Nutrition analysis and diagnosis and treatment decision support system based on artificial intelligence

Abstract

The invention discloses a nutrition analysis and diagnosis and treatment decision support system based on artificial intelligence, and the system comprises a data preprocessing module which is used for carrying out the formatting and standardization processing of collected diet pictures and health record information; the dietary image segmentation module is used for carrying out edge detection and segmentation on the standardized dietary image; the dietary feature extraction module is used for respectively extracting dietary picture features and health record features through a substream neural network; the feature fusion module is used for fusing the dietary picture features and the health record features; the feature optimization module is used for optimizing a substream neural network structure and weight parameters by adopting an ant algorithm; the nutritional ingredient analysis module is used for analyzing the nutritional ingredient condition of the user based on the fused joint features; and the personalized diagnosis and treatment decision module is used for generating a personalized nutrition analysis report and diagnosis and treatment suggestions in combination with the user health archive and the nutritional ingredient analysis result. Nutrition analysis and diagnosis and treatment are provided for the user.

Description

Technical Field

[0001] The present invention relates to the field of intelligent nutritional diagnosis and treatment technology, and in particular to an artificial intelligence-based nutritional analysis and diagnosis and treatment decision support system. Background Art

[0002] With the continuous improvement of people's living standards, health management and dietary nutrition have become hot topics of social concern. Reasonable nutritional intake can not only promote human health and prevent chronic diseases, but also play an important role in disease recovery and weight management. Traditional nutritional analysis and dietary diagnosis and treatment mainly rely on the manual evaluation and experience judgment of nutritionists. It is usually necessary to obtain users' diet and health information through questionnaires, diet diaries, manual records, etc., and professionals conduct nutritional analysis and formulate personalized recommendations. However, this method has many problems such as cumbersome data collection, strong subjectivity, low evaluation efficiency, and limited analysis dimensions. It is difficult to meet the needs of large-scale, personalized health management.

[0003] In recent years, with the rapid development of technologies such as artificial intelligence, big data, and computer vision, intelligent health management and dietary analysis have been widely researched. Some studies have proposed food classification and nutrient estimation methods based on image recognition. These methods automatically estimate food type, portion size, and corresponding nutritional content by identifying and analyzing user-uploaded meal images. While these methods have improved the automation of nutritional analysis to a certain extent, they still face numerous technical bottlenecks. First, traditional image segmentation and recognition algorithms lack accuracy for complex meal images with overlapping, occluded, and similarly colored ingredients, making it difficult to accurately extract the region and boundaries of a single ingredient. Second, existing nutritional analysis systems often focus solely on the food image itself, ignoring health profile information such as individual user differences, medical history, and lifestyle habits, which are crucial for personalized nutritional recommendations. Third, existing feature fusion methods often employ simple concatenation or weighting methods, failing to fully exploit the deep correlations between features from different sources, thus impacting the accuracy of subsequent nutritional analysis and treatment decisions.

[0004] In addition, although deep learning technology has achieved remarkable results in the fields of medical imaging and speech recognition, its application in the field of dietary nutrition analysis still faces many challenges. For example, deep neural network models have many parameters and complex structures, are prone to falling into local optimality, and lack effective automatic optimization mechanisms in the fusion of different data sources and personalized modeling. Some existing studies have attempted to introduce intelligent optimization algorithms such as genetic algorithms and particle swarm optimization to adjust network structure and parameters, but they are still not ideal in terms of feature diversity, model generalization ability, and computational efficiency. At the same time, the generation of nutritional component analysis and diagnosis and treatment recommendations often relies on static nutritional databases and rule bases, lacks dynamic personalized matching and real-time optimization, and is difficult to meet the health needs of diverse users.

[0005] In the existing technology, most nutritional analysis systems only have the processing capabilities of a single data source and lack the collaborative analysis of multimodal and multi-source data. Factors such as users' eating behaviors, health conditions, and living habits are complex and changeable. A single data dimension cannot fully reflect the user's true health conditions, resulting in insufficient scientificity and pertinence in the final analysis conclusions. As for the processing of food images, traditional methods are mostly limited to simple image classification or segmentation algorithms, which are difficult to cope with the lighting changes, different shooting angles, and complex food appearance in the actual dining environment. In addition, the use of health record information mostly stays at static attribute matching, lacks in-depth mining of dynamic features such as time series and behavioral patterns, resulting in inaccurate personalized diagnosis and treatment recommendations.

