Multimodal chronic disease risk prediction method based on AIGC technology

Abstract

The present invention relates to the technical field of chronic disease risk prediction. The present invention relates to a multimodal chronic disease risk prediction method based on AIGC technology. It comprises the following steps: S1, collecting historical diagnosis data, historical body shape data and daily physiological data of users; S2, using AIGC technology to extract symptom characteristics from historical diagnosis data; the present invention can more comprehensively capture the potential signs and development trends of chronic diseases by integrating multimodal information such as symptom difference feature data, image data of related parts and physiological fluctuation feature data. Data of different modalities complement each other. For example, symptom difference feature data can provide details about symptom changes and disease progression, image data can intuitively display the pathological conditions of body parts, and physiological fluctuation feature data reflects the dynamic changes of body functions. This multi-dimensional information fusion enables the risk prediction model to analyze chronic diseases from multiple angles.

Description

Technical Field

[0001] The present invention relates to the technical field of chronic disease risk prediction, and in particular to a multimodal chronic disease risk prediction method based on AIGC technology. Background Art

[0002] Chronic disease management is an important topic in today's medical field. It involves the prevention, diagnosis, treatment and control of long-term diseases such as cardiovascular disease, diabetes, chronic respiratory diseases, etc. Its main function is to identify potential risk factors for disease in advance, with the aim of achieving early intervention and personalized treatment to reduce the incidence of chronic diseases, improve the quality of life of patients and reduce the burden on the medical system.

[0003] In some places with advanced treatment levels, such as tertiary hospitals or advanced medical institutions, artificial intelligence technology has gradually been combined with modern medicine. For example, AIGC technology (Artificial Intelligence Generated Content), which is artificial intelligence generated content, refers to the use of artificial intelligence algorithms, such as generative adversarial networks (GANs), variational autoencoders (VAEs), and large language models (LLMs) in deep learning. In addition, chronic diseases are a long-term treatment process. During the long-term treatment process, medical institutions can obtain more complete physical data and corresponding treatment and diagnosis data of patients through some wearable sensor devices or regular physical examinations, and establish chronic disease databases and prediction models corresponding to patients by introducing artificial intelligence technologies such as AIGC to monitor the physical condition of patients.

[0004] However, the occurrence and development of chronic diseases is a complex and long-term process, which is affected by a combination of factors. In the diagnosis of common diseases, doctors often use a limited number of common dimensions to judge the patient's condition, which makes it difficult to accurately capture the potential risk signals of chronic diseases. Therefore, a multimodal chronic disease risk prediction method based on AIGC technology is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a multimodal chronic disease risk prediction method based on AIGC technology to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, a multimodal chronic disease risk prediction method based on AIGC technology is provided, which includes the following steps:

[0007] S1, collect the user's historical diagnosis data, historical body data and daily physiological data;

[0008] S2. Use AIGC technology to extract disease characteristics from historical diagnosis data, and analyze disease characteristics of the same part according to the difference in changes to obtain disease difference feature data between disease characteristics;

[0009] S3, combining the historical posture data with the historical diagnosis data for adjacent time periods, and then analyzing the relevant parts based on the disease characteristics, and assigning the disease characteristics to the relevant parts of the image data in the historical posture data according to the analysis results;

[0010] S4, extracting effective data from daily physiological data based on historical body posture data, then calculating development trends based on the effective physiological data, extracting abnormal fluctuations from the effective physiological data, and summarizing them into physiological fluctuation characteristic data;

[0011] S5. Collect chronic disease feature data to establish an AIGC model, perform multimodal fusion and prediction on the disease difference feature data, relevant part image data, and physiological fluctuation feature data, and set weights based on the importance of the representative features of chronic diseases, and then perform risk prediction scoring;

[0012] S6. Set the risk level corresponding to different risk prediction scores, and then obtain the risk level of the user and chronic disease based on the risk prediction score.

[0013] As a further improvement of the present technical solution, S1 obtains the user's historical diagnosis data from the hospital information storage port, obtains the user's historical body shape data through the user's physical examination and ID photo, and then obtains the user's daily physiological data through smart watches and wearable devices.

[0014] As a further improvement of the technical solution, the steps of S2 are as follows:

[0015] S2.1. Use AIGC technology to extract disease characteristics from historical diagnosis data, and obtain disease descriptions and examination results in diagnosis records as disease characteristics;

[0016] S2.2. Classify the symptom characteristics according to the described parts, and then analyze the symptom characteristics of the same part according to the change differences to obtain the symptom difference characteristic data of the same part.

