A multimodal sensor fusion-based core muscle group state evaluation and health care scheme intelligent matching system

Abstract

The application discloses a kind of multi-modal sensing fusion's core muscle group state evaluation and health care scheme intelligent matching system and method, belong to intelligent health care and AI muscle bone noninvasive detection technical field.System is by multi-modal data acquisition unit, AI feature extraction evaluation unit, longitudinal data storage unit, health care scheme matching output unit constitute, through adaptive high-frequency muscle bone ultrasonic probe, multi-frequency body composition module, RGBD depth camera fusion collection gluteus maximus, erector spinae, pectoralis major microcosmic muscle structure parameter and macroscopic body data;Rely on based on U-Net+ResNet-34 semantic segmentation network and random forest model automatic quantification muscle thickness, echo gray scale, high echo proportion and so on index, complete muscle group degenerative classification, annual attenuation rate calculation and trend prediction;Establish personal health care archives to realize long-term longitudinal tracking, and according to the evaluation result intelligent matching non-medical diet and home training health care scheme.The application solves the pain points that traditional core muscle group detection relies on artificial interpretation, cannot long-term dynamic monitoring, detection and health care service disjoint, can be widely applied to family, community and old-age care institution intelligent health care scene.

Description

Technical Field

[0001] This invention belongs to the fields of intelligent health care, non-invasive detection of human muscle groups, multimodal sensor fusion, and artificial intelligence algorithm technology. Specifically, it involves an integrated workstation system that non-invasively collects multi-dimensional physical parameters such as muscle thickness, echo grayscale distribution, and high echo area proportion for the three core muscle groups of gluteus maximus, erector spinae, and pectoralis major. It relies on artificial intelligence algorithms to automatically assess the condition of muscle groups and intelligently match non-medical daily diet and home activity health care guidance programs. Background Technology

[0002] In the process of population aging, degenerative changes in skeletal muscles in middle-aged and elderly people directly lead to a decline in limb mobility and a reduction in quality of life. Accurate detection and long-term monitoring of the structural status of the three core muscle groups—the gluteus maximus, erector spinae, and pectoralis major—is a core requirement for smart elderly care in homes and communities. Existing muscle detection and elderly care management technologies have several technical shortcomings, specifically as follows: First, the detection is not targeted enough. Conventional body composition analysis equipment can only count the total amount of skeletal muscle in the limbs as a whole. It cannot perform separate quantitative analysis of local core muscle groups such as the gluteus maximus, erector spinae, and pectoralis major. It cannot accurately obtain fine structural parameters such as muscle thickness, echo grayness, and the proportion of hyperechoic areas, which makes it difficult to meet the needs of specialized assessment of core muscle groups. Secondly, the level of intelligence in detection is low. Traditional musculoskeletal ultrasound detection relies entirely on professional physicians to manually interpret image data. The detection results are greatly affected by the operator's experience and technique, and it is impossible to achieve standardized, automated, and repeatable quantitative extraction of muscle group characteristic parameters, resulting in poor detection consistency. Third, there is a lack of longitudinal monitoring capabilities. Existing detection equipment only supports single static data acquisition and lacks functions for historical data storage, longitudinal comparison, rate of change calculation, and trend prediction, making it impossible to achieve long-term dynamic tracking of the state of the human core muscle groups. Fourth, there is a disconnect between testing and health and wellness services. Traditional testing equipment only outputs raw data reports and cannot match corresponding health and wellness guidance plans with muscle group test results, thus failing to form a closed-loop health and wellness service system that integrates data collection, plan matching, and long-term follow-up.

