referring to the differences in the distribution, density, functional status, etc. of immune cells between the periphery (marginal region) and the interior (central region) of the tumor.

[0038] Firstly, preliminarily evaluate the contribution weights of each feature to the efficacy of conservation treatment and identify potential key features; secondly, further confirm the prediction stability and contribution degree of the selected feature variables through cross-validation methods; finally, according to the screening results, clarify the most predictive combined feature combinations in clinicopathology and the immune microenvironment, and ultimately determine the key feature set for subsequent construction of a precision prediction model.

[0039] S3: Integrate the information of the multimodal clinicopathological database, construct a multimodal intelligent evaluation model for predicting the response to conservation treatment of endometrial cancer based on the immune feature database. By setting a multi-source data fusion mechanism, comprehensively analyze the correlation between clinicopathological features, tissue image structures, immune infiltration status, and molecular subtype patterns, identify the feature variables affecting the conservation treatment effect, construct a treatment response prediction index, and dynamically optimize and update the parameters of the multimodal intelligent evaluation model in combination with the treatment feedback data of real patients.

[0040] Furthermore, according to the constructed multimodal clinicopathological database, accurately extract the pathological feature data and molecular subtype feature data, and respectively perform standardization processing on the two types of data to ensure the quality and consistency of the data, forming a clear pathological feature set and the molecular subtype feature set .

[0041] For the pathological feature data set , construct a pathological feature transformation function , expressed as: ; For the molecular subtype feature set and the pathological feature set , construct a non-linear fusion function of molecular subtypes and pathological features , expressed as: ; Based on the pathological feature transformation function and the non-linear fusion function of molecular subtypes and pathological features , construct a multimodal intelligent evaluation model, expressed as: ; Among them, is the fusion prediction index, with a value range of 0 to 1. A value close to 1 indicates a good predicted treatment effect, and a value close to 0 indicates a poor effect. are the model parameters determined by fitting with actual data. is the feature adaptive coefficient, reflecting the dynamic contribution degree of each feature. is the set of pathological feature vectors. represents the value of the i-th pathological feature vector. represents the value of the k-th pathological feature vector, is the set of molecular subtype feature vectors, is the th eigenvalue of the molecular subtype, is the total number of pathological features, is the dimension of the pathological feature vector set, is the dimension of the molecular subtype feature vector set, and are the mean vectors of the molecular subtype features and pathological features respectively, represents the norm of the pathological feature vector, is the adjustment threshold of the pathological feature transformation function , and the specific data is shown in Table 1.

[0042] Table 1 Adjustment Threshold Setting Table Feature number K Pathological feature description Numerical range (normalized) Suggested threshold μ Basis for setting <![CDATA[P 1 > Tumor cell nuclear density ( / mm²) 0.00 – 1.00 0.65 Values above this indicate excessive cell proliferation and increased risk <![CDATA[P 2 > Glandular structure integrity score 0.00 – 1.00 0.45 Values below this indicate glandular structure damage and a potentially poor response <![CDATA[P 3 > CD8+ T cell tumor margin density ( / mm²) 0.00 – 1.00 0.55 Higher values indicate active immune infiltration, which is beneficial for treatment response <![CDATA[P 4 > Macrophage central distribution ratio 0.00 – 1.00 0.3 Higher distribution in the center may indicate immune escape and a poor response <![CDATA[P 5 > Nuclear atypia score 0.00 – 1.00 0.7 Significant distortion is often associated with high grade and a poor prognosis

[0043] Fusion Prediction Index is strictly limited in the range of . When the model output value is close to 1, it indicates that the patient's treatment effect is more likely to show a positive response. On the contrary, when it is close to 0, it indicates that the treatment effect is poor, which can guide individualized and precise clinical decision-making.

[0044] It should be noted that from the standardized database, the patient's clinical feature vector , multi-scale pathological image feature vector , immune microenvironment feature vector and molecular subtype information feature vector are accurately extracted, and it is ensured that all input data has undergone strict data quality control and standardization processing to meet the consistency and accuracy requirements of model input.

[0045] To achieve precise fusion between multi-modal features, a multi-modal fusion function is designed, which is expressed as: ; The multi-modal Attention mechanism is introduced, where is a learnable transformation matrix determined by clinical data training. The Attention mechanism accurately captures the internal correlations of each modal data by automatically calculating the attention weights between modal features. The parameter is used to precisely control the Attention fusion intensity, enabling the model to more accurately reflect the clinical actual situation.

