Cancer medical data analysis method based on large model technology

Abstract

The invention discloses a cancer medical data analysis method based on a large model technology, and relates to the technical field of medical data analysis, and the method comprises the following steps: 1, building a partition database model, carrying out the partition according to the number of hospitals, building a clinical database after the partition is completed, and collecting the clinical data, the method comprises the following steps of: 1, establishing a clinical database, and establishing a preprocessing model to process stored data, and 2, establishing a medical data analysis model after the clinical database is established, and analyzing the stored medical data. For cancer treatment scenes of a plurality of patients, based on an objective phenomenon that curative effects and adverse reactions of different patients may have significant differences even if pathological characteristics of tumors are similar or the same treatment scheme is adopted, a database is constructed to realize unified management of medical data by carrying out convergence statistics on medical data of various regions, and the medical data are stored in a database. And the stored data is analyzed and processed, so that a treatment scheme better matched with the characteristics of the patient is generated in an auxiliary manner.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, specifically a cancer medical data analysis method based on large model technology. Background Technology

[0002] Currently, cancer usually refers to malignant tumors, which are new organisms formed by the abnormal proliferation and differentiation of body cells under the influence of various carcinogenic factors. They are characterized by invading surrounding tissues and metastasizing through the blood or lymphatic system, thereby destroying the normal structure and function of organs and seriously threatening life and health. In clinical treatment, even when patients receive the same treatment plan or course, significant differences in efficacy and adverse reactions can still occur. These differences are often related to factors such as the biological heterogeneity of the tumor itself, differences in molecular subtyping, the state of the immune microenvironment, and individual physiological factors. Due to the lack of effective means to predict and screen treatment plans before treatment, some patients need to repeatedly adjust or change their treatment strategies during the course of treatment, leading to prolonged treatment cycles. In some cases, the failure to achieve effective control may even cause them to miss the optimal intervention window, ultimately increasing the risk of disease progression and treatment failure. Summary of the Invention

[0003] The purpose of this invention is to provide a cancer medical data analysis method based on large model technology to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a cancer medical data analysis method based on large model technology, comprising the following steps: Step 1: Build a partitioned database model. Partition the database according to the number of hospitals. After partitioning, build a clinical database to collect clinical data, build a preprocessing model to process the stored data, and build a correlation model to link the data of each region. Step 2: After the clinical database is built, a medical data analysis model is developed to analyze the stored medical data. After the analysis is completed, a treatment plan generation model is developed to generate treatment plans and treatment prediction plans. The treatment plans provide auxiliary references during the diagnosis process, and the treatment prediction plans predict the efficacy of treatment. After generation, the model is trained. Before training, a qualified value is set. If the final result during training is within the qualified value, the training is considered qualified and the training is stopped. The qualified value is dynamically adjusted according to the structure of the simulation training. Step 3: Collect feedback on the effects of treatment, conduct centralized analysis and processing after the feedback, formulate a dynamic adjustment model after analysis and processing, adjust the analysis method of treatment, and then adjust the treatment plan generation model based on the adjusted medical data analysis model. Machine learning methods are used to evaluate the performance of a model using test data that was not used in training, including accuracy and precision, store and label the trained data, and adjust predictions for new data.

[0005] Preferably, in step one, the partitioned data model is divided into regions, namely domestic regions and foreign regions. After the regions are divided, a clinical database is constructed, and a domestic database and a foreign database are built within the clinical database. The domestic database collects and stores medical data for the region, while the foreign database collects and stores medical data for the region and acquires relevant medical data from abroad, which is then translated and stored.

[0006] Preferably, in step one, the preprocessing model preprocesses the collected medical data, and the preprocessing includes the following: (1) Integrate information such as electronic medical records, transcriptomics data, genomics data, treatment records, and survival follow-up; (2) Standardization alignment is performed, which includes time alignment, term alignment and format conversion. Time alignment matches treatment events and laboratory results along the time axis to construct longitudinal time series data. Term alignment uses a standard coding system to eliminate semantic ambiguity. Format conversion extracts key information from unstructured text using NLP technology and converts it into structured fields. (3) Data deduplication and cleaning, including patient unique identification, logical verification and outlier handling. Patient unique identification is achieved by merging duplicate records using unique keys such as ID card number and hospitalization number to avoid data fragmentation of the same patient. Logical verification checks the rationality of the data. Outlier handling truncates or marks extreme values ​​and judges whether they are measurement errors based on clinical knowledge.

