Intelligent auxiliary system for differentiated thyroid cancer postoperative evaluation and iodine-131 treatment decision

Abstract

The invention relates to an intelligent auxiliary system for differentiated thyroid cancer postoperative evaluation and iodine-131 treatment decision, and belongs to the field of medical technology and artificial intelligence, and the intelligent auxiliary system comprises a medical record data collection module which is used for receiving an auxiliary evaluation instruction and obtaining medical record data; the pathological feature extraction module is used for extracting pathological features from the medical record data through a preset analysis model; the analysis result output module is used for analyzing the pathological features through the analysis model and a pre-constructed thyroid cancer postoperative diagnosis and treatment decision tree and outputting reference suggestions so that doctors can know the reference suggestions; wherein the reference suggestions at least comprise reference suggestions for iodine-131 treatment. The application has the effects of improving the thyroid cancer postoperative diagnosis and treatment quality of doctors and improving the diagnosis and treatment standardability and efficiency.

Description

Technical Field

[0001] The present application relates to the fields of medical technology and artificial intelligence, and particularly to an intelligent assistance system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment. Background Art

[0002] Thyroid cancer is one of the common and solid tumors with a rapidly increasing incidence and a long disease management cycle. The disease burden of thyroid cancer in China is heavy, and differentiated thyroid cancer, which accounts for the highest proportion, is the most representative. Differentiated thyroid cancer retains to a certain extent the specific iodine uptake function of thyroid follicular cells, providing a theoretical basis for the application of radioactive iodine-131 treatment. Iodine-131 can act on the residual or metastatic foci that cannot be resected surgically, exerting a radiation killing effect and effectively reducing the recurrence and death risks of the disease. However, at the same time, iodine-131 treatment can also cause radiation-related adverse reactions, such as neck pain, salivary gland injury, nausea, abdominal pain, and in a small number of users, peripheral blood picture suppression. Therefore, screening suitable candidates for iodine-131 treatment, giving play to its killing effect on the lesion tissue while avoiding its adverse reactions on non-lesion tissues, and avoiding over-treatment or under-treatment are the key links in the postoperative management of differentiated thyroid cancer.

[0003] Currently, due to the obvious regional differences in medical levels and the influence of subjective factors such as the knowledge level and diagnosis and treatment experience of physicians, the quality of postoperative evaluation of thyroid cancer and the decision-making quality of iodine-131 diagnosis and treatment are uneven, thus affecting the effectiveness of the diagnosis and treatment of thyroid cancer patients, so there is a need for improvement. Summary of the Invention

[0004] In order to improve the quality of postoperative diagnosis and treatment of thyroid cancer by physicians and enhance the standardization and efficiency of diagnosis and treatment, the present application provides an intelligent assistance system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment.

[0005] In a first aspect, the present application provides an intelligent assistance system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment, including: A medical record data collection module, configured to receive an auxiliary evaluation instruction and obtain medical record data; A pathological feature extraction module, configured to extract pathological features from the medical record data through a preset analysis model; An analysis result output module, configured to analyze the pathological features through the analysis model and a pre-constructed postoperative diagnosis and treatment decision tree of thyroid cancer, and output a reference suggestion for the physician to know the reference suggestion; wherein the reference suggestion at least includes a reference suggestion for iodine-131 treatment.

[0006] By adopting the above technical solution, based on a pre-built analysis model, case characteristics related to thyroid cancer are efficiently extracted from medical record data, and then the analysis and judgment of pathological features are realized based on a pre-built decision tree for the diagnosis and treatment of thyroid cancer after surgery. Among them, the decision tree for the diagnosis and treatment of thyroid cancer after surgery can be considered as the diagnosis and treatment logic of thyroid cancer sorted out based on medical materials such as existing guidelines for differentiated thyroid cancer and expert consensus. This logic includes reference suggestions for iodine-131. Therefore, based on this decision tree to guide the analysis model to achieve efficient analysis of pathological features, a set of unified and standardized decision-making suggestions can be finally provided to physicians to help them achieve efficient and accurate decision-making on the diagnosis and treatment plan for thyroid cancer after surgery.