[0006] When it comes to feature fusion, existing technologies often overlook the complex relationship between dietary image features and health profile features. Combining the two involves simple feature concatenation or weighted averaging, making it difficult to achieve deep interaction and effective fusion of information. This directly impacts the personalization and accuracy of subsequent nutritional analysis and treatment recommendations. Furthermore, feature fusion optimization relies primarily on manual experience and static parameter configuration, lacking an adaptive, intelligent, end-to-end optimization mechanism, making it difficult to achieve optimal results for different users and scenarios.

[0007] Furthermore, existing systems often present output using fixed templates or static text, lacking dynamic responses to users' individual needs and actual health goals, resulting in limited user experience and guidance. Nutrient analysis and treatment recommendations often lack scientific weighting and comprehensive assessment, failing to fully integrate users' health records, dietary behaviors, and nutritional needs to provide targeted health guidance.

[0008] Therefore, how to provide a nutritional analysis and diagnosis and treatment decision support system based on artificial intelligence is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0009] One objective of this invention is to develop an artificial intelligence-based nutritional analysis and treatment decision support system. This system utilizes data preprocessing, dietary image segmentation, and multi-source feature fusion, combined with an ant algorithm to optimize neural network parameters, to achieve efficient joint modeling and nutrient analysis of dietary images and health records. The system generates personalized nutritional analysis reports and treatment recommendations, and continuously optimizes decision-making strategies through closed-loop feedback, thereby enhancing the scientific and intelligent nature of nutritional health management.

[0010] An artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to an embodiment of the present invention includes:

[0011] Data preprocessing module, used to format and standardize the collected dietary images and health record information;

[0012] The meal image segmentation module is used to perform edge detection and segmentation on standardized meal images, extracting the edge contours and regional features of the ingredients;

[0013] A dietary feature extraction module is used to extract dietary image features and health profile features respectively through a sub-stream neural network;

[0014] Feature fusion module, used to fuse dietary image features and health profile features to generate a joint feature representation;

[0015] Feature optimization module, which uses the ant algorithm to optimize the sub-stream neural network structure and weight parameters and output the optimal feature fusion strategy;

[0016] Nutritional component analysis module, used to analyze the user's nutritional status based on the fused joint features;

[0017] The personalized diagnosis and treatment decision-making module is used to combine the user's health records and nutritional analysis results to generate personalized nutrition analysis reports and diagnosis and treatment recommendations.

[0018] Optionally, modules can be connected using the following methods:

[0019] S1. Collect the user's dietary image data and health record information and pre-process them to generate standardized dietary images and standardized health record information;

[0020] S2. Input the standardized meal image into an edge detection neural network to perform image segmentation, extract the edge contours and regional features of the food, and obtain a segmented food image;

[0021] S3, inputting the segmented food image into the first sub-stream neural network to extract dietary image features, and inputting the standardized health record information into the second sub-stream neural network to extract health record features, thereby obtaining dietary image features and health record features respectively;

[0022] S4, integrating dietary image features and health profile features to obtain the final joint feature representation;

[0023] S5. Use the ant algorithm to optimize the structural parameters and weight parameters of the first sub-stream neural network and the second sub-stream neural network, configure a tracking and optimization unit for each ant, optimize through centralized training and distributed execution mechanism, and output the optimal feature fusion strategy;

[0024] S6. Input the final joint feature representation into the nutritional component analysis and diagnosis and treatment decision unit, and output a personalized nutritional analysis report and diagnosis and treatment recommendations.

[0025] Optionally, the health record information includes user physical data, disease history, nutritional intake records and living habit data.

[0026] Optionally, the meal picture data represents image information of a user's daily meal, and a shooting time, a shooting location, and a shooting device related to the picture.