[0007] Preferably, in step one, the association model performs association processing on domestic and foreign databases. After the association is completed, a series adjustment design is performed to enable the sharing of medical data within each regional database. When data is queried across regions, minimum access permissions are set in advance. When the data transmitted in the query is sensitive data, encryption and desensitization processing is performed. When querying sensitive data, manual online query authorization is performed. When querying data, the key feature data of the medical data is sent to the association model, and the association model is used to query and obtain data from each region.

[0008] Preferably, in step two, the medical data analysis model re-analyzes the preprocessed medical data. During the analysis, a preprocessing analysis plan is formulated. Once the preprocessing analysis plan is established, an online expert mode is used for assisted analysis. The specific analysis steps are as follows: (1) The efficacy evaluation analysis includes descriptive statistical analysis and survival analysis. Among them, the descriptive statistical analysis ① calculates the statistics of core efficacy indicators, including the objective response rate, disease control rate, progression-free survival, and mean, median and confidence interval of overall survival. ② Comparative analysis compares the efficacy differences of different treatment regimens through t test / analysis of variance and chi-square test. Among them, the survival analysis ① draws the survival curve intuitively, compares the survival differences of different groups, and judges whether the difference is statistically significant through Log-rank test. ② Cox proportional hazards regression model: under the premise of controlling for confounding factors, analyzes the impact of a certain factor on patient survival and calculates the hazard ratio. (2) Analyze adverse reactions, including incidence statistics and grading analysis, correlation analysis and time-series correlation analysis, then perform gene expression difference analysis on transcriptome data, and perform mutation detection and annotation on genomic data.

[0009] Preferably, in step two, the treatment plan generation model classifies the results of the analysis of the treatment plan according to the level of treatment effect and treatment difficulty, and divides them into three levels: Level 1, Level 2, and Level 3. Level 1 indicates the most severe result, Level 2 indicates the moderate result, and Level 3 indicates the mildest result. After the level classification is completed, a treatment plan is formulated, which includes a first plan, a second plan, and a third plan. Level 1 difficulty corresponds to the first plan, Level 2 difficulty corresponds to the second plan, and Level 3 difficulty corresponds to the third plan. When relevant data is retrieved, a treatment plan of the corresponding level is generated based on the historical records of medical data. The final treatment plan serves as an auxiliary material for treatment and is used in conjunction with expert treatment. Treatment prediction plans are developed to predict the patient's condition in conjunction with the treatment plan. These prediction plans include a first prediction plan, a second prediction plan, and a third prediction plan. The first prediction plan is used to predict the first treatment plan, the second prediction plan is used to predict the second treatment plan, and the third prediction plan is used to predict the third treatment plan. The first prediction plan focuses on simultaneous prediction during and after treatment and develops a long-term health management plan. The second prediction plan focuses on predicting the treatment period first and then the post-treatment period and develops a medium- to long-term health management plan. The third prediction plan routinely predicts the medical period and post-treatment period and develops a medium- to short-term health management plan.

[0010] Preferably, in step three, problems and effects arising during the implementation of the treatment plan are collected and analyzed. When the collected problems reach half a month, the problems are analyzed in a concentrated manner. When the analysis result is judged to be serious, the problems are analyzed and processed immediately after three days. When the analysis result is judged to be moderate, the problems are analyzed and processed immediately after seven days. When the analysis result is judged to be mild, the problems are analyzed and processed in a concentrated manner after half a month. When the analysis result is judged to be normal, the analysis and processing are carried out in the conventional manner.

[0011] Preferably, in step three, the dynamic adjustment model is formulated based on the analyzed settings to develop an adjustment treatment plan. During the implementation of the adjustment treatment plan, experts are connected to participate in the analysis and formulation. The pre-treatment analysis plan is adjusted through the adjustment treatment plan, thereby adjusting the treatment plan. The adjustment treatment plan includes a first-level adjustment plan, a second-level adjustment plan, and a third-level adjustment plan. The first-level adjustment plan has the largest adjustment range, the second-level adjustment plan has a medium adjustment range, and the third-level adjustment plan has the smallest adjustment range. Each time an adjustment is made, the data at the time of the adjustment is collected centrally, and the effect of the adjustment is collected and formed into an adjustment database for centralized storage. Later, the stored historical data is analyzed, and the adjustment plan is then optimized.