[0007] Optionally, the content of the decision nodes corresponding to the pre-built decision tree for the diagnosis and treatment of thyroid cancer after surgery at least includes TNM staging analysis, recurrence risk stratification analysis, and dynamic assessment analysis; The analysis result output module is used to traverse the decision nodes included in the pre-built decision tree for the diagnosis and treatment of thyroid cancer after surgery based on the decision nodes included in the pre-built decision tree for the diagnosis and treatment of thyroid cancer after surgery, analyze the pathological features, and output reference suggestions. The reference suggestions at least include: the postoperative treatment effect, the determination result of whether the conditions for iodine-131 treatment are met, and the personalized iodine-131 treatment dose range obtained when the conditions for iodine-131 treatment are met.

[0008] By adopting the above technical solution, TNM staging classifies the tumor size (T), lymph node metastasis (N), and distant metastasis (M) according to the standards of the International Union Against Cancer (UICC) and the American Joint Committee on Cancer (AJCC), and is the basis for evaluating the severity of the user's condition; recurrence risk stratification divides the risk degree of the user's condition according to the user's pathological features (such as tumor size, lymph node metastasis, vascular invasion, extrathyroidal invasion, etc.) and postoperative examination results (such as Tg level, imaging examination, etc.), so as to decide whether further treatment (such as iodine-131 treatment) is needed. Dynamic assessment is based on dynamic data such as serum thyroglobulin (Tg), antithyroglobulin antibody (TgAb), and imaging examinations (such as ultrasound, CT, PET-CT) during postoperative follow-up to evaluate the changes in the user's condition and the postoperative treatment effect. By traversing the decision tree for thyroid cancer after surgery covering the above three analysis and assessment schemes, the determination of the postoperative diagnosis and treatment effect of thyroid cancer is comprehensively realized, and more accurate reference suggestions for iodine-131 treatment are obtained, improving the comprehensiveness of the diagnosis and treatment assessment.

[0009] Optionally, the medical record data collection module is used to conduct question-and-answer interactions with the user based on a preset AI large model; it is also used to obtain and feedback the inquiry questions raised by the user during the interaction process, and extract medical record data from the inquiry questions; among them, the inquiry questions at least include content related to the condition of thyroid cancer and other extended content not related to the condition of thyroid cancer; the medical record data at least includes surgical cases related to thyroid cancer, condition changes, vital signs, and case reports. It further includes a decision tree optimization module, which is used to regularly optimize the postoperative diagnosis and treatment decision tree of thyroid cancer based on the inquiry content.

[0010] By adopting the above technical solutions, the AI large model uses natural language processing (such as NLP) technology to conduct question-and-answer interactions with the user, and obtains the user's medical record data during the question-and-answer process, replacing manual collection, enhancing the flexibility and intelligence level of the system, and being able to accurately and efficiently select medical record data related to thyroid cancer from the complex medical record data, realizing the intelligent and efficient collection of the user's medical record data. In addition, the AI large model proposed in this application can not only answer the inquiry questions related to the condition of thyroid cancer raised by the user, but also help the user answer other extended content (such as management suggestions for the hospital, etc.), so as to enhance the user experience during the interaction with the user. Optionally, the medical record data collection module is further used to match the obtained medical record data with a preset postoperative diagnosis and treatment decision tree of thyroid cancer based on a preset AI large model, and determine whether there is missing data based on the matching result. If so, an omission prompt is output to prompt the user to supplement the corresponding missing data.

[0011] By adopting the above technical solutions, when extracting pathological data from the interaction process with the user using natural language processing technology, it also includes identifying and parsing the text content input by the user and extracting key information (such as disease description, past medical history, etc.). The extracted key information can be matched with the decision node content of the postoperative decision tree of thyroid cancer to determine whether there is missing material (i.e., missing data). By prompting to fill in the missing data, the integrity of the medical record data can be improved, providing accurate and complete basic data for subsequent diagnosis and treatment.

[0012] Optionally, the reference suggestion further includes the postoperative follow-up frequency. It further includes an auxiliary evaluation trigger module, which is used to record the reference suggestion whenever an auxiliary evaluation instruction is obtained and the corresponding reference suggestion is determined; and trigger the auxiliary evaluation instruction according to the postoperative follow-up frequency included in the reference suggestion.