[0046] Adaptive weight dynamic adjustment function , which is expressed as: ; Based on the fusion function and the adaptive weight adjustment function , a prediction fusion index is constructed which is expressed as: ; wherein, represents the output index of the final prediction model, with a value ranging from 0 to 1. Being close to 1 indicates a good response of the patient to conservation treatment, and being close to 0 indicates a poor response represents the logistic regression function represents the number of individual patients represents the number of feature dimensions is the adaptive adjustment factor, used to measure the contribution degree of the individual characteristics of the th patient is the multi-modal fusion function of innovative design respectively represent the clinical characteristics, multi-scale pathological image characteristics, immune microenvironment characteristics and molecular subtype information vectors of the th patient is the trainable transformation matrix corresponding to the modality is the feature fusion scale control factor respectively represent the global feature mean vectors corresponding to the modalities represents the Euclidean norm of the vector is the adaptive weight dynamic adjustment function is the th feature dimension adaptive weight adjustment coefficient

[0047] Prediction index An output value close to 1 indicates that the patient may show a better treatment response to conservation treatment under the comprehensive analysis of clinical characteristics, multi-scale pathological image characteristics, immune microenvironment characteristics and molecular subtype characteristics; on the contrary, the closer the predicted value is to 0, the worse the treatment response of the patient may be, and the treatment plan needs to be adjusted in time in clinical practice

[0048] After the prediction model is constructed, according to the clear inclusion criteria and exclusion criteria, independent of the model development process, new endometrial cancer conservation treatment patients who have not participated in the modeling are recruited to construct an independent validation cohort; the clinical characteristics of the patients in this cohort are uniformly and standardized , multi-scale pathological image characteristics , immune microenvironment characteristics and molecular subtype information are collected, and after the data is standardized and structured, it is substituted into the prediction model to obtain the predicted output value ;The predicted results of the model are strictly compared and analyzed with the actual clinical treatment follow-up results. Multiple performance indicators such as the area under the ROC curve (AUC), prediction accuracy (Accuracy), sensitivity (Sensitivity), and specificity (Specificity) are comprehensively used to evaluate the clinical generalization and accuracy of the prediction model; further analyze the consistency between the model prediction output and the actual treatment effect, clarify the correlation between the model output value and the patient's true treatment response, and judge the practical application reliability of the model.

[0049] Establish a real-time dynamic clinical feedback optimization loop to continuously optimize the accuracy and generalization performance of the prediction model: for each patient receiving endometrial cancer conservation treatment during the actual application process of the model, collect their treatment response data in real time, including the true clinical treatment response; perform a one-by-one matching analysis on the clinically actual treatment response data collected in real time and the model prediction output value to accurately identify possible prediction errors or feature biases in the prediction model; based on the error analysis results of the feedback data, use an adaptive feature weight dynamic adjustment function to adjust the feature weight parameters in real time; the optimized prediction model is continuously prospectively verified in a new independent clinical cohort, and the closed-loop process of "prediction - feedback - optimization - re-verification" is continuously carried out to ensure the long-term stability and clinical generalization ability of the prediction model.

[0050] Continuously implement the clinical feedback optimization loop of the prediction model: establish a standardized feedback data management platform to record the clinical prediction results and actual treatment follow-up results of patients in real time; regularly perform statistical analysis on the clinical feedback data to clarify the changing trend of prediction performance and diagnose possible prediction error patterns; continuously monitor and maintain the model performance, and periodically optimize and adjust the feature combination and parameter settings of the prediction model to ensure that the prediction model continuously adapts to the actual clinical changes.

[0051] Example 2 is the second example of the present invention, which is different from the previous two examples in that: When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0052] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a predefined list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0053] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), optical fiber devices, and portable compact disc read-only memories (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as appropriate, and then storing it in a computer memory.

[0054] It should be understood that each part of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0055] Embodiment 3 is the third embodiment of the present invention. This embodiment provides a system for predicting the response of endometrial cancer conservation treatment based on multi-dimensional pathological evaluation, including a multi-modal pathological data standardization construction and feature refinement extraction module, an intelligent fusion prediction module for molecular subtype information and pathological features, and a multi-modal fusion intelligent prediction model construction and clinical verification optimization module. The multi-modal pathological data standardization construction and feature refinement extraction module is used to standardize the collection and structured processing of the clinical pathological information, tissue pathological images, immune microenvironment features, and molecular subtype data of endometrial cancer patients to form a multi-modal clinical pathological database. The intelligent fusion prediction module for molecular subtype information and pathological features is used to determine the comprehensive role and contribution degree of pathological features and molecular subtypes in predicting the treatment response, and determine the predictive value of the fusion features for the patient's treatment response. The multi-modal fusion intelligent prediction model construction and clinical verification optimization module is used to construct a multi-modal intelligent evaluation model and optimize it through the verification of a clinical independent cohort and actual feedback.