[0012] Compared with the prior art, the beneficial effects of the present invention are: This invention addresses the clinical phenomenon that "even with similar tumor pathology and the same treatment plan, the efficacy of treatment can still vary significantly among different patients" in treatment scenarios involving multiple cancer patients. It aggregates and statistically analyzes medical data from various regions to build a database, enabling unified storage and standardized management of medical data. This data is then analyzed to form usable decision-making data. In subsequent treatment, doctors can quickly retrieve and filter matching data based on patient information to generate treatment plans more tailored to the patient's characteristics. This provides data support and efficacy prediction references for clinical medication and plan adjustments, thereby improving diagnostic and treatment efficiency, reducing redundant trials and unnecessary procedures, and shortening the treatment decision-making cycle. Attached Figure Description

[0013] Figure 1 A schematic diagram of the method flow is provided for embodiments of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Please see Figure 1 This invention provides a technical solution: a method for cancer medical data analysis based on large model technology, comprising the following steps: Step 1: Build a partitioned database model, partition the database according to the number of hospitals, build a clinical database to collect clinical data, build a preprocessing model to process the stored data, and build a correlation model to link the data of each partition. Step 2: After the clinical database is built, a medical data analysis model is developed to analyze the stored medical data. After the analysis is completed, a treatment plan generation model is developed to generate treatment plans and treatment prediction plans, and then the models are trained. Step 3: Collect feedback on the effects of treatment, conduct centralized analysis and processing after the feedback, formulate a dynamic adjustment model after analysis and processing, adjust the analysis method of treatment, and then adjust the treatment plan generation model based on the adjusted medical data analysis model. Machine learning methods are used to evaluate the performance of a model using test data that was not used in training, including accuracy and precision, store and label the trained data, and adjust predictions for new data.

[0016] In step one, the partitioned data model is divided into regions, namely domestic regions and foreign regions. After the regions are divided, a clinical database is constructed, and a domestic database and a foreign database are built inside the clinical database. The domestic database collects and stores medical data of the region, while the foreign database collects and stores medical data of the region. The foreign database also acquires relevant medical data from abroad, and then translates and stores the acquired data. Medical data sources include public databases and hospital systems within the region, as well as authorized interviews with relevant experts and attending physicians on a regular basis. When rare cancer medical data is collected, it is specially labeled, transfer learning, meta-learning, and data augmentation are applied. The medical data includes omics data, clinical data, follow-up data, and other data. Omics data includes genomics (DNA mutations and copy number variations) and transcriptomics (RNA expression), etc. Clinical data includes patients (age, gender), treatment reports (stage, grade, subtype), treatment history, and laboratory test results, etc. Follow-up data includes overall survival, progression-free survival, and time to recurrence, etc. Other data includes patient-reported outcomes and medical texts (physician notes, surgical reports).

[0017] In step one, the preprocessing model preprocesses the collected medical data, and the preprocessing includes the following: (1) Integrate information such as electronic medical records, transcriptomics data, genomics data, treatment records, and survival follow-up; (2) Standardization alignment is performed, which includes time alignment, terminology alignment and format conversion. Time alignment matches treatment events (such as chemotherapy cycles) and laboratory results along the time axis to construct longitudinal time series data. Terminology uses a standardized coding system (such as ICD-10 diagnostic coding, LOINC laboratory test coding, NCIt cancer terminology set) to eliminate semantic ambiguity. Format conversion extracts key information (such as tumor staging, molecular subtyping, transcriptome data analysis results, genomic data analysis results) from unstructured text (such as pathology reports) using NLP technology and converts it into structured fields. (3) Data deduplication and cleaning, including patient unique identification, logical verification and outlier handling. The patient unique identification is merged with duplicate records by unique keys such as ID card number and hospitalization number to avoid data fragmentation of the same patient. The logical verification checks the rationality of the data (such as whether the age and diagnosis time are contradictory, and whether the laboratory results are outside the physiological range). The outlier handling cuts off or marks extreme values ​​(such as abnormally large tumor volume) and judges whether it is a measurement error in combination with clinical knowledge. (4) Feature extraction and transformation include clinical features and genomic features. Clinical features include ① continuous variables such as tumor size and Ki-67 index, which can retain the original value or be binned; ② categorical variables such as cancer stage and molecular subtype, which need to be one-hot encoded or tag encoded. The genomic features include ① mutation status, which encodes gene mutations as binary variables, and ② expression level, which normalizes or performs differential expression analysis on RNA-seq data. (5) Feature selection and dimensionality reduction include ① preliminary screening by using variance threshold (e.g., removing features with variance <0.1) or correlation analysis (e.g., removing features with low correlation to the target variable), ② stepwise selection of the optimal feature subset by using recursive feature elimination (RFE) combined with model performance (e.g., random forest AUC), ③ automatic screening of important features using L1 regularization (Lasso), or dimensionality reduction of high-dimensional gene data by using PCA / t-SNE; (6) Process the data according to the time series, including time window division and time series feature extraction, and then perform data balancing and standardization. In the process, class imbalance is handled first, and then data standardization and normalization are performed. In addition, identical data is merged by creating a separate folder, placing two identical data items inside the same folder, and marking the folder. When abnormal data is encountered, it is preprocessed manually, and the preprocessing method is retained so that similar abnormal data can be processed in the same way in the future.