[0013] By adopting the above technical solution and combining the decision-making judgments in the three aspects of TNM staging analysis, recurrence risk stratification analysis, and dynamic assessment analysis covered by the decision tree for thyroid cancer after surgery mentioned above, the postoperative follow-up frequency for the user can be determined through recurrence risk stratification analysis. Correspondingly, this application proposes to automatically trigger an auxiliary assessment instruction for the corresponding user based on this follow-up frequency to achieve regular and dynamic diagnosis and treatment efficacy evaluation and reference suggestions for the user, and improve the timeliness of diagnosis and treatment efficacy evaluation and reference suggestions.

[0014] Optionally, the medical record data is divided into clinical data, genomics data, blood biochemical indicators, and imaging data; The pathological feature extraction module is used to extract key features from the clinical data through a natural language processing model; annotate gene sequencing data and extract mutation and expression features related to thyroid cancer from genomics data; standardize blood biochemical indicators through a machine learning model and identify abnormal features; extract visual features from the imaging data through a deep learning model; The pathological feature extraction module is also used to map all modal features to a unified feature space and perform fusion processing on all modal features through a pre-constructed multi-modal fusion model to generate pathological features.

[0015] By adopting the above technical solution, this application introduces a multi-modal fusion model, takes genomics data, blood biochemical indicators, imaging data, and clinical data as multi-modal data and inputs them into the pre-constructed multi-modal fusion model to generate a comprehensive feature representation (that is, finally generate pathological features), so as to provide more comprehensive and personalized reference suggestions by integrating multi-modal data, thereby assisting physicians in formulating more accurate and comprehensive treatment decisions.

[0016] Optionally, whenever the pathological features are obtained, the analysis result output module is also used to determine similar historical pathological features similar to the pathological features from the case features stored in the historical period based on a preset weighted similarity calculation method and the initial weight values preset for each modal data; calculate the difference between the reference suggestions corresponding to the similar historical cases and the reference suggestions analyzed by the analysis model for the currently obtained case features, and correct the initial weight values according to the difference, and then re-determine the similar historical pathological features corresponding to the current pathological features and their corresponding reference suggestions according to the corrected initial weight values; The analysis result output module is also used to output the corresponding similar historical pathological features and their corresponding reference suggestions when outputting the current pathological features and their corresponding reference suggestions.

[0017] By adopting the above technical solution, pathological features similar to the current pathological features in the historical period are found, and the initial weight value corresponding to the multimodal data is corrected by combining the differences between the reference suggestions corresponding to the aforementioned pathological features, and then the similar pathological features and their corresponding reference suggestions are re-determined based on the corrected initial weight value, and the aforementioned pathological features and their corresponding reference suggestions are output at the same time, so as to provide a more comprehensive diagnosis and treatment reference suggestion for the doctor.

[0018] Optionally, the analysis result output module is further configured to obtain the actual treatment decision input by the user after outputting the reference suggestion, and optimize the postoperative diagnosis and treatment decision tree of thyroid cancer based on the difference between the actual treatment decision and the output reference suggestion.

[0019] By adopting the above technical solution, after outputting the reference suggestion to assist the doctor in diagnosis and treatment, the postoperative diagnosis and treatment decision tree of thyroid cancer can be further optimized according to the actual treatment decision made by the doctor and the difference between the actual treatment decision and the reference suggestion output by the aforementioned analysis model, so as to reduce the error between the reference suggestion output by the analysis model and the actual treatment decision.

[0020] In a second aspect, the present application provides an intelligent auxiliary device for postoperative evaluation and iodine-131 treatment decision of differentiated thyroid cancer, including a memory and a processor. A computer program capable of being loaded and executed by the processor is stored on the memory, and the computer program includes instructions for implementing the functions of the intelligent auxiliary system for postoperative evaluation and iodine-131 treatment decision of differentiated thyroid cancer as described in the first aspect.

[0021] In a third aspect, the present application provides a computer-readable storage medium storing a computer program capable of being loaded and executed by a processor, and the computer program includes instructions for implementing the functions of the intelligent auxiliary system for postoperative evaluation and iodine-131 treatment decision of differentiated thyroid cancer as described in the first aspect.