[0017] As a further improvement of the technical solution, the steps of S3 are as follows:

[0018] S3.1. Obtain the upload time of each historical diagnosis data and the upload time of each historical posture data, and then combine each historical posture data with the most adjacent historical diagnosis data according to the upload time;

[0019] S3.2. Analyze the relevant parts of the disease characteristics at different upload times to obtain the relevant parts of the body that will be affected by the disease characteristics, and then extract images from the historical body data based on the obtained relevant parts, so as to obtain the relevant part image data corresponding to the disease characteristics at different upload times.

[0020] As a further improvement of the present technical solution, when S3.1 combines historical body shape data with historical diagnosis data, if the most adjacent historical diagnosis data of multiple historical body shape data are the same, the historical body shape data closest to the upload time of the historical diagnosis data is retained, and the remaining historical body shape data are deleted, so that one historical body shape data corresponds to one historical diagnosis data.

[0021] As a further improvement of the technical solution, the step of S4 is as follows:

[0022] S4.1. Perform normal analysis on daily physiological data according to historical body posture data, set a valid range according to the analysis results, and then effectively screen the daily physiological data according to the valid range to obtain valid physiological data within the valid range in the daily physiological data;

[0023] S4.2. Calculate the development trend according to the upload time of the valid physiological data to obtain the physiological development trend, then set the fluctuation abnormality threshold, extract the abnormal fluctuation of the valid physiological data in combination with the fluctuation abnormality threshold, and then summarize the physiological development trend and the abnormal fluctuation into physiological fluctuation characteristic data.

[0024] As a further improvement of the technical solution, the step of S5 is as follows:

[0025] S5.1. Collect chronic disease characteristic data to establish an AIGC model, and then combine the disease difference characteristic data, relevant part image data, and physiological fluctuation characteristic data for multimodal fusion and prediction;

[0026] S5.2. Extract the manifestation characteristics of each chronic disease, and obtain the importance of each phenotypic characteristic to the diagnosis of the chronic disease. Then set weights for different manifestation characteristics based on the importance, and then perform risk prediction scoring.

[0027] As a further improvement of the present technical solution, S5.2 simultaneously classifies the manifestation characteristics of chronic diseases according to historical diagnostic data, historical physical data and daily physiological data, and the weight of the manifestation characteristics represents the weight of the historical diagnostic data, historical physical data and daily physiological data in the risk prediction score of this chronic disease.

[0028] As a further improvement of the present technical solution, the S6 simultaneously displays the proportion of each feature of the historical diagnosis data, historical body shape data and daily physiological data in the chronic disease risk prediction score.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. In the multimodal chronic disease risk prediction method based on AIGC technology, by integrating multimodal information such as symptom difference feature data, image data of related parts, and physiological fluctuation feature data, it can more comprehensively capture the potential signs and development trends of chronic diseases. Data of different modalities complement each other. For example, symptom difference feature data can provide details about symptom changes and disease progression, image data can intuitively display the pathological conditions of body parts, and physiological fluctuation feature data reflects the dynamic changes of body functions. This multi-dimensional information fusion enables the risk prediction model to analyze chronic diseases from multiple angles, greatly improving the accuracy of prediction.

[0031] 2. In the multimodal chronic disease risk prediction method based on AIGC technology, the model established using AIGC technology has strong learning and pattern recognition capabilities. It can automatically learn complex features and patterns from a large amount of multimodal data, and dig out the deep connections between data without manually designing complex feature extraction rules. By continuously training and optimizing the AIGC model, it can adapt to the characteristics and data patterns of different types of chronic diseases, thereby achieving accurate prediction of chronic disease risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is the overall flow chart of the present invention;

[0033] Figure 2 A flowchart of the present invention for obtaining the difference characteristic data of symptoms in the same part;

[0034] Figure 3 A flowchart of the present invention for obtaining image data of relevant parts corresponding to disease characteristics at different upload times;

[0035] Figure 4 This is a flowchart of the present invention that combines physiological development trends with abnormal fluctuations to form physiological fluctuation characteristic data;

[0036] Figure 5 A flowchart of the risk prediction scoring process of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] See also Figure 1 - Figure 5 As shown, the purpose of this embodiment is to provide a multimodal chronic disease risk prediction method based on AIGC technology, comprising the following steps:

[0039] S1, collect the user's historical diagnosis data, historical body data and daily physiological data;

[0040] S1 obtains the user's historical diagnosis data from the hospital information storage port, obtains the user's historical body data through the user's physical examination and ID photo, and then obtains the user's daily physiological data through smart watches and wearable devices. The specific steps are as follows:

[0041] Obtain historical diagnostic data: Cooperate with the database management department of the hospital information system to obtain data access rights. According to the predetermined data field requirements, write SQL query statements or use relevant data extraction tools to extract the user's historical diagnostic data from the hospital database. The historical diagnostic data includes disease name, diagnosis date, diagnosing doctor, and related examination results.

