Postoperative weight loss health risk assessment system for weight loss surgery based on cloud computing

Abstract

The invention relates to an artificial intelligence technology, and provides a cloud computing-based postoperative weight loss health risk assessment system for a weight loss surgery, which comprises a data acquisition module, a data transmission and storage module, a feature extraction module, a health risk assessment module and a health intervention module, according to the system, semantic stratified analysis is carried out on multi-source data of a postoperative patient by constructing a self-adaptive heuristic-BERT model; on this basis, risk level classification is carried out based on postoperative feature data of the patient through an E-GRU-BiLST model, then an inter-class similarity matrix is introduced, and classification precision is optimized in combination with multi-class cross entropy loss and focus loss; by combining the technology and a cloud computing platform, the accuracy, dynamic adaptability and operability of postoperative health management are improved, the patient is assisted to recover health more efficiently, and the long-term health risk is reduced.

Description

Technical Field

[0001] The present invention relates to artificial intelligence technology, in particular to a health risk assessment system for weight loss after bariatric surgery based on cloud computing. Background Art

[0002] With the increasing severity of the obesity problem, bariatric surgery has become an important means to reduce morbid obesity. Although bariatric surgery can significantly reduce body weight and improve metabolic function, patients still face many health risks after surgery. Therefore, postoperative health management and risk assessment are crucial for the long-term health of patients. However, traditional postoperative weight loss health risk assessment still has the following deficiencies: First, traditional systems use decentralized or single-institution internal server storage for data storage, resulting in poor data compatibility between different institutions and different devices. At the same time, the security of remote data transmission in some systems is insufficient, and there may be data leakage and privacy risks. Second, traditional systems ignore the individual health status of patients and the dynamic changes in the postoperative recovery process, often lacking the ability to adapt to different patients and being difficult to adjust risk assessment strategies according to individual characteristics, resulting in limited accuracy of assessment results. Finally, most existing systems only provide basic health management suggestions and lack the ability to dynamically adjust health management plans according to patients' personal data. Summary of the Invention

[0003] The present invention provides a health risk assessment system for weight loss after bariatric surgery based on cloud computing, aiming to improve the intelligent level of postoperative health management, enhance the accuracy and real-time performance of health risk assessment, and thus optimize the postoperative recovery effect of patients. In the feature extraction module of the present invention, in-depth analysis is performed on multi-source data of postoperative patients to construct an adaptive heuristic-BERT model, which can efficiently extract key features in the data of postoperative patients, enhance context relevance, and improve the accuracy and comprehensiveness of data processing. On this basis, the system further constructs an E-GRU-BiLST model for health risk level classification, and by introducing an inter-class similarity matrix and combining multi-class cross-entropy loss and focal loss, the calculation accuracy of health risk scores is optimized. Through the combination of the above technologies, the present invention significantly improves the accuracy, dynamic adaptability, and operability of postoperative health risk assessment, providing intelligent support for the long-term health management of patients.

[0004] The present invention provides a health risk assessment system for weight loss after bariatric surgery based on cloud computing, which includes a data acquisition module, a data transmission and storage module, a feature extraction module, a health risk assessment module, and a health intervention module;

[0005] The data acquisition module collects postoperative weight loss data, including physical sign monitoring data, nutritional intake data, exercise data, lifestyle data, and postoperative follow-up records;

[0006] A data transmission and storage module constructs a cloud computing platform, uploads the post - weight - loss surgery data to the cloud using a security protocol, and stores it in the cloud computing platform; fills in missing values, cleans the data, and converts the format of the post - weight - loss surgery data to generate pre - processed post - weight - loss surgery data;

[0007] A feature extraction module introduces an adaptive heuristic information enhancement strategy to improve the ant colony algorithm, constructs an improved ant colony algorithm, uses the improved ant colony algorithm to optimize the BERT model, constructs an adaptive heuristic - BERT model, uses the adaptive heuristic - BERT model to process the pre - processed post - weight - loss surgery data, conducts semantic hierarchical analysis, enhances context relevance, and generates weight - loss feature data;

[0008] The adaptive heuristic - BERT model includes a BERT model and a Transformer network;

[0009] A health risk assessment module constructs an E - GRU - BiLST model, processes the weight - loss feature data through the E - GRU - BiLST model, and generates a health risk score and risk level classification;

[0010] A health intervention module conducts personalized health management, diet adjustment, exercise plan modification, health education, and psychological support according to the health risk score and risk level classification, ultimately reducing health risks, promoting patients' postoperative recovery, and improving the quality of life.