[0022] In summary, the present application includes the following beneficial technical effects: The present application aims to use large language model technology to extract and summarize key information from the medical record data of thyroid cancer users, and analyze and obtain suggestions for the postoperative management of differentiated thyroid cancer based on existing expert consensus, clinical guidelines, comprehensive postoperative TNM staging, recurrence risk stratification and dynamic assessment system, so as to provide efficient and comprehensive reference opinions for doctors' diagnosis and treatment decisions, optimize the quality of doctors' postoperative diagnosis and treatment of thyroid cancer, and improve the standardization and efficiency of diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0024] Figure 1 It is a structural block diagram of an intelligent assistant for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment disclosed in the embodiments of the present application.

[0025] Explanation of reference numerals: 201, medical record data collection module; 202, pathological feature extraction module; 203, analysis result output module; 204, decision tree optimization module; 205, auxiliary evaluation trigger module. Detailed implementation manners

[0026] The embodiments of the present application disclose an intelligent assistant system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment (hereinafter referred to as the auxiliary diagnosis and treatment system for short), which includes a medical record data collection module 201, a pathological feature extraction module 202, and an analysis result output module 203. Among them, the medical record data collection module 201 is used to receive an auxiliary evaluation instruction, and during the interaction process, obtain and feedback the inquiry questions raised by the user, and extract medical record data from the inquiry questions; among them, the inquiry questions at least include content related to the condition of thyroid cancer and other extended content not related to the condition of thyroid cancer; the medical record data at least includes surgical cases related to thyroid cancer, condition changes, vital signs, and case reports. It is also used to match the obtained medical record data with a preset postoperative diagnosis and treatment decision tree for thyroid cancer based on a preset AI large model, and determine whether there is missing data based on the matching result. If so, an omission prompt is output to prompt the user to supplement the corresponding missing data; among them, the medical record data is divided into clinical data, genomics data, blood biochemical indexes, and imaging data.

[0027] The pathological feature extraction module 202 is used to extract key features from clinical data through a natural language processing model; annotate gene sequencing data and extract mutation and expression features related to thyroid cancer from genomics data; standardize blood biochemical indexes through a machine learning model and identify abnormal features, and extract visual features from imaging data through a deep learning model; map all modal features to a unified feature space, and perform fusion processing on all modal features through a pre-constructed multi-modal fusion model to generate pathological features.

[0028] Among them, the content of the decision nodes corresponding to the post-thyroid cancer diagnosis and treatment decision tree includes at least TNM staging analysis, recurrence risk stratification analysis, and dynamic assessment analysis.

[0029] The analysis result output module 203 is used to traverse the decision nodes included in the pre-constructed post-thyroid cancer diagnosis and treatment decision tree through a preset AI large model based on the decision nodes included in the post-thyroid cancer diagnosis and treatment decision tree, analyze the pathological features, and output reference suggestions. The reference suggestions at least include: the postoperative treatment effect, the determination result of whether the conditions for iodine-131 treatment are met, and the personalized iodine-131 treatment dose range obtained when the conditions for iodine-131 treatment are met.

[0030] It also includes a decision tree optimization module 204, which is used to optimize the post-thyroid cancer diagnosis and treatment decision tree regularly based on the inquiry content.

[0031] In implementation, the medical record data collection module 201 pre-creates an interactive interface for users to upload and input medical record data and interact with the system. The medical record data collection module 201 conducts question-and-answer interactions with users through natural language processing technologies (such as NLP). During the interaction process, it is used to answer the questions raised by users or ask questions to users based on a preset question bank to obtain the feedback content of users. Thus, medical record data is obtained from the questions raised by users and the output feedback content. The medical record data is divided into clinical data (such as TNM staging, pathological reports (such as postoperative pathological examination reports), imaging data (such as ultrasound, CT, PET-CT, diagnostic iodine-131 imaging)), genomics data (such as gene mutations (BRAF, RAS, TERT, etc.), gene expression profiles, molecular typing), blood biochemical indicators (thyroglobulin (Tg), thyroglobulin antibody (TgAb), thyroid stimulating hormone (TSH), thyroid function indicators (TSH, FT3, FT4, etc.)), and other data (such as the surgical medical records of the patient (such as admission records, operation records), changes in the condition, vital signs, etc.).