[0027] Optionally, the S2 comprises the following specific steps:

[0028] S21, converting the standardized meal picture into a gray-scale image to obtain a gray-scale meal picture;

[0029] S22, performing noise removal processing on the gray-scale meal picture to obtain a denoised meal picture;

[0030] S23, inputting the denoised meal picture into an edge detection neural network to obtain an initial edge map of the meal picture;

[0031] S24, extracting edge contour information of food materials based on the initial edge map of the meal picture, and outputting an edge contour map of the meal picture;

[0032] S25, performing region feature analysis on the edge contour map of the meal picture to separate different food material regions, and obtaining a separated food material region map;

[0033] S26, performing food material region labeling on the separated food material region map, and outputting a final segmented food material image.

[0034] Optionally, the S3 comprises the following specific steps:

[0035] S31, performing multi-scale transformation processing on the segmented food material image to generate food material image feature maps of multiple scales;

[0036] S32, filtering and weighting the food material image feature maps of multiple scales by using an attention mechanism to obtain fused food material image feature maps;

[0037] S33, inputting the fused food material image feature maps into a first sub-stream neural network, performing deep feature extraction through a multi-layer residual structure, and outputting meal picture features;

[0038] S34, performing time series modeling on the standardized health record information, extracting time series features in the user's health record, and generating sequence health record features;

[0039] S35, inputting the sequence health record features into a second sub-stream neural network, combining a graph convolution structure to perform associated feature extraction, and obtaining health record features.

[0040] Optionally, the S4 comprises the following specific steps:

[0041] S41, performing adaptive feature selection on the dietary image features, screening out key dietary image features related to health profile features, and obtaining a key dietary image feature set;

[0042] S42, generating a multi-dimensional health record feature representation by performing multi-layer nonlinear mapping on the health record features;

[0043] S43, using a bidirectional gating mechanism to interact the key dietary image feature set and the multidimensional health profile feature representation to obtain an interactive feature matrix;

[0044] S44, perform multi-head attention fusion on the interactive feature matrix to form a multi-level fusion feature representation;

[0045] S45, inputting the multi-level fusion feature representation into the joint feature compression unit, extracting the main feature components through low-rank decomposition, and generating a simplified joint feature vector;

[0046] S46. Output the simplified joint feature vector as the final joint feature representation.

[0047] Optionally, S5 includes the following specific steps:

[0048] S51, initialize the ant algorithm, assign a tracking unit and an optimization unit to each ant, set the initial values ​​of the neural network structure parameters and weight parameters, and generate an ant individual parameter set;

[0049] S52, using the ant individual parameter set to train the first sub-stream neural network and the second sub-stream neural network respectively, to obtain a neural network training result set;

[0050] S53, performing performance evaluation on the neural network training result set, and generating a performance evaluation data set based on the performance of each ant individual in the feature fusion task;

[0051] S54, based on the performance evaluation data set, using the path update mechanism of the ant algorithm, adjusting the neural network structure parameters and weight parameters in the ant individual parameter set to obtain an updated ant individual parameter set;

[0052] S55, using the updated ant individual parameter set, combined with centralized training and distributed execution mechanisms, cyclically executing training and evaluation steps until a preset convergence condition is met, and outputting the optimal ant individual parameter set;

[0053] S56. Apply the optimal ant individual parameter set to the first sub-stream neural network and the second sub-stream neural network to generate and output the optimal feature fusion strategy.

[0054] Optionally, S6 includes the following specific steps:

[0055] S61, applying the optimal feature fusion strategy to the final joint feature representation to generate fused joint feature data;

[0056] S62, inputting the fused joint feature data into a nutrient component analysis unit to extract multi-dimensional nutrient component feature information;

[0057] S63, performing feature correction on the multi-dimensional nutritional component characteristic information, performing data correction in combination with a reference nutritional database, and obtaining corrected nutritional characteristic data;

[0058] S64, inputting the corrected nutritional characteristic data into the personalized demand matching unit, performing characteristic matching in combination with the user's health file information, and generating a personalized matching result;

[0059] S65. Based on the personalized matching results, the diagnosis and treatment knowledge related to the user's condition is filtered through the diagnosis and treatment rule engine to generate preliminary diagnosis and treatment recommendations;

[0060] S66. Integrate the preliminary results of diagnosis and treatment recommendations with the personalized matching results, and use the comprehensive assessment unit to output personalized nutrition analysis reports and diagnosis and treatment recommendations.