[0003] Existing publicly available technologies mostly focus on the detection of routine muscle groups in the limbs, such as the gastrocnemius and quadriceps. There is currently no technology specifically designed for the multi-parameter automatic detection and status grading assessment of the three core muscle groups: the gluteus maximus, erector spinae, and pectoralis major. Furthermore, there is no integrated system that combines muscle group detection data to intelligently match daily health and wellness plans, indicating a significant technological gap in the industry. Summary of the Invention

[0004] Purpose of the invention This invention addresses the shortcomings of existing technologies, such as low accuracy in core muscle group detection, reliance on manual interpretation, lack of longitudinal tracking capabilities, and separation between detection and rehabilitation services. It provides a multimodal sensing fusion system for core muscle group status assessment and intelligent matching of rehabilitation plans. This invention integrates ultrasound sensing, body composition detection, and depth vision acquisition technologies to achieve automated and accurate extraction of multi-dimensional parameters from the three core muscle groups. It utilizes AI algorithms to perform graded assessment of muscle group status, automatically calculates trends in muscle group status changes, and intelligently matches standardized home-based rehabilitation guidance plans, constructing a fully automated closed-loop rehabilitation service. Technical solution A multimodal sensing fusion core muscle group status assessment and health care plan intelligent matching system is characterized by comprising a multimodal data acquisition unit, an AI feature extraction and status assessment unit, a longitudinal comparison and storage unit, and a health care plan intelligent matching output unit, all of which are integrated into an integrated health care workstation device. Multimodal data acquisition unit The multimodal data acquisition unit is used to acquire microstructural parameters of the human core muscle groups and macroscopic posture and vital signs data, including an adaptive high-frequency musculoskeletal ultrasound probe, a multi-frequency multi-electrode body component detection module, an RGBD depth vision camera and a basic vital signs input module. The adaptive high-frequency musculoskeletal ultrasound probe has wideband adaptive adjustment capability and sub-millimeter spatial resolution, and is used to acquire parameters such as muscle thickness, ultrasound echo grayscale texture, and proportion of intramuscular hyperechoic areas of the gluteus maximus, erector spinae, and pectoralis major muscles. The multi-frequency multi-electrode body component detection module is used to collect the total amount of human skeletal muscle, bioimpedance, and phase angle parameters as benchmark reference data for muscle group status assessment. The RGBD depth vision camera is used to acquire human body contours, chest shape, and pelvic tilt angle, and to quantify and output body offset parameters. The basic vital signs input module is used to input basic information such as the subject's age, height, weight, and average daily sedentary time, providing auxiliary input features for the AI ​​evaluation model. AI Feature Extraction and State Evaluation Unit The AI ​​feature extraction and state evaluation unit is the core data processing module of the system, which includes an image segmentation network module, a multi-parameter fusion evaluation model, and a longitudinal comparison sub-module. The image segmentation network module uses a deep learning semantic segmentation algorithm to automatically identify and select the regions of interest for the three core muscle groups, remove interference signals such as fat and bone, and accurately extract four-dimensional quantitative parameters such as muscle thickness, mean echo gray value, proportion of high echo region, and body posture offset. The multi-parameter fusion evaluation model completes feature fusion analysis based on four-dimensional acquired parameters, classifies the muscle group structural state into four standard levels, and achieves accurate classification of muscle group degenerative state. The longitudinal comparison submodule is used to store the detection data of each time, automatically calculate the annual rate of change of muscle group thickness, generate a state change trend curve, and predict the trend of muscle group state change in the next 1-3 years. Vertical comparison of storage units The longitudinal comparison storage unit builds a dedicated personal health and wellness electronic record, uniformly storing original collected images, quantitative feature parameters, status grading results, and trend data. It supports cross-cycle data comparison, data export, and long-term follow-up records, enabling full life-cycle tracking of the core muscle group status. Intelligent matching output unit for health and wellness solutions The intelligent matching output unit of the health and wellness program associates with the AI ​​assessment and grading results. Based on various quantitative parameters of the muscle groups, it automatically matches and pushes non-medical health and wellness reference programs, including personalized dietary nutrition guidance, home training and stretching guidance for core muscle groups, and automatically updates the follow-up plan, reminding of the re-examination time and monitoring focus. This invention also provides a method for intelligently matching core muscle group status assessment and rehabilitation programs using multimodal sensor fusion, comprising the following steps: S1. Subjects enter their basic personal information through the vital signs entry module and then complete the ultrasound image acquisition of the three core muscle groups, the multi-frequency body composition detection, and the three-dimensional body posture scan acquisition in sequence. S2. The system automatically segments the target muscle group region through an AI image segmentation network and extracts four-dimensional standardized physical parameters; S3. The multi-parameter fusion evaluation model outputs muscle group status grading results, and the longitudinal comparison sub-module calculates the annual rate of change of muscle group status and generates a trend prediction report. S4. Based on the quantitative data from muscle group detection, the system intelligently matches and outputs corresponding dietary nutrition and home activity health care guidance plans; S5. All detection data, evaluation results, and health care plans are automatically archived, and historical data can be retrieved for subsequent review to dynamically update evaluation results and trends.