[0032] In addition, the questions raised by users can be other questions at the non-medical level, such as medical management suggestions, psychological counseling, etc. The AI large model is used to match the answer content for the corresponding questions from a preset answer library based on the questions raised by users and output it to the interactive interface. The decision tree optimization module 204 is used to regularly add decision content according to the obtained inquiry questions and feedback it to the management decision terminal for the management to perform addition, deletion, and modification editing on the decision tree based on the new decision content, so as to realize the optimization of the decision tree.

[0033] The pathological feature extraction module 202 is used to extract features from medical record data in the form of input text, pictures, images, etc. through a pre-constructed analysis model (such as a large AI model). Specifically: extract key features (such as tumor size, lymph node metastasis, etc.) from clinical data by using natural language processing technology, use bioinformatics tools (such as GATK, ANNOVAR) to annotate gene sequencing data in genomic data, and extract mutation and expression features related to thyroid cancer; standardize blood biochemical indicators through rules or machine learning models, and use computer vision technology (such as convolutional neural network CNN) to extract image features (such as tumor morphology, lymph node status, etc.), so as to obtain features of multiple modalities, map the features of the foregoing different modalities to a unified feature space, and then use a multimodal fusion model (such as multimodal transformer, graph neural network GNN) to fuse the multimodal features to generate a comprehensive feature representation, that is, finally obtain pathological features.

[0034] Next, the post-thyroidectomy diagnosis and treatment decision tree is the diagnosis and treatment logic of thyroid cancer sorted out based on existing medical materials such as differentiated thyroid cancer guidelines and expert consensus. The decision tree covers three aspects: TNM staging, recurrence risk stratification, and dynamic assessment. In addition, decision nodes related to multimodal data can be arranged based on the data corresponding to the pathological features in the decision tree. For example, there is a decision node A with the node content: whether the BRAF V600E mutation is detected, and a lower-level node a of decision node A, and the content of lower-level node a is: the tumor has strong invasiveness and may require more aggressive treatment; for example, there is a decision node B with the node content: whether the TERT promoter mutation is detected, and a lower-level node b of decision node B, and the content of lower-level node b is: indicating a higher recurrence risk, and it is recommended to set the postoperative follow-up frequency to x. Another example is a decision node C with the node content: whether the Tg level continues to rise, and the content of lower-level node c is: there may be residual lesions or recurrence... The large AI model is used to traverse the post-thyroidectomy diagnosis decision tree based on the extracted pathological features, and finally output reference suggestions, and the reference suggestions include at least the postoperative treatment effect, the determination result of whether the iodine-131 treatment condition is met, and the iodine-131 treatment dose range obtained when the iodine-131 treatment condition is met.

[0035] In addition, the medical record data collection module 201 is also used to compare the extracted pathological features with the specific content corresponding to the decision nodes in the thyroid cancer diagnosis and treatment decision tree. If there is data that is included in the decision node but does not exist in the current pathological features, then this data is considered to be missing data, and the auxiliary diagnosis and treatment system is used to send a missing data prompt to the user through the interaction interface to prompt the user to supplement the corresponding missing data.

[0036] Optionally, the reference suggestions further include the postoperative follow-up frequency; It further includes an auxiliary evaluation trigger module 205, which is used to record the reference suggestions whenever an auxiliary evaluation instruction is obtained and the corresponding reference suggestions are determined; and trigger the auxiliary evaluation instruction according to the postoperative follow-up frequency included in the reference suggestions.

[0037] In implementation, as can be seen from the foregoing, in the thyroid cancer diagnosis and treatment decision tree, there are decision nodes for determining specific postoperative follow-up frequency values. Therefore, when the analysis model traverses the thyroid cancer diagnosis and treatment decision tree, it will output reference suggestions including the postoperative follow-up frequency. Correspondingly, this application proposes that the auxiliary diagnosis and treatment system is further used to regularly trigger the auxiliary evaluation instruction related to the user according to the corresponding postoperative follow-up frequency, and push the reference suggestions corresponding to the output of the auxiliary evaluation instruction to the corresponding user intelligent terminal to remind the user to return for a follow-up visit regularly.

[0038] Optionally, the analysis result output module 203 is further used to obtain the actual treatment decision input by the user, and optimize the thyroid cancer postoperative diagnosis and treatment decision tree based on the difference between the actual treatment decision and the output reference suggestions.