[0042] Obtain historical body data: Cooperate with the physical examination center of the hospital to obtain the user's physical examination report. The physical examination data includes height, weight, body fat percentage, body dimensions, body assessment results, etc. For paper reports, OCR technology can be used to convert them into editable text data before extraction. Then, by obtaining the user's ID photo, image processing technology and human posture estimation algorithm are used to perform body analysis and detect the positions of key nodes of the human body in the photo;

[0043] Daily physiological data: Connect with smart watches and wearable devices, and set a reasonable data collection frequency based on the needs of chronic disease risk prediction and the performance of the equipment. For example, for heart rate data, it can be set to collect every 5 minutes. At the same time, the wearable device temporarily stores the collected data in the local device memory, and then transmits the data to the connected cloud server at a certain time interval.

[0044] S2. Use AIGC technology to extract disease characteristics from historical diagnosis data, and analyze disease characteristics of the same part according to the difference in changes to obtain disease difference feature data between disease characteristics;

[0045] The steps of S2 are as follows:

[0046] S2.1. Use AIGC technology to extract disease characteristics from historical diagnosis data, and obtain disease descriptions and examination results in diagnosis records as disease characteristics;

[0047] S2.2, classify the symptom characteristics according to the described parts, and then analyze the symptom characteristics of the same part according to the change differences to obtain the symptom difference characteristic data of the same part. The specific steps are as follows:

[0048] Model training: Use the pre-trained AIGC model and fine-tune it according to the specific disease feature extraction task. For example, when using BERT to extract features from disease description text, you can use the annotated medical text data for further training based on the pre-trained BERT, and then prepare the annotated data, that is, manually annotate a part of the disease description text to annotate the key features related to the disease. For the description of heart disease symptoms, annotate key features such as "chest pain", "palpitations", and "dyspnea";

[0049] Feature extraction: Input the cleaned and preprocessed disease description text into the trained language model BERT to obtain the feature vector representation of the text. For image inspection results, input the image data into the CNN model. After convolution, pooling and other operations, obtain the feature map or feature vector of the image. These features can reflect the lesion characteristics, organ morphology and other information in the image;

[0050] Part classification and feature integration: Based on medical knowledge and the distribution of common diseases, a part classification system is determined, which can be divided into head, neck, chest, abdomen, limbs and other part categories. Then, the extracted disease features are annotated by part classification, and the disease features of the same part are integrated;

[0051] Feature comparison and difference calculation: Multiple disease feature instances of the same part are calculated according to the comparison method;

[0052] For image features, there are chest X-ray results of multiple patients. SSIM is used to calculate the similarity between every two images, and the similarity value is used as the difference measure of image features;

[0053] For text-based descriptions of symptoms, such as "the patient described a mild cough the first time and a severe cough with sputum the second time", the two texts are converted into vectors and their difference is calculated using cosine similarity.

[0054] S3, combining the historical posture data with the historical diagnosis data for adjacent time periods, and then analyzing the relevant parts based on the disease characteristics, and assigning the disease characteristics to the relevant parts of the image data in the historical posture data according to the analysis results;

[0055] The steps for S3 are as follows:

[0056] S3.1. Obtain the upload time of each historical diagnosis data and the upload time of each historical posture data, and then combine each historical posture data with the most adjacent historical diagnosis data according to the upload time;

[0057] S3.1 When combining historical posture data with historical diagnosis data, if the historical diagnosis data that is most adjacent to multiple historical posture data is the same, the historical posture data that is closest to the upload time of the historical diagnosis data is retained, and the remaining historical posture data are deleted, so that one historical posture data corresponds to one historical diagnosis data.

[0058] S3.2. Analyze the relevant parts of the disease characteristics at different upload times, obtain the relevant parts of the body that the disease characteristics will affect, and then extract images from the historical body data based on the obtained relevant parts, so as to obtain the relevant part image data corresponding to the disease characteristics at different upload times. The specific steps are as follows:

[0059] Symptom characteristic data collection: Extract symptom characteristic information at different upload times from historical diagnosis data, collect and organize medical knowledge about the association between different symptoms and human body parts. For example, we know that heart disease usually affects the chest and surrounding areas, which may manifest as chest pain, palpitations, etc., while arthritis mainly affects the joints, causing joint pain, swelling, etc. We organize this medical knowledge into a knowledge base that associates symptoms with body parts, so as to facilitate subsequent analysis of the body parts of symptom characteristics.