[0011] Furthermore, the process of the feature extraction module generating weight - loss feature data specifically includes the following steps:

[0012] Step S1: Text tokenization processing: Tokenize the pre - processed post - weight - loss surgery data to generate a tokenization result, and convert the tokenization result into a token sequence to obtain text token data;

[0013] Step S2: Input token sequence: Input the text token data into the embedding layer of the BERT model to generate an initial embedding representation and obtain token embedding data;

[0014] Step S3: Masked language processing: Randomly mask 25% of the tokens in the token embedding data to obtain masked token positions, predict the masked token positions, and generate MLM semantic feature data;

[0015] Step S4: Next sentence prediction processing: Divide the text token data into sentence pairs, calculate the continuity relationship of each pair of sentences, and obtain NSP feature data;

[0016] Step S5: Adaptive Ant Colony Optimization of Transformer: Optimize the Transformer network using an improved ant colony algorithm to construct an adaptive ant colony-Transformer network. Input the MLM semantic feature data and NSP feature data into the adaptive ant colony-Transformer network to generate context semantic embedding features;

[0017] Step S6: Feature Output: Extract the overall global semantic features and local semantic features from the context semantic embedding features to generate weight-reduced feature data.

[0018] Furthermore, step S5 specifically includes the following steps:

[0019] Step S51: Initialize Network Weights: Use the MLM semantic feature data and NSP feature data as inputs to initialize the attention weights of the adaptive ant colony-Transformer network;

[0020] Step S52: Assign Initial Pheromones: According to the attention weights, assign initial pheromone values to each attention head of the multi-head Attention in the adaptive ant colony-Transformer network to obtain a pheromone distribution map;

[0021] Step S53: Optimize the Path Selection Strategy: Combine the pheromone distribution map, simulate ants' path selection between different attention heads through the ant colony algorithm, and introduce a semantic similarity factor and a patient health priority factor to dynamically adjust the heuristic information in the ant colony algorithm, optimizing the path selection strategy of the ant colony algorithm to generate path pheromone accumulation data and contribution degree data;

[0022] Step S54: Dynamic Weight Adjustment: Dynamically adjust the attention weights of the attention heads according to the path pheromone accumulation data and contribution degree data, weaken the path weights of the low-attention heads in the contribution degree data, and strengthen the path weights of the high-contribution attention heads in the contribution degree data to obtain attention weight distribution data. Based on the attention weight distribution data, generate context semantic embedding features.

[0023] Furthermore, the process of constructing the E-GRU-BiLST model specifically includes: establishing a GRU-BiLST model, introducing inter-class similarity, and combining a multi-class cross-entropy loss and a focal loss joint strategy to optimize the learnable parameters of the GRU-BiLST model to construct an E-GRU-BiLST model.

[0024] Adopting the above solution, the beneficial effects of the present invention are as follows:

[0025] The present invention realizes the precise assessment and personalized management of the health risks after bariatric surgery, and improves the comprehensive data collection and intelligent analysis capabilities of the postoperative data; by constructing an adaptive heuristic-BERT model, key features are extracted from the multi-source data of postoperative patients; this technology enhances the context relevance of the data, enabling the system to more accurately identify the core factors affecting postoperative health conditions, and solving the problems of insufficient semantic understanding and insufficient information mining depth in the traditional system during the data processing process; through efficient information extraction and optimized data processing mechanisms, the present invention ensures the high-quality input of health risk assessment data, providing a solid data foundation for subsequent risk prediction and health intervention;