[0018] In step one, the association model performs association processing on domestic and foreign databases. After the association is completed, a series adjustment design is carried out to enable the sharing of medical data within the databases in each region. When data is queried across regions, minimum access permissions are set in advance. When the data transmitted in the query is sensitive data, encryption and desensitization processing is performed. When querying sensitive data, manual online query authorization is required. When querying data, the key feature data of the medical data is sent to the association model, and the association model is used to query and obtain data from each region. The system sets limits on the number of access attempts. If an attempt to access and obtain treatment data exceeds three times within a month, subsequent attempts will be subject to a manual online query to retrieve relevant medical data and supporting evidence from previous access attempts. If no issues are found during the query, authorization to continue the query is granted. If any issues are found during the query, authorization to continue the query is suspended. If the number of queries exceeds twelve within six months, a manual remote query will be immediately performed, linking the query to the local medical system. If no issues are found during the query, authorization to continue will be granted. If issues are found during the query, authorization to continue the query is suspended.

[0019] In step two, the medical data analysis model re-analyzes the preprocessed medical data. During the analysis, a preprocessing analysis plan is formulated. Once the preprocessing analysis plan is established, an online expert mode is used for assisted analysis. The specific analysis steps are as follows: (1) The efficacy evaluation analysis includes descriptive statistical analysis and survival analysis. The descriptive statistical analysis ① calculates the statistical measures of the core efficacy indicators, including the objective response rate, disease control rate, progression-free survival, and mean, median and confidence interval of overall survival. The comparative analysis compares the efficacy differences of different treatment regimens through t test / analysis of variance and chi-square test. The survival analysis ① draws the survival curves intuitively, compares the survival differences of different groups, and judges whether the differences are statistically significant through Log-rank test. The Cox proportional hazards regression model: under the premise of controlling for confounding factors (such as age, gender, tumor stage), it analyzes the impact of a certain factor (such as targeted drug use, gene mutation status) on patient survival and calculates the hazard ratio. (2) Analyze adverse reactions, including incidence statistics and grading analysis, correlation analysis and time-series correlation analysis, and then perform gene expression difference analysis on transcriptome data and mutation detection and annotation on genomic data; (3) Multimodal fusion includes early fusion and late fusion. In early fusion, features and gene expression data are concatenated and then input into the model. In late fusion, image and molecular models are trained separately, and finally the results are weighted and fused to predict the results. (4) Identify gene mutations (such as TP53) that are associated with specific treatment features (such as mitotic counts), and construct a composite prognostic model by combining treatment image features with molecular data (such as PD-L1 expression); After the analysis plan is formulated, a training model is set up to train the analysis plan. The results of each simulation training are collected, and a training database is set up to collect the data in a centralized form. The data is then classified according to half a month, one month, three months, half a year, and year. When the stored data exceeds half a month, it is automatically classified into the one-month category. When the stored data exceeds one month, it is automatically classified into the three-month category. When the stored data exceeds three months, it is automatically classified into the six-month category. When the stored data exceeds six months, it is automatically classified into the year category.