[0056] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological evaluation, characterized in that: include: Collect the patient's basic clinical information, histopathological image data, immune microenvironment indicators and molecular subtype information to build a multimodal clinical pathology database, standardize the entry and management of multimodal clinical pathology database information, and establish a dynamic update and quality control mechanism; Perform multi-scale analysis and quantitative processing on image data and immune features in the multimodal clinical pathology database. Through the image preprocessing process, color correction, background removal, and image enhancement are performed on tissue slice images. Multi-scale spatial segmentation is used to identify tumor areas at the macro scale, extract glandular structures and cell-dense areas at the tissue scale, and identify cell nuclear morphology and cell density information at the cell scale. The distribution characteristics, infiltration density, aggregation degree, and spatial interaction of immune cells are extracted, and the spatial characteristic patterns of the immune microenvironment are expressed in a structured manner to build an immune feature database. By integrating the information of multimodal clinical pathology database, a multimodal intelligent evaluation model for predicting the response to conservative treatment of endometrial cancer was constructed based on the immune signature database. By setting a multi-source data fusion mechanism, the correlation between clinical pathological characteristics, tissue image structure, immune infiltration status and molecular subtype patterns was comprehensively analyzed, the characteristic variables that affect the effect of conservative treatment were identified, and a treatment response prediction index was constructed. The multimodal intelligent evaluation model was dynamically optimized and its parameters updated in combination with real patient treatment feedback data.

2. The method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to claim 1, characterized in that: The image preprocessing process includes analyzing the image, standardizing the image color, unifying the resolution, and removing the background noise.

3. The method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to claim 2, characterized in that: The multi-scale spatial segmentation includes three levels: Identify the overall tumor area at the macroscopic level; Identify glands and tissue structures at the tissue scale; Identify cellular morphological features and nuclear density at the cellular scale.

4. The method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to claim 3, characterized in that: The spatial characteristic patterns of the immune microenvironment include the density value of immune cells, the spatial distribution pattern, the co-localization relationship of cell types, and the infiltration difference between the edge and the central area.

5. The method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to claim 4, characterized in that: The multimodal intelligent assessment model includes feature contribution analysis, feature screening and combination methods to identify associated variables among clinical information, image information, immune indicators and molecular subtype information.

6. The method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to claim 5, characterized in that: The molecular subtype pattern is obtained by sequencing technology or immunohistochemistry, including collecting molecular features, converting and structuring the molecular features into a standardized format, and matching and fusing modality data at the patient level; Molecular characteristics included POLE gene mutation status, mismatch repair protein status, and p53 expression pattern.

7. The method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to claim 6, characterized in that: The dynamic optimization and parameter updating of the multimodal intelligent assessment model in combination with real patient treatment feedback data includes constructing the multimodal intelligent assessment model, implementing prospective verification through an independent clinical verification cohort, and continuously adjusting the structure and parameters of the multimodal intelligent assessment model based on actual treatment effect feedback.

8. A system using the method for predicting the response to conservation treatment of endometrial cancer based on multi-dimensional pathological assessment as described in any one of claims 1 to 7, characterized in that: It includes a module for standardization and refinement of multimodal pathology data, a module for intelligent fusion prediction of molecular subtype information and pathological features, and a module for intelligent prediction model construction and clinical verification optimization of multimodal fusion; The multimodal pathology data standardization construction and feature refinement extraction module is used to standardize the collection and structured processing of clinical pathology information, histopathological images, immune microenvironment characteristics and molecular subtype data of endometrial cancer patients to form a multimodal clinical pathology database; The intelligent fusion prediction module of molecular subtype information and pathological characteristics is used to determine the comprehensive role and contribution of pathological characteristics and molecular subtypes in predicting treatment response, and determine the predictive value of fusion characteristics for patient treatment response; The multimodal fusion intelligent prediction model construction and clinical verification optimization module is used to construct a multimodal intelligent evaluation model, and is optimized through verification and actual feedback from clinical independent cohorts.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the response to conservative treatment of endometrial cancer based on multi-dimensional pathological assessment according to any one of claims 1 to 7 are implemented.

Images