Large model analysis system and method for endometrial cancer data management
By constructing a large-scale model analysis system, the problem of relying on doctors' experience in molecular subtyping detection of endometrial cancer patients has been solved. Objective detection method recommendations and adaptive updates have been achieved, improving detection accuracy and model update efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-02
- Publication Date
- 2026-05-15
AI Technical Summary
In molecular subtyping of endometrial cancer patients, current technologies lack data-driven decision support, leading to reliance on doctors' personal experience in selecting testing methods. Furthermore, the scarcity of samples makes it difficult to update large models, thus failing to achieve optimal decision support.
A large-scale model analysis system is constructed. By determining the total amount of sample data, dividing the feature data information, forming matching sub-intervals, using decision trees to build a training model, forming recommendation decisions, and achieving adaptive updates when the sample is scarce.
It provides objective recommendations for testing methods, reduces differences in doctors' experience, minimizes resource waste, shortens patient waiting time, improves model update efficiency, and reduces medical costs.
Smart Images
Figure CN122050503A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of large-scale data model analysis technology, specifically a large-scale model analysis system and method for endometrial cancer data management. Background Technology
[0002] Endometrial cancer is a group of epithelial malignant tumors that occur in the endometrium. Currently, the incidence rate is on a continuous upward trend. Therefore, the prognostic analysis of endometrial cancer patients is becoming increasingly important. With current technology, molecular subtyping has become a standard procedure and a key part of clinical management for newly diagnosed endometrial cancer patients. Molecular subtyping is generally divided into four types: POLE mutation type, MMR deficiency type, p53 mutation type (p53abn), and nonspecific molecular alteration type (NSMP).
[0003] In molecular typing, when MMR deficiency is detected, there are generally two clinical methods: PCR (using PCR technology to detect the length of multiple microsatellite loci in the genome) and immunohistochemistry (using immunohistochemistry to detect the expression of MMR core proteins). PCR is the gold standard with good detection results, but in actual clinical practice, the usage rate of immunohistochemistry far exceeds that of PCR, mainly because its concordance with PCR exceeds 90%, but it is cheaper, faster, and more in line with actual needs. Therefore, in actual molecular typing, the choice between the two methods is made based solely on the doctor's personal experience, lacking specific data to support decision-making. At the same time, due to the scarcity of data samples, there are also many drawbacks in updating large-scale models. Therefore, how to introduce large-scale data models to achieve optimal decision support to meet the needs of digital healthcare under the premise of scarce sample size is one of the problems that needs to be solved. Summary of the Invention
[0004] The purpose of this invention is to provide a large-scale model analysis system and method for endometrial cancer data management, in order to solve the problems raised in the prior art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a large-scale model analysis method for endometrial cancer data management, the method comprising: Determine the total amount of sample data; each group of sample data should include the patient's characteristic data information, the judgment results of molecular subtyping data determined by immunohistochemistry, and record 1 for those that conform to the molecular subtyping given by PCR and 0 for those that do not conform. Based on the patient's characteristic data, the features are divided, the anastomosis rate under each feature division is calculated, and several anastomosis sub-intervals are formed based on the anastomosis rate. A training model is built to generate the probability that the matching rate falls within different matching sub-intervals based on different feature data information, and a recommendation decision is generated and fed back to the administrator.
[0006] The method also includes: Based on the current total amount of sample data, a grouped fitting method is adopted to establish a functional relationship between the amount of newly added sample data and the increase in the accuracy of the large model under different total amounts of sample data, forming update nodes and realizing the adaptive update of the large model.
[0007] According to the above technical solution, the calculation of the matching rate under each feature segmentation, and the formation of several matching sub-intervals based on the matching rate, includes: The concordance rate under each feature segmentation refers to the ratio of the number of samples that conform to the molecular subtyping given by PCR under each feature segmentation to the number of samples under that feature segmentation. After obtaining the matching rate under all feature divisions, the minimum and maximum values are taken to form a matching interval. The matching interval is then divided equally to form several matching sub-intervals. Except for the last matching sub-interval, the endpoint values of each matching sub-interval include the preceding value but not the following value. During the equal division process, an even number of equal divisions is used.