[0039] In implementation, the physician can input the actual treatment decision based on the reference suggestions output by the analysis model. When the optimization condition is met, the auxiliary diagnosis and treatment system will optimize the thyroid cancer postoperative diagnosis and treatment decision tree. Among them, the determination method of whether the optimization condition is met can be: the physician manually compares the difference between the actual treatment decision and the output reference suggestions and determines whether the optimization condition is met, or the auxiliary diagnosis and treatment system uses the similarity comparison technology to determine the similarity between the actual treatment decision and the output reference suggestions. When the similarity is less than or equal to the preset similarity, it is considered that the optimization condition is met. The specific optimization method can be: determine the content in the reference suggestions output by the analysis model that is inconsistent with the actual treatment decision (hereinafter referred to as the optimization content), determine the decision node corresponding to the optimization content in the thyroid cancer postoperative diagnosis and treatment decision tree (hereinafter referred to as the node to be optimized), and then determine the content in the actual treatment decision corresponding to the optimization content (hereinafter referred to as the correction content), and supplement the correction content to the node content corresponding to the node to be optimized, or replace the node content corresponding to the node to be optimized with the correction content; to update the content of the thyroid cancer postoperative diagnosis and treatment decision tree. In other embodiments, if there is content in the actual treatment decision that is not included in the output reference suggestions (hereinafter referred to as the new content), the physician will be provided with the optimization permission to add a decision node corresponding to the new content in the thyroid cancer postoperative diagnosis and treatment decision tree to update the thyroid cancer postoperative diagnosis and treatment decision tree.

[0040] Optionally, the analysis result output module 203 is also used to determine similar historical pathological features similar to the pathological features from case features stored in historical periods based on a preset weighted similarity calculation method and an initial weight value preset for each modal data whenever a pathological feature is obtained; calculate the difference between the reference suggestions corresponding to the similar historical cases and the reference suggestions obtained by the analysis model for the currently obtained case features, and correct the initial weight value according to the difference, and then re-determine the similar historical pathological features corresponding to the current pathological features and their corresponding reference suggestions according to the corrected initial weight value; The analysis result output module 203 is further configured to output corresponding similar historical pathological features and their corresponding reference suggestions when outputting current pathological features and their corresponding reference suggestions.

[0041] In implementation, whenever a pathological feature is extracted, the analysis result output module 203 will store the pathological feature. At the same time, the analysis result output module 203 is also used to find similar historical pathological features corresponding to the current pathological features from the pathological features stored in the historical period based on a similarity measurement method (such as cosine similarity, Euclidean distance, Jaccard index). Similar historical pathological features refer to pathological features with the highest similarity to the current case features. Since the pathological features in the previous article contain multimodal data, the similarity measurement method here can be a weighted similarity calculation method, that is, an initial weight is assigned to each modality according to the importance of different modalities. After finding similar historical pathological features according to the weighted similarity calculation method corresponding to the initial weights, the analysis result output module 203 will be further used to analyze the differences between the reference suggestions corresponding to the current pathological features obtained by the analysis model and the reference suggestions corresponding to the similar historical pathological features. The specific analysis methods may include: 1. Classification decision difference: If the reference suggestion contains a classification result (such as whether iodine-131 treatment is needed), the classification error (such as 0 / 1 error) is used as the difference evaluation method; 2. Numerical decision difference: If the reference suggestion is a numerical result (such as iodine-131 treatment dose), the mean square error (such as MSE) or absolute error (MAE) can be used as the difference evaluation method.

[0042] Then, the reference suggestions are used to re-adjust the initial weight values of each modality data in the pathological features according to the differences (the initial weight values refer to the weight values assigned to each modality data when using the weighted similarity calculation method to find the similar historical case features). The specific weight adjustment methods may include: 1. Gradient descent method: Using the difference as the loss function, update the weights through the gradient descent method; 2. Heuristic adjustment: Adjust the proportion according to the magnitude of the difference; Repeat the above steps until the weights converge or reach the preset maximum weight number; Then, use the adjusted initial weight values and the weighted similarity calculation method to re-determine the similar historical pathological features and their corresponding reference suggestions, so as to help provide the most similar historical pathological features and their reference suggestions by adjusting the weights; After obtaining the reference suggestions corresponding to the current pathological features through the analysis model subsequently, the corresponding similar historical pathological features and their corresponding reference suggestions can be output simultaneously to provide more reference basis for doctors. In summary, through the dynamic weight adjustment mechanism based on differences, the diagnostic assistance system can automatically optimize the weight values of each modality feature in the similarity calculation, thereby improving the accuracy and personalization of similarity matching. This function can significantly enhance the practicality and intelligent level of the case similarity matching system.