[0061] The beneficial effects of the present invention are:

[0062] In response to the shortcomings of existing nutritional analysis and personalized diagnosis and treatment decision-making systems in terms of multimodal data fusion, dynamic optimization and intelligent decision-making, the present invention proposes a new intelligent system that integrates dietary image segmentation, health record modeling and feature depth optimization. Compared with the traditional single data source, manual evaluation and templated recommendation methods, the present invention can automatically collect and standardize the user's dietary images and health record data, and through the collaborative optimization of deep learning and ant algorithms, it can achieve efficient fusion of multi-source features and a deep understanding of diet and health behaviors. The system can not only achieve precise segmentation of multiple ingredients and scientific analysis of nutritional components in complex dietary scenarios, but also dynamically perceive the health status and eating habits of individual users, and automatically generate nutritional analysis reports and diagnosis and treatment recommendations that are both scientific and personalized, greatly improving the accuracy of analysis and the pertinence of services.

[0063] By introducing an intelligent optimization algorithm to adaptively adjust the neural network structure and parameters, the present invention effectively solves the problems of insufficient model generalization ability and low feature fusion efficiency, enabling the system to continuously output high-quality personalized health recommendations for different users and different dietary environments. At the same time, the system's built-in closed-loop feedback mechanism can dynamically optimize feature modeling and decision-making strategies based on user execution results and changes in health indicators, ensuring the scientific nature and timeliness of nutritional analysis and treatment recommendations. Through visual display and intelligent decision-making support functions, both users and professionals can intuitively understand the analysis results, greatly enhancing the interactive experience and execution enthusiasm of health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0065] Figure 1 This is a method flow chart of an artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system proposed by the present invention;

[0066] Figure 2 This is a system flow chart of an artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system proposed by the present invention;

[0067] Figure 3 This is a data flow diagram of the artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system proposed by the present invention. DETAILED DESCRIPTION

[0068] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0069] refer to Figure 1-3 , an artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system, including:

[0070] Data preprocessing module, used to format and standardize the collected dietary images and health record information;

[0071] The meal image segmentation module is used to perform edge detection and segmentation on standardized meal images, extracting the edge contours and regional features of the ingredients;

[0072] A dietary feature extraction module is used to extract dietary image features and health profile features respectively through a sub-stream neural network;

[0073] Feature fusion module, used to fuse dietary image features and health profile features to generate a joint feature representation;

[0074] Feature optimization module, which uses the ant algorithm to optimize the sub-stream neural network structure and weight parameters and output the optimal feature fusion strategy;

[0075] Nutritional component analysis module, used to analyze the user's nutritional status based on the fused joint features;

[0076] The personalized diagnosis and treatment decision-making module is used to combine the user's health records and nutritional analysis results to generate personalized nutrition analysis reports and diagnosis and treatment recommendations.

[0077] This method combines data preprocessing, image segmentation, and deep fusion of multi-source features, using an ant algorithm to dynamically optimize the structure and parameters of substream neural networks. This allows for efficient joint modeling of dietary images and health profile features. The system accurately analyzes individual nutritional components and automatically generates personalized treatment recommendations, effectively enhancing the scientific and intelligent nature of nutritional assessments.

[0078] In this embodiment, the modules are connected through the following methods:

[0079] S1. Collect the user's dietary image data and health record information and pre-process them to generate standardized dietary images and standardized health record information;

[0080] S2. Input the standardized meal image into an edge detection neural network to perform image segmentation, extract the edge contours and regional features of the food, and obtain a segmented food image;

[0081] S3, inputting the segmented food image into the first sub-stream neural network to extract dietary image features, and inputting the standardized health record information into the second sub-stream neural network to extract health record features, thereby obtaining dietary image features and health record features respectively;

[0082] S4, integrating dietary image features and health profile features to obtain the final joint feature representation;

[0083] S5. Use the ant algorithm to optimize the structural parameters and weight parameters of the first sub-stream neural network and the second sub-stream neural network, configure a tracking and optimization unit for each ant, optimize through centralized training and distributed execution mechanism, and output the optimal feature fusion strategy;

[0084] S6. Input the final joint feature representation into the nutritional component analysis and diagnosis and treatment decision unit, and output a personalized nutritional analysis report and diagnosis and treatment recommendations.