[0008] According to the above technical solution, the construction of the training model to form the probability that the matching rate falls within different matching sub-intervals based on different feature data information includes: Based on the decision tree concept, a training model is constructed by progressively adding multiple trees, with the total number of trees set to T. The prediction function of the t-th tree for the matching sub-interval k is expressed as: ;in, The feature vector of the patient; the leaf weight vector of the t-th tree is expressed as: The leaf node function is expressed as: ; Soon Mapping to leaf number ; in, It is a recursive binary decision tree, based on the split feature index formed when the administrator divides the features; For any new patient, whose feature vector is represented by Y, the original score of each anastomotic sub-interval is... for: ; in, It is the initial bias of the matching sub-interval k; Features The weight value of a leaf in the matching sub-interval k of the t-th tree is expressed by the corresponding leaf node function mentioned above. The final prediction model is constructed by progressively adding multiple trees. For each matching sub-interval k: ; in, This represents the learning rate, used to control the contribution of each tree. After iterating through T trees, the resulting score is converted into a probability expression. The anastomosing interval corresponding to the maximum probability is taken as the probability interval for judging whether the molecular subtyping data determined by immunohistochemistry for new patients matches the molecular subtyping given by PCR.
[0009] According to the above technical solution, the step of forming a recommendation decision feedback to the administrator port means that after obtaining the probability range, if the minimum value of the probability range is greater than the system preset threshold, a recommendation to use immunohistochemistry is formed and sent to the administrator port; if the maximum value of the probability range is less than the system preset threshold, a decision not to use immunohistochemistry is formed and sent to the administrator port; if the system preset threshold is within the probability range, an adaptive decision is formed and sent to the administrator port, which is then further output by the administrator.
[0010] According to the above technical solution, S6-1. Based on the determined total amount of sample data, and using a fixed test group to test the trained model, a first accuracy rate is obtained. S6-2. From the determined total amount of sample data, M groups of samples are randomly selected as new samples. M is greater than the number of samples set by the system. The remaining NM are identified as the first sample group data. Based on the first sample group data and the training model, a fixed test group is used to test the training model to form a second accuracy. The second accuracy is compared with the first accuracy. If the second accuracy is higher than the first accuracy, the data group is deleted. The difference between the second accuracy and the first accuracy is calculated and marked as the accuracy increase value. M groups of sample data are randomly selected L1 times, and the average of all accuracy increase values is taken as the accuracy increase value under the first sample group data. M and L1 are both set by the system. S6-3. Change the number of new samples and repeat step S6-2 to generate several groups of first sample group data along with their corresponding number of new samples and accuracy increase values; after normalizing the data, use a multinomial fitting method to generate the following regression equation: ; in, The normalized value of the increase in accuracy; , These represent the coefficients for the first sample group data and the number of newly added samples, respectively. , These represent the normalized values of the first sample group data and the number of newly added samples, respectively. Represents the error term; Based on regression, after determining the current number of samples, the system plans the accuracy increase value that needs to be achieved for each model update, and feeds back the corresponding predicted value of the number of new samples based on the accuracy increase value. The predicted value is used as the model update node, that is, when the number of new sample data reaches the predicted value, the large model is automatically updated.
[0011] According to the above technical solution, the feature segmentation is performed by the administrator, and the patient's feature data information includes patient age, ER_status, Ki67, HER2_score, and PR_status.
[0012] This application also includes a large-scale model analysis system for endometrial cancer data management, the system comprising: The sample data management module is used to determine the total amount of sample data. Each group of sample data should include the patient's characteristic data information and the judgment results of the molecular subtyping data determined by immunohistochemistry. If it matches the molecular subtyping given by PCR, it is recorded as 1, and if it does not match, it is recorded as 0. The interval division module divides features based on the patient's feature data, calculates the anastomosis rate under each feature division, and forms several anastomosis sub-intervals based on the anastomosis rate; The model training module constructs and trains a model, generating probabilities that the matching rate falls within different matching sub-intervals based on different feature data information. This information is then used to generate recommendation decisions and feed them back to the administrator. The system also includes: The adaptive update module, based on the currently determined total amount of sample data, uses a grouped fitting method to establish a functional relationship between the amount of newly added sample data and the increase in the accuracy of the large model under different total amounts of sample data, forming update nodes to achieve adaptive updates of the large model.