[0005] Beneficial effects 1. Targeted Innovation in Detection: Breaking through the limitations of traditional limb muscle group detection, it focuses specifically on the three core muscle groups: gluteus maximus, erector spinae, and pectoralis major, achieving automated and precise quantification of the microstructural parameters of local muscle groups, filling the technological gap in refined intelligent detection of specific core muscle groups. 2. Standardized Innovation in Detection: Relying on deep learning algorithms, the system achieves automatic segmentation and parameter extraction of ultrasound images, completely eliminating reliance on manual interpretation and solving the problems of strong subjectivity and poor repeatability of traditional detection results, thus achieving standardized and accurate output of detection data. 3. Dynamic tracking innovation: Establish a longitudinal data comparison mechanism to quantify the rate of change in muscle group status and predict development trends, upgrading from single-time detection to long-term dynamic monitoring, and meeting the long-term health and wellness monitoring needs of middle-aged and elderly people. 4. Multimodal Fusion Innovation: Integrating three modalities of data—ultrasound microstructure detection, body composition bioparameter detection, and depth visual posture detection—it achieves a multi-dimensional comprehensive assessment from microscopic muscle structure to macroscopic posture, resulting in more comprehensive and accurate assessment results. 5. Closed-loop service innovation: It integrates data collection, intelligent assessment, solution matching, and long-term follow-up, addressing the industry pain point of the disconnect between traditional testing and health guidance, and achieving an intelligent closed-loop health service. 6. Wide range of applications: The device is not a medical or health care device; its output is only for daily health reference and does not involve disease diagnosis or treatment. It can be widely used in homes, community health centers, and elderly care institutions, demonstrating strong practicality and universality. Attached Figure Description

[0006] Figure 1 This is a block diagram of the overall system architecture of the present invention. Figure 2 Assessment-Recommendation Closed-Loop Workflow Diagram Figure 3 Schematic diagram of region of interest segmentation for ultrasound target muscle group Figure 4 Schematic diagram of the status assessment level output interface Detailed Implementation