[0085] This method uses a multi-step hierarchical process, edge detection, and a two-stream neural network to extract features from dietary images and health records, and fuses them into a joint feature representation. An ant algorithm is then used to globally optimize the network structure and parameters, enabling personalized feature modeling. The resulting output of nutritional reports and treatment recommendations is more accurate and scientific, effectively enhancing the intelligence and service efficiency of personalized nutrition and health management.

[0086] In this embodiment, the health record information includes user's physical data, disease history, nutritional intake records and living habit data.

[0087] This method integrates user data such as vital signs, medical history, nutritional intake, and lifestyle habits to construct a multidimensional health profile, providing a comprehensive picture of an individual's health status. This method effectively improves the targeted and scientific nature of nutritional assessments and treatment recommendations, making the output more tailored to the user's actual needs and optimizing personalized health management outcomes.

[0088] In this embodiment, the meal picture data includes image information of the user's daily meals, as well as the shooting time, location, and shooting equipment related to the pictures.

[0089] This method collects images of users' daily meals and, by incorporating the time, location, and equipment used to capture them, enhances the spatiotemporal labeling and environmental information of these meal images. This method effectively improves the accuracy and adaptability of dietary behavior analysis, making nutritional assessments more targeted and practical.

[0090] In this embodiment, S2 includes the following specific steps:

[0091] S21, converting the standardized meal picture into a grayscale image to obtain a grayscale meal picture;

[0092] S22, performing noise removal processing on the grayscale meal image to obtain a denoised meal image;

[0093] S23, inputting the denoised meal image into an edge detection neural network to obtain an initial edge map of the meal image;

[0094] S24, extracting edge contour information of the food ingredients based on the initial edge map of the meal image, and outputting an edge contour map of the meal image;

[0095] S25, performing regional feature analysis using the edge contour map of the meal image to separate different food regions and obtain a separated food region map;

[0096] S26: label the separated food area map and output the final segmented food image.

[0097] This method uses a step-by-step grayscale conversion, noise removal, edge detection, and region separation process on meal images to accurately extract and label ingredient edges and regions. This method improves the accuracy of ingredient segmentation, provides a high-quality data foundation for subsequent nutritional analysis and personalized recommendations, and significantly enhances the system's intelligent recognition and analysis capabilities.

[0098] In this embodiment, S3 includes the following specific steps:

[0099] S31, performing multi-scale transformation processing on the segmented food image to generate food image feature maps of multiple scales;

[0100] S32. Using an attention mechanism to filter and weight the food image feature maps at multiple scales to obtain a fused food image feature map;

[0101] S33, inputting the fused food image feature map into the first sub-stream neural network, performing deep feature extraction through a multi-layer residual structure, and outputting meal image features;

[0102] S34, performing time series modeling on the standardized health record information, extracting time series features in the user's health record, and generating sequence health record features;

[0103] S35. Input the sequence health record features into the second sub-stream neural network, combine the graph convolution structure to extract the associated features, and obtain the health record features.

[0104] This method uses multi-scale transformation and an attention mechanism to filter and fuse food image features. Combining a multi-layer residual structure with a graph convolutional network, it deeply extracts temporal and correlation features from dietary images and health records. This method enhances the richness and relevance of feature representation, making subsequent nutritional analysis more comprehensive and accurate, and effectively enhancing the scientific nature and reliability of personalized health management.

[0105] In this embodiment, S4 includes the following specific steps:

[0106] S41, performing adaptive feature selection on the dietary image features, screening out key dietary image features related to health profile features, and obtaining a key dietary image feature set;

[0107] S42, generating a multi-dimensional health record feature representation by performing multi-layer nonlinear mapping on the health record features;

[0108] S43, using a bidirectional gating mechanism to interact the key dietary image feature set and the multidimensional health profile feature representation to obtain an interactive feature matrix;

[0109] S44, perform multi-head attention fusion on the interactive feature matrix to form a multi-level fusion feature representation;

[0110] S45, inputting the multi-level fusion feature representation into the joint feature compression unit, extracting the main feature components through low-rank decomposition, and generating a simplified joint feature vector;

[0111] S46. Output the simplified joint feature vector as the final joint feature representation.