[0026] The present invention further constructs an E-GRU-BiLST health risk assessment model, which performs risk scoring on the postoperative characteristic data of patients through deep learning and classifies the health risk levels; based on the GRU-BiLSTM architecture, this model introduces an inter-class similarity matrix, and combines multi-class cross-entropy loss and focal loss to optimize the calculation accuracy of health risks, improving the system's ability to distinguish different risk categories and prediction accuracy; it solves the problem that traditional risk assessment methods rely on fixed rules and are difficult to dynamically adapt to individual differences, enhances the system's ability to accurately identify individual health conditions, and enables it to meet the needs of patients in different postoperative recovery stages; the risk assessment system of the present invention not only has higher prediction accuracy, but also can provide more targeted health management suggestions for different risk levels, avoiding a one-size-fits-all health management approach and improving the personalized level of postoperative management;

[0027] The overall architecture of the present invention is based on a cloud computing platform, realizing real-time data transmission, storage and dynamic analysis, and improving the scalability and cross-platform compatibility of the system; combining the above technical features, the present invention effectively improves the health management level of postoperative patients, reduces potential health risks, and provides precise, efficient and intelligent technical support for the long-term health maintenance of patients. Brief Description of the Drawings

[0028] Figure 1 It is a schematic diagram of a health risk assessment system model for postoperative weight loss after bariatric surgery based on cloud computing proposed by the present invention;

[0029] Figure 2 It is a schematic diagram of the effect of the learnable parameters of the optimized GRU-BiLST model in Examples 6 and 7. Detailed Embodiments

[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] Example 1. According to Figure 1 , the present invention provides a health risk assessment system for weight loss after bariatric surgery based on cloud computing. The system includes a data collection module, a data transmission and storage module, a feature extraction module, a health risk assessment module, and a health intervention module;

[0032] The data collection module collects data after bariatric surgery, including physical sign monitoring data, nutritional intake data, exercise data, lifestyle data, and postoperative follow-up records;

[0033] The data transmission and storage module constructs a cloud computing platform, uploads the data after bariatric surgery to the cloud using a security protocol, and stores it in the cloud computing platform; fills in missing values, cleans the data, and converts the format of the data after bariatric surgery to generate preprocessed data after bariatric surgery;

[0034] The feature extraction module introduces an adaptive heuristic information enhancement strategy to improve the ant colony algorithm, constructs an improved ant colony algorithm, uses the improved ant colony algorithm to optimize the BERT model, constructs an adaptive heuristic - BERT model, uses the adaptive heuristic - BERT model to process the preprocessed data after bariatric surgery, performs semantic hierarchical analysis, enhances context relevance, and generates weight loss feature data;

[0035] The adaptive heuristic - BERT model includes a BERT model and a Transformer network;

[0036] The health risk assessment module constructs an E - GRU - BiLST model, processes the weight loss feature data through the E - GRU - BiLST model, and generates a health risk score and risk level classification;

[0037] The health intervention module conducts personalized health management, diet adjustment, exercise plan change, health education, and psychological support according to the health risk score and risk level classification, ultimately reducing health risks, promoting postoperative recovery of patients, and improving the quality of life.

[0038] Example 2. This example is based on Example 1. In this example, the physical sign monitoring data, nutritional intake data, exercise data, and lifestyle data specifically include:

[0039] The physical sign monitoring data includes weight change, heart rate, blood sugar, and blood pressure;

[0040] The nutritional intake data includes diet records and nutritional component analysis;

[0041] The exercise data includes the number of steps, activity duration, and calories burned;

[0042] The lifestyle data includes sleep quality and mental state.

[0043] Example 3: This example is based on Example 2. In this example, the process of the feature extraction module generating weight loss feature data specifically includes the following steps:

[0044] Step S1: Text tokenization processing: Tokenize the preprocessed post-weight loss surgery data to generate a tokenization result, and convert the tokenization result into a token sequence to obtain text token data;

[0045] Step S2: Input token sequence: Input the text token data into the embedding layer of the BERT model to generate an initial embedding representation and obtain token embedding data;

[0046] Step S3: Masked language processing: Randomly mask 25% of the tokens in the token embedding data to obtain the masked token positions, predict the masked token positions, and generate MLM semantic feature data;

[0047] Step S4: Next sentence prediction processing: Divide the text token data into sentence pairs, calculate the continuity relationship of each pair of sentences, and obtain NSP feature data;

[0048] Step S5: Adaptive ant colony optimization of Transformer: Use an improved ant colony algorithm to optimize the Transformer network, construct an adaptive ant colony-Transformer network, input the MLM semantic feature data and NSP feature data into the adaptive ant colony-Transformer network, and generate context semantic embedding features;

[0049] Step S6: Feature output: Extract the overall global semantic features and local semantic features from the context semantic embedding features to generate weight loss feature data.