[0043] Optionally, the analysis result output module 203 is further configured to determine whether there is a first association relationship based on the changes in the initial weight values of all modality data before and after adjustment and based on a preset association relationship analysis model, that is, to determine whether there is an association relationship in the adjustment amplitude of the initial weight values corresponding to two or more modality data. If so, store the first association relationship to improve the weight value adjustment efficiency based on the first association relationship between modality data when the weight value needs to be adjusted subsequently. For example, when adjusting the initial weight value of one modality data, adaptively adjust the initial weight value of the modality data that has a first association relationship with this modality data.

[0044] Optionally, the analysis result output module 203 is further configured to analyze the initial weight values corresponding to each modality data in the pathological features under different change trends (i.e., different values) based on each historical stored pathological feature and its corresponding finally adjusted initial weight values of each modality data, and determine the association relationship (hereinafter referred to as the second association relationship) between the specific data value of each modality data and the value of the initial weight value based on a preset association relationship analysis model (such as the causal forest algorithm), record the second association relationship, and when it is necessary to determine the similar historical pathological features for the newly obtained pathological features subsequently, directly adjust the numerical value of the corresponding initial weight value based on the specific data value of each modality data in the newly obtained pathological feature and its corresponding second association relationship, further improving the weight adjustment efficiency.

[0045] Optionally, the present application also pre-constructs an iodine treatment dose determination model. The diagnosis and treatment assistance system is used to regularly collect pathological features within a specified period and the corresponding optimal iodine treatment dose range for each of the foregoing pathological features (the optimal iodine treatment dose range is the iodine treatment dose range included in the actual decision-making suggestions finally input by the physician). Then, the foregoing pathological features are used as inputs, and the corresponding optimal iodine treatment dose range for the pathological features is used as outputs to implement the training of the iodine treatment dose determination model. During the training process, a loss function is predefined, such as mean square error (MSE) or cross-entropy loss, to quantify the gap between the model prediction and the actual treatment dose. At the same time, an optimization algorithm (such as gradient descent method) is used to minimize the loss function, thereby adjusting the parameters of the model. Regularization techniques (such as L1 regularization, L2 regularization or dropout) can also be introduced to prevent the model from overfitting. Therefore, the iodine treatment dose determination model can be used to assist the analysis model to further optimize the output of the optimal iodine treatment dose, so as to make personalized recommendations based on the specific conditions of the user (such as genomic variations, lesion size, biochemical indicators, etc.).

[0046] The embodiment of the present application also discloses an intelligent assistance device for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment. The intelligent assistance device for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment includes a memory and a processor. A computer program that can be loaded and executed by the processor is stored on the memory, and the computer program includes instructions for implementing the functions of the intelligent assistance system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment as described above.

[0047] The embodiment of the present application also discloses a computer-readable storage medium that stores a computer program that can be loaded and executed by the processor, and the computer program includes instructions for implementing the functions of the intelligent assistance system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment as described above. The computer-readable storage medium includes, for example, various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0048] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0049] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting the protection scope of the application. Obviously, the described embodiments are only partial embodiments of the present application, rather than all embodiments. Based on these embodiments, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope to be protected by the present application.

Claims

1. An intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision-making, characterized in that, Including: A medical record data collection module (201) for receiving an auxiliary evaluation instruction and obtaining medical record data; A pathological feature extraction module (202) for extracting pathological features from the medical record data through a preset analysis model; An analysis result output module (203) for analyzing the pathological features through the analysis model and a pre-constructed post-thyroid cancer treatment decision tree, and outputting reference suggestions for a doctor to know the reference suggestions; wherein, the reference suggestions at least include reference suggestions for iodine-131 treatment.