[0112] This method extracts key dietary image features and multidimensional health profile features through adaptive feature selection and multi-layer nonlinear mapping. It also employs a bidirectional gating mechanism and multi-head attention to achieve deep information interaction and multi-level fusion. Combined with low-rank decomposition for feature compression, it significantly improves the accuracy and effectiveness of the combined features, providing solid data support for personalized health interventions and intelligent recommendations.

[0113] In this embodiment, S5 includes the following specific steps:

[0114] S51, initialize the ant algorithm, assign a tracking unit and an optimization unit to each ant, set the initial values ​​of the neural network structure parameters and weight parameters, and generate an ant individual parameter set;

[0115] S52, using the ant individual parameter set to train the first sub-stream neural network and the second sub-stream neural network respectively, to obtain a neural network training result set;

[0116] S53, performing performance evaluation on the neural network training result set, and generating a performance evaluation data set based on the performance of each ant individual in the feature fusion task;

[0117] S54, based on the performance evaluation data set, using the path update mechanism of the ant algorithm, adjusting the neural network structure parameters and weight parameters in the ant individual parameter set to obtain an updated ant individual parameter set;

[0118] S55, using the updated ant individual parameter set, combined with centralized training and distributed execution mechanisms, cyclically executing training and evaluation steps until a preset convergence condition is met, and outputting the optimal ant individual parameter set;

[0119] S56. Apply the optimal ant individual parameter set to the first sub-stream neural network and the second sub-stream neural network to generate and output the optimal feature fusion strategy.

[0120] This paper introduces an ant algorithm to intelligently optimize neural network structural and weight parameters. Combining centralized training with distributed execution, this method dynamically adjusts the network structure to efficiently improve feature fusion performance. This method enables a global search for optimal parameter sets, enhancing the adaptability and accuracy of feature fusion strategies, and providing the system with more optimized feature representation and decision support.

[0121] In this embodiment, S6 includes the following specific steps:

[0122] S61, applying the optimal feature fusion strategy to the final joint feature representation to generate fused joint feature data;

[0123] S62, inputting the fused joint feature data into a nutrient component analysis unit to extract multi-dimensional nutrient component feature information;

[0124] S63, performing feature correction on the multi-dimensional nutritional component characteristic information, performing data correction in combination with a reference nutritional database, and obtaining corrected nutritional characteristic data;

[0125] S64, inputting the corrected nutritional characteristic data into the personalized demand matching unit, performing characteristic matching in combination with the user's health file information, and generating a personalized matching result;

[0126] S65. Based on the personalized matching results, the diagnosis and treatment knowledge related to the user's condition is filtered through the diagnosis and treatment rule engine to generate preliminary diagnosis and treatment recommendations;

[0127] S66. Integrate the preliminary results of diagnosis and treatment recommendations with the personalized matching results, and use the comprehensive assessment unit to output personalized nutrition analysis reports and diagnosis and treatment recommendations.

[0128] This invention applies an optimal feature fusion strategy to joint feature representation, combined with multidimensional nutrient analysis and correction, to achieve precise matching of personalized needs. Leveraging a diagnosis and treatment rule engine and a comprehensive assessment mechanism, it generates personalized nutrition analysis reports and treatment recommendations, significantly improving the scientific, targeted, and intelligent nature of nutritional assessment and health intervention, providing users with more accurate health guidance services.

[0129] Example 1:

[0130] In order to verify the feasibility of the present invention in implementation, the present invention was applied to the clinical nutrition management scenario of the Nutrition Department of the Third People's Hospital of a certain city. The hospital has long been responsible for the dietary guidance and health management of patients with chronic diseases such as diabetes and hypertension. The number of patients is huge, the diet types are complex, and the individual differences are obvious, which brings great challenges to traditional manual nutrition assessment and personalized diagnosis and treatment recommendations. In the past, the methods of manual questionnaires, paper diet records and manual analysis by nutritionists had prominent problems such as difficult data collection, many omissions of information, strong subjectivity of manual evaluation, slow response speed, and inaccurate recommendations. It is difficult to meet the current needs of chronic disease management and personalized health services.