[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: This application can solve the significant subjectivity of the traditional model that relies solely on the doctor's personal experience to select testing methods. By establishing objective decision-making criteria through data-driven approaches, it reduces selection bias caused by differences in doctors' experience. Under the premise of ensuring testing accuracy, the system prioritizes recommending economical and fast testing methods, providing a data foundation for the doctor's final decision. At the same time, it can avoid resource waste, shorten patient waiting time, and reduce overall medical costs. Furthermore, based on the current situation of scarce sample data, this application proposes a process for adaptive update nodes of the model, effectively reducing the difficulty of later operation and maintenance and improving the overall model update efficiency. Attached Figure Description
[0014] Figure 1 This is a flowchart illustrating a large-scale model analysis method for endometrial cancer data management according to the present invention. Detailed Implementation
[0015] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0016] Example: Figure 1 As shown in this embodiment, a large-scale model analysis method for endometrial cancer data management is provided. This method is primarily applicable to medical environments where sample data is scarce and data-driven decision support is lacking. Specifically, it includes: In this embodiment, the total amount of available sample data is determined and denoted as N groups. Each group of sample data should include the patient's characteristic data information. The molecular subtyping data determined by immunohistochemistry is used. If the molecular subtyping results match the molecular subtyping given by PCR, they are denoted as 1, and if they do not match, they are denoted as 0. Define the patient's characteristic data as denoted as ,in, The total number of feature data participating in the training is represented. In this embodiment, patient age, ER_status (estrogen receptor status), Ki67, HER2_score, and PR_status (progesterone receptor status) are used as training features. In actual testing, other feature data may also be included, such as race, region, etc. For ease of expression, this application selects five common features for processing. Taking any single characteristic as an example, such as patient age, with less than 25 years old as the standard, all sample data under the age of 25 are collected to form the concordance rate under the less than 25-year-old standard, which is the ratio of the number of samples that meet the molecular subtyping given by PCR to the number of samples under 25 years old. Characteristic criteria are established for each characteristic, such as ER_status, which can be set to whether it is positive, whether Ki67 is greater than 25, etc. The specific characteristic settings can be adjusted according to the system. After obtaining the matching rate under all feature criteria, the minimum and maximum values are taken to form a matching interval. The matching interval is then divided into several matching sub-intervals. Except for the last matching sub-interval, the endpoint values of each matching sub-interval include the preceding value but not the following value. In the process of dividing, an even number of equal parts is generally used to avoid gaps. In this embodiment, taking a four-part division as an example, assuming the matching interval is 85% to 97%, the matching sub-intervals after the four-part division are represented as: 85% to 88% (excluding 88%), 88% to 91% (excluding 91%), 91% to 94% (excluding 94%), and 94% to 97%. The data is processed using a multi-classification model as follows: Let the total number of trees be T, where the prediction function of the t-th tree for the matching sub-interval k is expressed as: ;in, The feature vector of the patient; the leaf weight vector of the t-th tree is expressed as: The leaf node function is expressed as: ; Soon Mapped to leaf number; For any new patient, whose feature vector is represented by Y, the original score for each anastomotic sub-interval is: ; in, It is the initial bias of the matching sub-interval k; Features The weight value of a leaf in the matching sub-interval k of the t-th tree is expressed by the corresponding leaf node function mentioned above. The final prediction model is constructed by progressively adding multiple trees. For each matching sub-interval k: ; in, This represents the learning rate, which is used to control the contribution of each tree. Specific examples are as follows: In the initialization phase, t=0, the distribution of the four matching sub-intervals in the training set is assumed to be [40%, 30%, 20%, 10%]. The distribution is calculated as the ratio of the number of data points within the matching sub-interval to the total sample size. Therefore, the initial score of the matching sub-interval corresponding to 40% (based on prior concepts) is expressed as: The rest are handled in the same way. Taking a single tree as an example, we construct the first tree for each feature