2. The intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision-making according to claim 1, characterized in that The content of the decision nodes corresponding to the pre-constructed post-thyroid cancer treatment decision tree at least includes TNM staging analysis, recurrence risk stratification analysis, and dynamic evaluation analysis; The analysis result output module (203) is used to traverse the decision nodes included in the pre-constructed post-thyroid cancer treatment decision tree through a preset AI large model based on the decision nodes included in the pre-constructed post-thyroid cancer treatment decision tree, analyze the pathological features, and output reference suggestions, and the reference suggestions at least include: postoperative treatment effect, determination result of whether the iodine-131 treatment condition is met, and personalized iodine-131 treatment dose range obtained when the iodine-131 treatment condition is met.

3. The intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision-making according to claim 1, wherein The medical record data collection module (201) is used to perform question-and-answer interaction with a user based on a preset AI large model; it is also used to obtain and feedback the inquiry questions raised by the user during the interaction process, and extract medical record data from the inquiry questions; wherein, the inquiry questions at least include content related to the condition of thyroid cancer and other extended content not related to the condition of thyroid cancer; the medical record data at least includes surgical cases related to thyroid cancer, condition changes, vital signs, and case reports; It also includes a decision tree optimization module (204) for regularly optimizing the post-thyroid cancer treatment decision tree based on the inquiry content.

4. The intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision-making according to claim 3, wherein The medical record data collection module (201) is also used to match the obtained medical record data with the pre-constructed post-thyroid cancer treatment decision tree based on a preset AI large model, and determine whether there is missing data based on the matching result. If so, an omission prompt is output to prompt the user to supplement the corresponding missing data.

5. The intelligent auxiliary system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment according to claim 2, characterized in that, The reference suggestions also include the postoperative follow-up frequency; It also includes an auxiliary evaluation trigger module (205) for recording the reference suggestions whenever an auxiliary evaluation instruction is obtained and the corresponding reference suggestions are determined; And triggering an auxiliary evaluation instruction according to the postoperative follow-up frequency included in the reference suggestions.

6. The intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision-making according to claim 1, wherein The medical record data is divided into clinical data, genomics data, blood biochemical indicators, and imaging data; The pathological feature extraction module (202) is used to extract key features from the clinical data through a natural language processing model; annotate gene sequencing data and extract mutation and expression features related to thyroid cancer from the genomics data; perform standardization processing on blood biochemical indicators through a machine learning model and identify abnormal features; extract visual features from the imaging data through a deep learning model; The pathological feature extraction module (202) is further configured to map all modality features to a unified feature space, and fuse all modality features through a pre-constructed multi-modal fusion model to generate pathological features.

7. The intelligent auxiliary system for the postoperative evaluation of differentiated thyroid cancer and the decision-making of iodine-131 treatment according to claim 6, wherein The analysis result output module (203) is further configured to, whenever pathological features are obtained, determine similar historical pathological features similar to the pathological features from the case features stored in the historical period based on a preset weighted similarity calculation method and the initial weight values preset for each modality data; calculate the difference between the reference suggestions corresponding to the similar historical cases and the reference suggestions analyzed by the analysis model for the currently obtained case features, and correct the initial weight values according to the difference, and then re-determine the similar historical pathological features corresponding to the current pathological features and their corresponding reference suggestions according to the corrected initial weight values; The analysis result output module (203) is further configured to output the corresponding similar historical pathological features and their corresponding reference suggestions when outputting the current pathological features and their corresponding reference suggestions.

8. The intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision-making according to claim 1, wherein The analysis result output module (203) is further configured to, after outputting the reference suggestions, obtain the actual treatment decision input by the user, and optimize the postoperative diagnosis and treatment decision tree of thyroid cancer based on the difference between the actual treatment decision and the output reference suggestions.

9. An intelligent auxiliary device for postoperative evaluation of differentiated thyroid cancer and decision-making of iodine-131 treatment, characterized in that, It includes a memory and a processor. A computer program capable of being loaded and executed by the processor is stored on the memory. The computer program contains instructions for implementing the functions of the intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program capable of being loaded and executed by the processor is stored. The computer program contains instructions for implementing the functions of the intelligent auxiliary system for postoperative evaluation of differentiated thyroid cancer and iodine-131 treatment decision as described in any one of claims 1 to 7.

Images