[0020] In step two, the treatment plan generation model categorizes the results of the analysis of the treatment plan according to the levels of treatment effect and difficulty, into three levels: Level 1, Level 2, and Level 3. Level 1 indicates the most severe result, Level 2 indicates a moderate result, and Level 3 indicates the mildest result. After the level classification is completed, a treatment plan is formulated, which includes Plan 1, Plan 2, and Plan 3. Plan 1 corresponds to Level 1 difficulty, Plan 2 corresponds to Level 2 difficulty, and Plan 3 corresponds to Level 3 difficulty. When relevant data is retrieved, a treatment plan corresponding to the level is generated based on the historical records of medical data. The final treatment plan serves as an auxiliary material for treatment and is used in conjunction with expert treatment. Treatment prediction plans are developed to predict the patient's condition in conjunction with the treatment plan. These prediction plans include a first prediction plan, a second prediction plan, and a third prediction plan. The first prediction plan is used to predict the first treatment plan, the second prediction plan is used to predict the second treatment plan, and the third prediction plan is used to predict the third treatment plan. The first prediction plan focuses on simultaneous prediction during and after treatment and develops a long-term health management plan. The second prediction plan focuses on predicting the treatment period first and then the post-treatment period and develops a medium- to long-term health management plan. The third prediction plan routinely predicts the medical period and post-treatment period and develops a medium- to short-term health management plan.

[0021] In step three, problems and effects arising during the implementation of the treatment plan are collected and analyzed. When the collected problems reach half a month, they are analyzed in a concentrated manner. If the analysis result is serious, the problems are analyzed and processed immediately after three days. If the analysis result is moderate, the problems are analyzed and processed immediately after seven days. If the analysis result is mild, the problems are analyzed and processed in a concentrated manner after half a month. If the analysis result is normal, the analysis and processing are carried out in the conventional manner.

[0022] In step three, the model is dynamically adjusted. Based on the analyzed settings, an adjustment plan is formulated. During the implementation of the adjustment plan, experts are connected to participate in the analysis and formulation. The pretreatment analysis plan is adjusted through the adjustment plan, thereby adjusting the treatment plan. The adjustment plan includes a first-level adjustment plan, a second-level adjustment plan, and a third-level adjustment plan. The first-level adjustment plan has the largest adjustment range, the second-level adjustment plan has a medium adjustment range, and the third-level adjustment plan has the smallest adjustment range. Each time an adjustment is made, the data at the time of the adjustment is collected centrally, and the effect of the adjustment is collected and formed into an adjustment database for centralized storage. Later, the stored historical data is analyzed, and the adjustment plan is then optimized.

[0023] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0024] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing cancer medical data based on large model technology, characterized by Includes the following steps: Step 1: Build a partitioned database model, partition the database according to the number of hospitals, build a clinical database to collect clinical data, build a preprocessing model to process the stored data, and build a correlation model to link the data of each partition. Step 2: After the clinical database is built, a medical data analysis model is developed to analyze the stored medical data. After the analysis is completed, a treatment plan generation model is developed to generate a treatment prediction plan. The treatment prediction plan is used to train the efficacy of treatment. Before training, a qualified value is set. When the final result during training is within the qualified value, it is considered qualified, and the training operation is stopped. The qualified value is dynamically adjusted according to the structure of the simulation training. Step 3: Collect feedback on the effects of treatment, conduct centralized analysis and processing after the feedback, formulate a dynamic adjustment model after analysis and processing, adjust the analysis method of treatment, and then adjust the treatment plan generation model based on the adjusted medical data analysis model. Machine learning methods are used to evaluate the performance of a model using test data that was not used in training, including accuracy and precision, store and label the trained data, and adjust predictions for new data.

2. The cancer medical data analysis method based on large model technology according to claim 1, characterized in that: In step one, the partitioned data model is divided into regions, namely domestic regions and foreign regions. After the regions are divided, a clinical database is constructed, and a domestic database and a foreign database are built within the clinical database. The domestic database collects and stores medical data for the region, while the foreign database collects and stores medical data for the region and acquires relevant medical data from abroad, which is then translated and stored.

3. The cancer medical data analysis method based on large model technology according to claim 2, characterized in that: In step one, the preprocessing model preprocesses the collected medical data, and the preprocessing includes the following: (1) Integrate information such as electronic medical records, transcriptomics data, genomics data, treatment records, and survival follow-up; (2) Standardization alignment is performed, which includes time alignment, term alignment and format conversion. Time alignment matches treatment events and laboratory results along the time axis to construct longitudinal time series data. Term alignment uses a standard coding system to eliminate semantic ambiguity. Format conversion extracts key information from unstructured text using NLP technology and converts it into structured fields. (3) Data deduplication and cleaning, including patient unique identification, logical verification and outlier handling. Patient unique identification is achieved by merging duplicate records using unique keys such as ID card number and hospitalization number to avoid data fragmentation of the same patient. Logical verification checks the rationality of the data. Outlier handling truncates or marks extreme values ​​and judges whether they are measurement errors based on clinical knowledge.