[0060] Symptom feature classification and body part labeling: For each symptom feature, natural language processing technology is used to search for matching keywords in the symptom part knowledge base. If the symptom feature is complex or contains multiple symptoms, it can be split into multiple keywords for matching. Based on the results of keyword matching, the body parts that may be affected by each symptom feature are labeled;

[0061] Filtering body data: Based on the part information obtained from the analysis of the parts related to the disease characteristics, filter out the records related to these parts in the historical body data, and then store the filtered body data in a new data set for subsequent image extraction operations;

[0062] Image extraction and association: For the body data records corresponding to the relevant parts that have been screened out, the image data therein is extracted, and the extracted images are associated with the corresponding disease characteristics. Relevant information of the disease characteristics can be added to the metadata of the image data, or an association table can be created to record the correspondence between the image and the disease characteristics.

[0063] S4, extracting effective data from daily physiological data based on historical body posture data, then calculating development trends based on the effective physiological data, extracting abnormal fluctuations from the effective physiological data, and summarizing them into physiological fluctuation characteristic data;

[0064] The steps of S4 are as follows:

[0065] S4.1. Perform normal analysis on daily physiological data based on historical body posture data, set a valid range based on the analysis results, and then effectively screen the daily physiological data based on the valid range to obtain valid physiological data within the valid range in the daily physiological data. The specific steps are as follows:

[0066] Feature selection and extraction: Select features that may be related to daily physiological data from historical body data, quantify and standardize the selected body features for subsequent analysis and modeling. For height and weight, the body mass index can be calculated as a comprehensive feature, and then linear regression is used to establish the relationship between heart rate (daily physiological data) and weight (historical body data);

[0067] Based on the prediction results of the model: Use the established association model to predict the historical body shape data to obtain the corresponding predicted value of daily physiological data. For example, for each individual's weight data, the corresponding heart rate prediction value is predicted by the above linear regression model. According to the predicted value and the actual daily physiological data, the error range or confidence interval is calculated, and then the confidence interval or error range is used as a reference for the effective range.

[0068] Data screening: For each daily physiological data record, traverse it in chronological order or individually, check whether the physiological indicator value corresponding to each data point is within its valid range, and store the screened valid physiological data.

[0069] S4.2. Calculate the development trend according to the upload time of the valid physiological data, obtain the physiological development trend, then set the fluctuation abnormality threshold, extract the abnormal fluctuation of the valid physiological data in combination with the fluctuation abnormality threshold, and then summarize the physiological development trend and the abnormal fluctuation into the physiological fluctuation characteristic data. The specific steps are as follows:

[0070] Calculate the development trend: According to the characteristics of physiological data and the purpose of analysis, select the appropriate time series analysis method to calculate the effective physiological data and obtain the development trend of physiological data over time. For example, use the moving average method to calculate the development trend of heart rate, calculate the moving average values ​​at different time points, and draw the moving average curve to intuitively observe the trend change of heart rate. If linear regression trend analysis is used, after obtaining the linear regression equation of heart rate and time, the slope can be used to determine whether the heart rate is rising, falling, or maintaining a stable trend, as well as the size of the trend;

[0071] Setting abnormal fluctuation threshold: Statistical analysis is performed on valid physiological data to calculate basic statistics, such as mean, standard deviation, median, quartile, etc. These statistics can reflect the central tendency and dispersion of the data, provide a reference for setting the threshold, and then determine the abnormal fluctuation threshold based on the characteristics of the data and actual needs;

[0072] Abnormal fluctuation detection: For each valid physiological data point, the detection is performed according to the set abnormal fluctuation threshold, and the value of the data point is compared with the threshold to determine whether it is an abnormal fluctuation. The detected abnormal fluctuation data point is marked or stored in a special abnormal data set;

[0073] Summary of physiological fluctuation characteristic data: Integrate the calculated physiological development trend information and the results of abnormal fluctuation detection so that the information is matched in time and individual dimensions, so as to be accurately summarized into physiological fluctuation characteristic data.