relative to the matching sub-interval, defined as follows: First matching interval (40%) if ER_status<0.5: If age < 50: leaf_weight = -0.2 else: leaf_weight = 0.1 else: If Ki67 < 20: leaf_weight = 1.8 else: leaf_weight = 0.5 Second anastomotic interval (30%) if ER_status<0.5: leaf_weight = -0.3 else: If Ki67 < 20: leaf_weight = 0.7 else: leaf_weight = 1.5 Third anastomotic interval (20%) If HER2_score < 2: leaf_weight = -0.4 else: leaf_weight = 2.0 Fourth matching sub-interval (10%) If ER_status > 0.5: leaf_weight = -1.0 else: if PR_status>0.5: leaf_weight = -0.5 else: leaf_weight = 0.8 In a real model, there should be multiple trees. In this embodiment, it is only a brief example, defined as one tree per matching sub-interval per round. If patient Y's characteristics are: ER_status = 1 (positive), age = 55, Ki67 = 25, HER2_score = 1, and PR_status = 0; then substituting these values into the above equations, we can further obtain: First anastomotic interval: ER_status=1≥0.5, Ki67=25≥20 → leaf_weight=0.5; Second anastomotic interval: ER_status=1≥0.5, Ki67=25≥20 → leaf_weight=1.5 Third anastomotic interval: HER2_score=1<2 → leaf_weight=-0.4 Fourth matching interval: ER_status=1>0.5 → leaf_weight=-1.0 With a learning rate of 0.3, further scores under the first tree can be calculated based on the initial scores of the matching sub-intervals. First anastomotic interval:
[0017] Second anastomotic interval:
[0018] Third anastomotic interval:
[0019] Fourth anastomotic interval:
[0020] Thus, the iteration of the first tree is complete. Subsequent iterations follow the same method as the first, and will not be repeated here. Each iteration builds upon the previous one, continuing until T trees are reached. The predicted output is: ; Here, S represents the example output of the first iteration, and this embodiment uses the first iteration as an illustration. The probability that the data obtained from immunohistochemistry for patient Y falls within the first anastomotic region is: ; The rest The algorithm is the same; the maximum probability is taken as the final prediction output. The system can make a judgment based on an adjustable threshold. For example, if the final prediction result falls in the first anastomotic interval, and the system requires an accuracy of 95%, then the system will decide that immunohistochemistry cannot replace PCR, for the doctor's reference.
[0021] In practical use, due to the lack of medical data sharing among medical institutions and the issue of verifying patient privacy data, sample data is often scarce. Therefore, there is a lack of opportunities to proactively update the large model. For example, current model updates are often periodic, with sample data updated every six months. However, such updates often suffer from significant problems in the fine-tuning effect due to the variable number of new sample data. Therefore, in the embodiments of this application, an additional method for updating a large data model for endometrial cancer is presented. Taking the total amount of currently available sample data N as an example, the above-mentioned prediction model is determined, and a fixed test group is used to test the determined prediction model to form a first accuracy rate. From the currently available samples, M groups (M is greater than the system-set quantity threshold) of samples are randomly selected (as new samples). The remaining NM are identified as the first sample group data. Based on the first sample group data, the above prediction model is determined. The determined prediction model is tested using a fixed test group to form a second accuracy. The second accuracy is compared with the first accuracy. If the second accuracy is higher than the first accuracy, the data in that group is deleted. The difference between the second accuracy and the first accuracy is calculated and marked as the accuracy increase value. M groups of sample data are randomly selected L1 times, and the average of all accuracy increase values is taken as the accuracy increase value under the first sample group data. Change the number of new samples and repeat the above steps to generate several groups of first sample group data along with their corresponding number of new samples and accuracy increase values; after normalizing the data, use a multinomial fitting method to generate the following regression equation: ; in, The normalized value of the increase in accuracy; , These represent the coefficients for the first sample group data and the number of newly added samples, respectively. , These represent the normalized values of the first sample group data and the number of newly added samples, respectively. Represents the error term; Based on the above regression formula, after determining the current number of samples, the system can plan the accuracy increase value that needs to be achieved for each model update, and feed back the corresponding predicted value of the number of new samples based on the accuracy increase value. The predicted value is used as the model update node, that is, when the number of new sample data reaches the predicted value, the large model is automatically updated.