[0050] Example 4: This example is based on Example 2. In this example, the process of the feature extraction module generating weight loss feature data specifically includes the following steps:

[0051] Step C1: Tokenize the preprocessed post-weight loss surgery data to generate a tokenization result, and convert the tokenization result into a token sequence to obtain text token data;

[0052] Step C2: Input the text token data into the embedding layer of the BERT model to generate an initial embedding representation and obtain token embedding data;

[0053] Step C3: Randomly mask 25% of the tokens in the masked token-embedded data to obtain the masked token positions, predict the masked token positions, and generate MLM semantic feature data;

[0054] Step C4: Divide the text token data into sentence pairs, calculate the continuity relationship of each pair of sentences, and obtain NSP feature data;

[0055] Step C5: Process the MLM semantic feature data and NSP feature data through a Transformer network to generate context semantic embedding features;

[0056] Step S6: Feature output: Extract the overall global semantic features and local semantic features from the context semantic embedding features to generate weight reduction feature data.

[0057] Example 5, this example is based on Example 3. In this example, Step S5 specifically includes the following steps:

[0058] Step S51: Initialize network weights: Use the MLM semantic feature data and NSP feature data as inputs to initialize the attention weights of the adaptive ant colony-Transformer network;

[0059] Step S52: Allocate initial pheromones: According to the attention weights, allocate initial pheromone values to each attention head of the multi-head Attention in the adaptive ant colony-Transformer network to obtain a pheromone distribution map;

[0060] Step S53: Optimize the path selection strategy: Combine the pheromone distribution map, simulate the path selection of ants between different attention heads through the ant colony algorithm, and introduce a semantic similarity factor and a patient health priority factor to dynamically adjust the heuristic information in the ant colony algorithm, optimize the path selection strategy of the ant colony algorithm, and generate path pheromone accumulation data and contribution degree data. The formula used is as follows:

[0061] ;

[0062] Where, and represent the attention head index, represents the path to of the dynamic optimization heuristic information, , and represent the weights of the adjustment factors, represents the attention head and semantic similarity, which measures the correlation between attention heads through cosine similarity; represents the patient health priority function; Represents traditional heuristic information;

[0063] Path selection probability formula:

[0064] ;

[0065] Wherein, Represents the probability that the ant moves from attention head to attention head The probability of, Represents the path to The pheromone concentration of, Represents the weight coefficient of the pheromone concentration, Represents the weight coefficient of the heuristic information; Represents the target path, that is, the target attention head, Represents the path to The pheromone concentration of, Represents the path to The heuristic information of, Represents the set of valid target paths, Represents the normalization term;

[0066] Step S54: Dynamic weight adjustment: According to the path pheromone accumulation data and contribution degree data, dynamically adjust the attention weights of the attention heads, weaken the path weights of the low attention heads in the contribution degree data, and strengthen the path weights of the high contribution attention heads in the contribution degree data to obtain the attention weight distribution data. Based on the attention weight distribution data, generate context semantic embedding features. The formula used is as follows:

[0067] Updated path pheromone formula:

[0068] ;

[0069] Wherein, Represents the iteration index, Represents the path to At the The pheromone concentration of the iteration, Represents the path to At the The pheromone concentration of the iteration, Represents the pheromone evaporation rate, Represents the path to The pheromone increment during the iteration process.

[0070] Example six, according toFigure 2 , this embodiment is based on Embodiment 5. In this embodiment, the process of constructing the E-GRU-BiLST model specifically includes: establishing a GRU-BiLST model, introducing the similarity between categories, and optimizing the learnable parameters of the GRU-BiLST model by combining the multi-class cross-entropy loss and the focal loss joint strategy to construct the E-GRU-BiLST model. The formula used is as follows:

[0071] ;

[0072] Among them, represents the total loss function, represents the total number of samples in the dataset, that is, the total amount of weight loss feature data, represents the total number of categories in the target classification task, that is, the total number of health risk level categories, represents the sample index, represents the category index, that is, the true health risk level, represents the predicted category index, that is, the health risk level predicted by the model, represents the weight of the category similarity matrix, and the value range is [0, 1], represents the sample For the category true label, that is, one-hot encoding, represents the sample belongs to the category predicted probability; and represent the weight parameters, represents the category weight, represents the focusing parameter, represents the focus factor.