[0074] S5. Collect chronic disease feature data to establish an AIGC model, perform multimodal fusion and prediction on the disease difference feature data, relevant part image data, and physiological fluctuation feature data, and set weights based on the importance of the representative features of chronic diseases, and then perform risk prediction scoring;

[0075] The steps of S5 are as follows:

[0076] S5.1. Collect chronic disease characteristic data to establish an AIGC model, and then combine the disease difference characteristic data, relevant part image data, and physiological fluctuation characteristic data for multimodal fusion and prediction;

[0077] Data related to chronic diseases are collected from multiple channels including hospital information systems, electronic medical records, wearable devices, and questionnaires.

[0078] S5.2. Extract the manifestation characteristics of each chronic disease, and obtain the importance of each phenotypic characteristic to the diagnosis of the chronic disease. Then set weights for different manifestation characteristics based on the importance, and then perform risk prediction scoring.

[0079] S5.2 also classifies the manifestation characteristics of chronic diseases according to historical diagnostic data, historical physical data, and daily physiological data. The weight of the manifestation characteristics represents the weight of the historical diagnostic data, historical physical data, and daily physiological data in the risk prediction score of the chronic disease. The multimodal fusion formula is as follows:

[0080] ;

[0081] in, They are different modalities, namely, the disease difference feature vector, image feature vector, and physiological fluctuation feature vector. n1, n2, and n3 are the dimensions of different feature vectors. It is a fused feature vector formed by sequentially concatenating these three vectors, and its dimension is n1+n2+n3;

[0082] ;

[0083] Among them, w1, w2, and w3 represent the weights of different modal feature vectors during fusion, which are used to adjust the importance of different modal features. The weighted feature vector More emphasis is placed on modal features with higher importance;

[0084] ;

[0085] Among them, y i For the i The prediction results of the model for chronic diseases, w i is the weight of the model during fusion, N is the total number of models, y Z The final fusion prediction result is obtained by weighted summation of the prediction results of each model, which integrates the information of different modal models;

[0086] The formula for extracting features and analyzing importance of chronic disease manifestations is as follows:

[0087] ;

[0088] Among them, C is the number of categories, the sample set is in node m, P k,m is the proportion of the k-th class samples in the total samples at node m, G m It is used to measure the impurity of sample m at the node. The smaller the value, the purer the sample and the better the classification effect.

[0089] When feature j is selected to split node m, two child nodes m1 and m2 are obtained after the split. s The ratio of the number of samples in (S=1,2) to the number of samples in the parent node m is p s,m , child node m s The Gini index is G m,s , then the Gini index reduction of the feature when the node is split is The formula is as follows:

[0090] ;

[0091] in, It is the degree of reduction of the Gini index when the feature is split at node m, reflecting the contribution of the feature to the improvement of classification purity. The larger the value, the better the effect of the feature when splitting at this node.

[0092] When there are T decision trees in the random forest, the reduction in the Gini index of feature j in the tth decision tree is (calculated by the above node splitting), then the importance score of the feature in the random forest is I j The formula is as follows:

[0093] ;

[0094] Where T is the number of decision trees in the random forest, I j The importance score of feature j in the entire random forest is obtained by averaging the reduction of the Gini index of the feature in each tree, reflecting the overall importance of the feature for classification in the random forest model;

[0095] The formula for weight setting and risk prediction score is as follows:

[0096] ;

[0097] Where M is the total number of performance characteristics, I j is the importance score of feature j, which is obtained by the importance score formula in the random forest above. is the sum of all feature importance scores, A j represents the relative importance weight of feature j in risk prediction. After normalization, the sum of all feature weights is 1. j Between 0 and 1;

[0098] ;

[0099] Among them, p is the probability of a patient suffering from a chronic disease that the model initially predicts based on multimodal data, and A j is the weight of feature j, f j is the actual value of the jth feature in a specific sample, which is the basis for measuring the impact of this feature on the risk of chronic diseases. j Therefore, even if the weight of the feature is fixed, each patient will eventually get a different risk prediction score due to the difference in their own feature values. S is the risk prediction score obtained after comprehensively considering the model's initial prediction probability and the weights and actual values ​​of each performance feature. The higher the score, the greater the risk of chronic disease. It is used to quantitatively assess the patient's chronic disease risk.

[0100] S6. Set the risk level corresponding to different risk prediction scores, and then obtain the risk level of the user and chronic disease based on the risk prediction score.

[0101] The risk prediction score is between 0 and 1. You can set 0-0.5 as low risk, 0.-0.7 as medium risk, and above 0.7 as high risk.