[0022] This embodiment also includes a large-scale model analysis system for endometrial cancer data management, the system comprising: The sample data management module is used to determine the total amount of sample data. Each group of sample data should include the patient's characteristic data information and the judgment results of the molecular subtyping data determined by immunohistochemistry. If it matches the molecular subtyping given by PCR, it is recorded as 1, and if it does not match, it is recorded as 0. The interval division module divides features based on the patient's feature data, calculates the anastomosis rate under each feature division, and forms several anastomosis sub-intervals based on the anastomosis rate; The model training module constructs and trains a model, generating probabilities that the matching rate falls within different matching sub-intervals based on different feature data information. This information is then used to generate recommendation decisions and feed them back to the administrator. The system also includes: The adaptive update module, based on the currently determined total amount of sample data, uses a grouped fitting method to establish a functional relationship between the amount of newly added sample data and the increase in the accuracy of the large model under different total amounts of sample data, forming update nodes to achieve adaptive updates of the large model.
[0023] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the platforms, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0024] The use of prefixes such as "first" and "second" in this application embodiment is solely for distinguishing different descriptive objects and does not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes to distinguish descriptive objects in this application embodiment does not constitute a limitation on the described objects. The description of the described objects is found in the claims or the context of the embodiments, and the use of such prefixes should not constitute unnecessary restrictions.
[0025] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0026] In the various embodiments of this application, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions between the various embodiments are consistent and can be referenced by each other. Technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.
[0027] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0028] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0029] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A large-scale model analysis method for endometrial cancer data management, characterized by: The method includes: Determine the total amount of sample data; each group of sample data should include the patient's characteristic data information, the judgment results of molecular subtyping data determined by immunohistochemistry, and record 1 for those that conform to the molecular subtyping given by PCR and 0 for those that do not conform. Based on the patient's characteristic data, the features are divided, the anastomosis rate under each feature division is calculated, and several anastomosis sub-intervals are formed based on the anastomosis rate. A training model is built to generate the probability that the matching rate falls within different matching sub-intervals based on different feature data information, and a recommendation decision is generated and fed back to the administrator.
2. The large-scale model analysis method for endometrial cancer data management according to claim 1, characterized in that: The method also includes: Based on the current total amount of sample data, a grouped fitting method is adopted to establish a functional relationship between the amount of newly added sample data and the increase in the accuracy of the large model under different total amounts of sample data, forming update nodes and realizing the adaptive update of the large model.
3. The large-scale model analysis method for endometrial cancer data management according to claim 1, characterized in that: The calculation of the matching rate under each feature segmentation, and the formation of several matching sub-intervals based on the matching rate, includes: The concordance rate under each feature segmentation refers to the ratio of the number of samples that conform to the molecular subtyping given by PCR under each feature segmentation to the number of samples under that feature segmentation. After obtaining the matching rate under all feature divisions, the minimum and maximum values are taken to form a matching interval. The matching interval is then divided equally to form several matching sub-intervals. Except for the last matching sub-interval, the endpoint values of each matching sub-interval include the preceding value but not the following value. During the equal division process, an even number of equal divisions is used.