4. The cancer medical data analysis method based on large model technology according to claim 3, characterized in that: In step one, the association model performs association processing on domestic and foreign databases. After the association is completed, a series adjustment design is carried out to enable the sharing of medical data within the databases in different regions. When data is queried across regions, minimum access permissions are set in advance. When the data being queried is sensitive, it is encrypted and desensitized. When querying sensitive data, manual online query authorization is required. When querying data, the key feature data of the medical data is sent to the association model, and the association model is used to query and obtain data from various regions.

5. The cancer medical data analysis method based on large model technology according to claim 4, characterized in that: In step two, the medical data analysis model re-analyzes the preprocessed medical data. During the analysis, a preprocessing analysis plan is formulated. Once the preprocessing analysis plan is established, an online expert mode is used for assisted analysis. The specific analysis steps are as follows: (1) The efficacy evaluation analysis includes descriptive statistical analysis and survival analysis. Among them, the descriptive statistical analysis ① calculates the statistics of core efficacy indicators, including the objective response rate, disease control rate, progression-free survival, and mean, median and confidence interval of overall survival. ② Comparative analysis compares the efficacy differences of different treatment regimens through t test / analysis of variance and chi-square test. Among them, the survival analysis ① draws the survival curve intuitively, compares the survival differences of different groups, and judges whether the difference is statistically significant through Log-rank test. ② Cox proportional hazards regression model: under the premise of controlling for confounding factors, analyzes the impact of a certain factor on patient survival and calculates the hazard ratio. (2) Analyze adverse reactions, including incidence statistics and grading analysis, correlation analysis and time-series correlation analysis, then perform gene expression difference analysis on transcriptome data, and perform mutation detection and annotation on genomic data.

6. The cancer medical data analysis method based on large model technology according to claim 5, characterized in that: In step two, the treatment plan generation model categorizes the results of the analysis of the treatment plan according to the levels of treatment effect and difficulty, into three levels: Level 1, Level 2, and Level 3. Level 1 indicates the most severe result, Level 2 indicates a moderate result, and Level 3 indicates the mildest result. After the level classification is completed, a treatment plan is formulated, which includes Plan 1, Plan 2, and Plan 3. Plan 1 corresponds to Level 1 difficulty, Plan 2 corresponds to Level 2 difficulty, and Plan 3 corresponds to Level 3 difficulty. When relevant data is retrieved, a treatment plan corresponding to the level is generated based on the historical records of medical data. The final treatment plan serves as an auxiliary material for treatment and is used in conjunction with expert treatment. Treatment prediction plans are developed to predict the patient's condition in conjunction with the treatment plan. These plans include a first prediction plan, a second prediction plan, and a third prediction plan. The first prediction plan is used to predict the first treatment plan, the second prediction plan is used to predict the second treatment plan, and the third prediction plan is used to predict the third treatment plan. The first prediction plan focuses on simultaneous prediction during and after treatment and develops a long-term health management plan. The second prediction plan first focuses on prediction during treatment and then prediction after treatment and develops a medium- to long-term health management plan. The third prediction plan routinely predicts during and after treatment and develops a medium- to short-term health management plan.

7. The cancer medical data analysis method based on large model technology according to claim 6, characterized in that: In step three, problems and effects arising during the implementation of the treatment plan are collected and analyzed. When the collected problems reach half a month, they are analyzed in a concentrated manner. If the analysis result is serious, the problems are analyzed and processed immediately after three days. If the analysis result is moderate, the problems are analyzed and processed immediately after seven days. If the analysis result is mild, the problems are analyzed and processed in a concentrated manner after half a month. If the analysis result is normal, the analysis and processing are carried out in the conventional manner.

8. The cancer medical data analysis method based on large model technology according to claim 7, characterized in that: In step three, the model is dynamically adjusted. Based on the analyzed settings, an adjustment plan is formulated. During the implementation of the adjustment plan, experts are connected to participate in the analysis and formulation. The pretreatment analysis plan is adjusted through the adjustment plan, thereby adjusting the treatment plan. The adjustment plan includes a first-level adjustment plan, a second-level adjustment plan, and a third-level adjustment plan. The first-level adjustment plan has the largest adjustment range, the second-level adjustment plan has a medium adjustment range, and the third-level adjustment plan has the smallest adjustment range. Each time an adjustment is made, the data at the time of the adjustment is collected centrally, and the effect of the adjustment is collected and formed into an adjustment database for centralized storage. Later, the stored historical data is analyzed, and the adjustment plan is then optimized.

Images