[0102] S6 also displays the proportion of each feature of historical diagnostic data, historical body shape data and daily physiological data in the chronic disease risk prediction score.

[0103] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A multimodal chronic disease risk prediction method based on AIGC technology, characterized by: The following steps are involved: S1, collect the user's historical diagnosis data, historical body data and daily physiological data; S2. Use AIGC technology to extract disease characteristics from historical diagnosis data, and analyze disease characteristics of the same part according to the difference in changes to obtain disease difference feature data between disease characteristics; S3, combining the historical posture data with the historical diagnosis data for adjacent time periods, and then analyzing the relevant parts based on the disease characteristics, and assigning the disease characteristics to the relevant parts of the image data in the historical posture data according to the analysis results; S4, extracting effective data from daily physiological data based on historical body posture data, then calculating development trends based on the effective physiological data, extracting abnormal fluctuations from the effective physiological data, and summarizing them into physiological fluctuation characteristic data; S5. Collect chronic disease feature data to establish an AIGC model, perform multimodal fusion and prediction on the disease difference feature data, relevant part image data, and physiological fluctuation feature data, and set weights based on the importance of the representative features of chronic diseases, and then perform risk prediction scoring; S6. Set the risk level corresponding to different risk prediction scores, and then obtain the risk level of the user and the chronic disease based on the risk prediction score; The steps of S2 are as follows: S2.

1. Use AIGC technology to extract disease characteristics from historical diagnosis data, and obtain disease descriptions and examination results in diagnosis records as disease characteristics; S2.2, classifying the symptom characteristics according to the described parts, and then analyzing the symptom characteristics of the same part according to the change differences, to obtain the symptom difference characteristic data of the same part; The steps of S3 are as follows: S3.

1. Obtain the upload time of each historical diagnosis data and the upload time of each historical posture data, and then combine each historical posture data with the most adjacent historical diagnosis data according to the upload time; S3.2, analyzing the relevant parts of the disease characteristics at different upload times, obtaining the relevant parts of the body that the disease characteristics will affect, and then extracting images from the historical body data based on the obtained relevant parts, so as to obtain the image data of the relevant parts corresponding to the disease characteristics at different upload times; In the above S3.1, when combining the historical posture data with the historical diagnosis data, if the historical diagnosis data that are most adjacent to the multiple historical posture data are the same, the historical posture data that is closest to the upload time of the historical diagnosis data is retained, and the remaining historical posture data are deleted, so that one historical posture data corresponds to one historical diagnosis data; The steps of S4 are as follows: S4.

1. Perform normal analysis on daily physiological data according to historical body posture data, set a valid range according to the analysis results, and then effectively screen the daily physiological data according to the valid range to obtain valid physiological data within the valid range in the daily physiological data; S4.

2. Calculate the development trend according to the upload time of the valid physiological data to obtain the physiological development trend, then set the fluctuation abnormality threshold, extract the abnormal fluctuation of the valid physiological data in combination with the fluctuation abnormality threshold, and then summarize the physiological development trend and the abnormal fluctuation into physiological fluctuation characteristic data.

2. The multimodal chronic disease risk prediction method based on AIGC technology according to claim 1, characterized in that: The S1 obtains the user's historical diagnosis data from the hospital information storage port, obtains the user's historical body shape data through the user's physical examination and ID photo, and then obtains the user's daily physiological data through smart watches and wearable devices.

3. The multimodal chronic disease risk prediction method based on AIGC technology according to claim 1, characterized in that: The steps of S5 are as follows: S5.

1. Collect chronic disease characteristic data to establish an AIGC model, and then combine the disease difference characteristic data, relevant part image data, and physiological fluctuation characteristic data for multimodal fusion and prediction; S5.

2. Extract the manifestation characteristics of each chronic disease, and obtain the importance of each phenotypic characteristic to the diagnosis of the chronic disease. Then set weights for different manifestation characteristics based on the importance, and then perform risk prediction scoring.

4. The multimodal chronic disease risk prediction method based on AIGC technology according to claim 3 is characterized by: S5.2 also classifies the manifestation characteristics of chronic diseases according to historical diagnostic data, historical physical data and daily physiological data. The weight of the manifestation characteristics represents the weight of the historical diagnostic data, historical physical data and daily physiological data in the chronic disease risk prediction score.

5. The multimodal chronic disease risk prediction method based on AIGC technology according to claim 1 is characterized in that: The S6 simultaneously displays the proportion of each feature of the historical diagnosis data, historical body shape data and daily physiological data in the chronic disease risk prediction score.

Images