4. The large-scale model analysis method for endometrial cancer data management according to claim 1, characterized in that: The construction of the training model, which generates the probability that the concordance rate falls within different concordance sub-intervals based on different feature data information, includes: Based on the decision tree concept, a training model is constructed by progressively adding multiple trees, with the total number of trees set to T. The prediction function of the t-th tree for the matching sub-interval k is expressed as: ;in, The feature vector of the patient; the leaf weight vector of the t-th tree is expressed as: The leaf node function is expressed as: ； Soon Mapping to leaf number ; in, It is a recursive binary decision tree, based on the split feature index formed when the administrator divides the features; For any new patient, whose feature vector is represented by Y, the original score of each anastomotic sub-interval is... for: ； in, It is the initial bias of the matching sub-interval k; Features The weight value of a leaf in the matching sub-interval k of the t-th tree is expressed by the corresponding leaf node function mentioned above. The final prediction model is constructed by progressively adding multiple trees. For each matching sub-interval k: ； in, This represents the learning rate, which is used to control the contribution of each tree. After iterating through T trees, the resulting score is converted into a probability expression. The anastomosing interval corresponding to the maximum probability is taken as the probability interval for judging whether the molecular subtyping data determined by immunohistochemistry for new patients matches the molecular subtyping given by PCR.
5. The large-scale model analysis method for endometrial cancer data management according to claim 4, characterized in that: The process of generating a recommendation decision and feeding it back to the administrator means that after obtaining the probability range, if the minimum value of the probability range is greater than the system's preset threshold, a recommendation to use immunohistochemistry is generated and sent to the administrator; if the maximum value of the probability range is less than the system's preset threshold, a decision not to use immunohistochemistry is generated and sent to the administrator; if the system's preset threshold is within the probability range, an adaptive decision is generated and sent to the administrator for further output.
6. The large-scale model analysis method for endometrial cancer data management according to claim 2, characterized in that: S6-1. Based on the determined total amount of sample data, and using a fixed test group to test the trained model, a first accuracy rate is obtained. S6-2. From the determined total amount of sample data, M groups of samples are randomly selected as new samples. M is greater than the number of samples set by the system. The remaining NM are identified as the first sample group data. Based on the first sample group data and the training model, a fixed test group is used to test the training model to form a second accuracy. The second accuracy is compared with the first accuracy. If the second accuracy is higher than the first accuracy, the data group is deleted. The difference between the second accuracy and the first accuracy is calculated and marked as the accuracy increase value. M groups of sample data are randomly selected L1 times, and the average of all accuracy increase values is taken as the accuracy increase value under the first sample group data. M and L1 are both set by the system. S6-3. Change the number of new samples and repeat step S6-2 to generate several groups of first sample group data along with their corresponding number of new samples and accuracy increase values; after normalizing the data, use a multinomial fitting method to generate the following regression equation: ； in, The normalized value of the increase in accuracy; , These represent the coefficients for the first sample group data and the number of newly added samples, respectively. , These represent the normalized values of the first sample group data and the number of newly added samples, respectively. Represents the error term; Based on regression, after determining the current number of samples, the system plans the accuracy increase value that needs to be achieved for each model update, and feeds back the corresponding predicted value of the number of new samples based on the accuracy increase value. The predicted value is used as the model update node, that is, when the number of new sample data reaches the predicted value, the large model is automatically updated.
7. The large-scale model analysis method for endometrial cancer data management according to claim 1, characterized in that: The feature segmentation is customized by the administrator, and the patient's feature data information includes patient age, ER_status, Ki67, HER2_score, and PR_status.
8. A large-scale model analysis system for endometrial cancer data management, characterized in that: The system includes: The sample data management module is used to determine the total amount of sample data. Each group of sample data should include the patient's characteristic data information and the judgment results of the molecular subtyping data determined by immunohistochemistry. If it matches the molecular subtyping given by PCR, it is recorded as 1, and if it does not match, it is recorded as 0. The interval division module divides features based on the patient's feature data, calculates the anastomosis rate under each feature division, and forms several anastomosis sub-intervals based on the anastomosis rate; The model training module builds and trains a model, generating the probability that the matching rate falls within different matching sub-intervals based on different feature data information, and then feeds back the recommendation decision to the administrator port.
9. A large-scale model analysis system for endometrial cancer data management according to claim 8, characterized in that: The system also includes: The adaptive update module, based on the currently determined total amount of sample data, uses a grouped fitting method to establish a functional relationship between the amount of newly added sample data and the increase in the accuracy of the large model under different total amounts of sample data, forming update nodes to achieve adaptive updates of the large model.