[0073] Embodiment 7, according to Figure 2 , this embodiment is based on Embodiment 5. In this embodiment, the process of constructing the E-GRU-BiLST model specifically includes: establishing a GRU-BiLST model, and optimizing the learnable parameters of the GRU-BiLST model by combining the cross-entropy loss and the focal loss joint strategy to construct the E-GRU-BiLST model.

[0074] The above describes the present invention and its implementation manners. This description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto; all in all, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A cloud computing-based bariatric surgery postoperative weight loss health risk assessment system, comprising a data transmission and storage module, wherein the data transmission and storage module generates pre-processed bariatric surgery data; characterized in that: The system also includes a feature extraction module and a health risk assessment module; Feature extraction module, building an adaptive heuristic-BERT model, using the adaptive heuristic-BERT model to process the post-weight loss data and generate weight loss feature data; The health risk assessment module builds the E-GRU-BiLST model, processes the weight loss feature data through the E-GRU-BiLST model, and generates a health risk score and risk level classification.

2. According to claim 1, a cloud computing-based weight loss health risk assessment system for postoperative weight loss surgery, characterized in that: The adaptive heuristic-BERT model includes the BERT model and the Transformer network.

3. According to claim 2, a cloud computing-based bariatric surgery postoperative weight loss health risk assessment system is characterized by: The feature extraction module generates weight loss feature data, which specifically includes the following steps: Step S1: Segment the pre-processed post-weight loss surgery data to obtain text labelled data; Step S2: Input text tag data into the BERT model to obtain tag embedding data; Step S3: randomly mask the marker embedding data to generate MLM semantic feature data; Step S4: Divide the text tag data into sentence pairs to obtain NSP feature data; Step S5: construct an adaptive ant colony-Transformer network, input the MLM semantic feature data and the NSP feature data into the adaptive ant colony-Transformer network, and generate contextual semantic embedding features; Step S6: Extract semantic features from contextual semantic embedding features to generate weight loss feature data.

4. The cloud computing-based post-operative weight loss health risk assessment system for bariatric surgery according to claim 3, characterized in that: Step S5 specifically includes the following steps: Step S51: using the MLM semantic feature data and the NSP feature data as input to initialize the attention weights of the adaptive ant colony-Transformer network; Step S52: assigning an initial pheromone value to each attention head of the multi-head Attention in the adaptive ant colony-Transformer network according to the attention weight, and obtaining a pheromone distribution map; Step S53: generating path pheromone accumulation data and contribution degree data in combination with the pheromone distribution map; Step S54: Generate contextual semantic embedding features based on the path pheromone accumulation data and contribution degree data.

5. The cloud computing-based post-operative weight loss health risk assessment system for bariatric surgery according to claim 4, characterized in that: Step S53 specifically includes: combining the pheromone distribution map, simulating ants to select paths between different attention heads through the ant colony algorithm, introducing semantic similarity factors and patient health priority factors, dynamically adjusting the heuristic information in the ant colony algorithm, optimizing the path selection strategy of the ant colony algorithm, and generating path pheromone accumulation data and contribution degree data.

6. The cloud computing-based post-operative weight loss health risk assessment system for bariatric surgery according to claim 4, characterized in that: Step S54 specifically includes: dynamically adjusting the attention weight of the attention head according to the path pheromone accumulation data and the contribution degree data, obtaining the attention weight distribution data, and generating contextual semantic embedding features based on the attention weight distribution data.

7. The cloud computing-based post-operative weight loss health risk assessment system for bariatric surgery according to claim 1, characterized in that: The process of building the E-GRU-BiLST model specifically includes: establishing the GRU-BiLST model, introducing the similarity between categories, and optimizing the learnable parameters of the GRU-BiLST model by combining the multi-classification cross entropy loss and focal loss joint strategy to build the E-GRU-BiLST model.

Images