[0131] In this embodiment, the hospital's nutrition department selected 100 chronic disease patients as pilot subjects, covering common chronic diseases such as diabetes, hypertension, and hyperlipidemia, with an age distribution between 35 and 70 years old. Each patient uploads pictures of three meals a day through the hospital's self-service health management APP and fills in health file information, including basic physical signs, medical history, nutritional intake records, and living habits. The system automatically formats and standardizes the pictures and archival data uploaded by the patients to eliminate data deviations caused by different shooting equipment, picture resolutions, and information entry methods. Subsequently, the system segments the pictures through a deep learning model, extracts the edge and regional features of each meal's staple food, side dishes, vegetables, meat and other ingredients, and realizes accurate identification of multiple ingredients in complex dishes. At the same time, the patient's health file information undergoes time series modeling and feature extraction, and the system can dynamically capture the physical sign changes and eating behavior patterns of patients with chronic diseases.

[0132] In the multi-source feature fusion stage, the system uses adaptive feature selection and multi-head attention mechanisms to deeply fuse image features with health record features, fully exploring the correlation between the two. By introducing the ant algorithm to optimize the network structure and weight parameters, the generalization ability of the model in different patients and different dietary scenarios is effectively improved. The nutritional component analysis module combines the fused features to automatically and accurately analyze the patient's daily intake of energy, protein, fat, carbohydrates, vitamins, minerals and other multi-dimensional nutrients, and calibrates them with the national standard nutrition database to ensure the scientific nature of the analysis results. Based on the patient's health record and nutritional analysis results, the personalized diagnosis and treatment decision module combines the diagnosis and treatment rule engine with the demand matching mechanism to automatically generate personalized, dynamically adjusted nutrition analysis reports and diagnosis and treatment recommendations, and push them to patients and attending nutritionists in the form of a combination of pictures and texts, greatly improving patients' understanding and execution.

[0133] Table 1 Comparison of optimization effects of nutrition analysis and diagnosis and treatment decision support system based on artificial intelligence

[0134]

[0135] Table 1 demonstrates how this system, through a closed-loop decision-making feedback mechanism, continuously optimizes feature extraction and fusion strategies, ensuring that diagnosis and treatment recommendations are more tailored to patients' actual needs. During this period, the hospital also invited third-party experts to manually review the system reports, and the consistency rate between the system's automatically generated recommendations and the expert recommendations reached over 95%. Patient satisfaction surveys revealed that 92% of patients found the system reports easy to understand, and 91% expressed a willingness to use the system for long-term health management.

[0136] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An artificial intelligence-based nutritional analysis and diagnosis and treatment decision support system, characterized in that: include: Data preprocessing module, used to format and standardize the collected dietary images and health record information; The meal image segmentation module is used to perform edge detection and segmentation on standardized meal images, extracting the edge contours and regional features of the ingredients; A dietary feature extraction module is used to extract dietary image features and health profile features respectively through a sub-stream neural network; Feature fusion module, used to fuse dietary image features and health profile features to generate a joint feature representation; Feature optimization module, which uses the ant algorithm to optimize the sub-stream neural network structure and weight parameters and output the optimal feature fusion strategy; Nutritional component analysis module, used to analyze the user's nutritional status based on the fused joint features; The personalized diagnosis and treatment decision-making module is used to combine the user's health records and nutritional analysis results to generate personalized nutrition analysis reports and diagnosis and treatment recommendations.

2. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 1, characterized in that: The modules are implemented as follows: S1. Collect the user's dietary image data and health record information and pre-process them to generate standardized dietary images and standardized health record information; S2. Input the standardized meal image into an edge detection neural network to perform image segmentation, extract the edge contours and regional features of the food, and obtain a segmented food image; S3, inputting the segmented food image into the first sub-stream neural network to extract dietary image features, and inputting the standardized health record information into the second sub-stream neural network to extract health record features, thereby obtaining dietary image features and health record features respectively; S4, integrating dietary image features and health profile features to obtain the final joint feature representation; S5. Use the ant algorithm to optimize the structural parameters and weight parameters of the first sub-stream neural network and the second sub-stream neural network, configure a tracking and optimization unit for each ant, optimize through centralized training and distributed execution mechanism, and output the optimal feature fusion strategy; S6. Input the final joint feature representation into the nutritional component analysis and diagnosis and treatment decision unit, and output a personalized nutritional analysis report and diagnosis and treatment recommendations.

3. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 2, characterized in that: The health record information includes user's physical data, disease history, nutritional intake records and living habit data.

4. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 2, characterized in that: The meal picture data includes image information of the user's daily meals, as well as the shooting time, location, and shooting equipment related to the pictures.

5. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 2, characterized in that: The S2 includes the following specific steps: S21, converting the standardized meal picture into a grayscale image to obtain a grayscale meal picture; S22, performing noise removal processing on the grayscale meal image to obtain a denoised meal image; S23, inputting the denoised meal image into an edge detection neural network to obtain an initial edge map of the meal image; S24, extracting edge contour information of the food ingredients based on the initial edge map of the meal image, and outputting an edge contour map of the meal image; S25, performing regional feature analysis using the edge contour map of the meal image to separate different food regions and obtain a separated food region map; S26: label the separated food area map and output the final segmented food image.

6. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 2, characterized in that: The S3 includes the following specific steps: S31, performing multi-scale transformation processing on the segmented food image to generate food image feature maps of multiple scales; S32. Using an attention mechanism to filter and weight the food image feature maps at multiple scales to obtain a fused food image feature map; S33, inputting the fused food image feature map into the first sub-stream neural network, performing deep feature extraction through a multi-layer residual structure, and outputting meal image features; S34, performing time series modeling on the standardized health record information, extracting time series features in the user's health record, and generating sequence health record features; S35. Input the sequence health record features into the second sub-stream neural network, combine the graph convolution structure to extract the associated features, and obtain the health record features.

7. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 2, characterized in that: The S4 includes the following specific steps: S41, performing adaptive feature selection on the dietary image features, screening out key dietary image features related to health profile features, and obtaining a key dietary image feature set; S42, generating a multi-dimensional health record feature representation by performing multi-layer nonlinear mapping on the health record features; S43, using a bidirectional gating mechanism to interact the key dietary image feature set and the multidimensional health profile feature representation to obtain an interactive feature matrix; S44, perform multi-head attention fusion on the interactive feature matrix to form a multi-level fusion feature representation; S45, inputting the multi-level fusion feature representation into the joint feature compression unit, extracting the main feature components through low-rank decomposition, and generating a simplified joint feature vector; S46. Output the simplified joint feature vector as the final joint feature representation.

8. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 2, characterized in that: The S5 includes the following specific steps: S51, initialize the ant algorithm, assign a tracking unit and an optimization unit to each ant, set the initial values ​​of the neural network structure parameters and weight parameters, and generate an ant individual parameter set; S52, using the ant individual parameter set to train the first sub-stream neural network and the second sub-stream neural network respectively, to obtain a neural network training result set; S53, performing performance evaluation on the neural network training result set, and generating a performance evaluation data set based on the performance of each ant individual in the feature fusion task; S54, based on the performance evaluation data set, using the path update mechanism of the ant algorithm, adjusting the neural network structure parameters and weight parameters in the ant individual parameter set to obtain an updated ant individual parameter set; S55, using the updated ant individual parameter set, combined with centralized training and distributed execution mechanisms, cyclically executing training and evaluation steps until a preset convergence condition is met, and outputting the optimal ant individual parameter set; S56. Apply the optimal ant individual parameter set to the first sub-stream neural network and the second sub-stream neural network to generate and output the optimal feature fusion strategy.

9. The artificial intelligence-based nutrition analysis and diagnosis and treatment decision support system according to claim 2, characterized in that: The S6 comprises the following specific steps: S61, applying the optimal feature fusion strategy to the final joint feature representation to generate fused joint feature data; S62, inputting the fused joint feature data into a nutrient component analysis unit to extract multi-dimensional nutrient component feature information; S63, performing feature correction on the multi-dimensional nutritional component characteristic information, performing data correction in combination with a reference nutritional database, and obtaining corrected nutritional characteristic data; S64, inputting the corrected nutritional characteristic data into the personalized demand matching unit, performing characteristic matching in combination with the user's health file information, and generating a personalized matching result; S65. Based on the personalized matching results, the diagnosis and treatment knowledge related to the user's condition is filtered through the diagnosis and treatment rule engine to generate preliminary diagnosis and treatment recommendations; S66. Integrate the preliminary results of diagnosis and treatment recommendations with the personalized matching results, and use the comprehensive assessment unit to output personalized nutrition analysis reports and diagnosis and treatment recommendations.

Images