[0007] The present invention will be further described in detail below with reference to the embodiments. Example 1 A smart matching system for assessing and managing the condition of the gluteus maximus, erector spinae, and pectoralis major muscles based on multimodal sensor fusion is proposed. The hardware components are mounted on a health and wellness workstation host: a wideband adaptive high-frequency linear ultrasound probe (frequency range 7–15MHz, adaptively adjustable); a multi-frequency, multi-electrode body composition module (frequency points: 5kHz, 50kHz, 250kHz, 500kHz); an RGBD depth camera (depth resolution 640×480@30fps, angle measurement accuracy ≤2°); an industrial control host with a built-in AI image segmentation network (based on U-Net architecture, with ResNet-34 as the backbone network) and a multi-parameter fusion evaluation model (based on random forest algorithm, fusing four-dimensional features including muscle thickness, grayscale mean, high-echo area proportion, and body posture offset); and a touch-screen interactive display for the workstation. AI Model Training and Validation The image segmentation network was trained using a dataset containing at least 3000 ultrasound images of the gluteus maximus, erector spinae, and pectoralis major muscles. All images were manually annotated by at least three radiologists with over five years of experience in musculoskeletal ultrasound as a reference standard. The segmentation network achieved a Dice coefficient ≥0.85 on the validation set. The multi-parameter fusion evaluation model was calibrated using the muscle cross-sectional area measured by MRI and the proportion of hyperechoic regions as reference standards. The accuracy of each physical parameter extraction was verified, with an intraclass correlation coefficient (ICC) ≥0.80 with the MRI measurement results. Method for calculating the proportion of high echo regions Gray-level statistics were performed on the pixels within the segmented muscle region. A dynamic gray-level threshold was set as the mean gray-level value within the muscle region plus 1.5 times the standard deviation. The percentage of hyperechoic regions (%) was calculated as the sum of the areas of connected regions above this threshold relative to the total area of ​​the muscle region. This method can effectively distinguish hyperechoic structures within muscle from normal muscle tissue echoes and shows good correlation with fat signal regions measured by MRI. Subject testing procedure: 1) Subjects entered basic information such as age, height, weight, and daily sedentary time on a touchscreen; 2) Operators used an adaptive high-frequency ultrasound probe to sequentially acquire ultrasound images of the gluteus maximus, erector spinae, and pectoralis major muscles—the gluteus maximus was acquired in the prone position via the midpoint of the line connecting the posterior superior iliac spine and the greater trochanter of the femur; the erector spinae was acquired in the prone position via the L3–L4 spinous processes 2cm lateral to them; and the pectoralis major was acquired in the supine position via the midclavicular line at the 4th intercostal space—an AI image segmentation network automatically measured the muscle thickness of each muscle group (accurate to 0.1mm), extracted the mean echo grayscale value (0–255 grayscale value) and the percentage of hyperechoic areas (%); 3) Subjects stood with both feet on a multi-electrode component platform, and multi-frequency impedance and phase angle data were acquired in 15 seconds; 4) An RGBD camera scanned and acquired pelvic tilt angle and thoracic posture data, and calculated the posture offset (°); 5) AI The feature extraction and state assessment unit integrates all feature data and outputs structural feature grading results: 0 / Ⅰ / Ⅱ / Ⅲ levels. It calculates the annual rate of change of muscle thickness (mm / year) and outputs the estimated trend of muscle thickness change in the next 1–3 years; 6) Based on quantitative parameters such as muscle thickness value and the proportion of hyperechoic areas, the system automatically matches daily nutritional dietary references (such as examples of high-quality protein foods and examples of calcium-rich food combinations) and home activity guidance programs (such as graphic guidance on gluteus maximus activation training, erector spinae stretching training, and pectoralis major flexibility training), which are pushed and displayed through the touch screen and can be exported and printed; 7) All collected reports and health care reference programs are saved to personal electronic files, and historical data are automatically retrieved for comparison during the next review. Example 2 As a workstation expansion module, this system can be independently configured as a desktop health and wellness data acquisition terminal. Without the configuration of physiotherapy hardware, it can only perform functions such as collecting muscle group physical characteristics, pushing health and wellness reference plans, and managing records, making it suitable for use in community health and wellness centers and elderly care institutions.

Claims

1. A multimodal sensor fusion-based intelligent matching system for core muscle group status assessment and rehabilitation programs, characterized in that, The device comprises four units: a multimodal data acquisition unit, an AI feature extraction and status assessment unit, a longitudinal comparison and storage unit, and a health and wellness solution intelligent matching and output unit. These four units are integrated into a single health and wellness workstation. The multimodal data acquisition unit includes an adaptive high-frequency musculoskeletal ultrasound probe, a multi-frequency multi-electrode body component detection module, an RGBD depth vision camera, and a basic vital sign recording module. An adaptive high-frequency musculoskeletal ultrasound probe acquires muscle thickness, mean ultrasound echo grayscale value, and proportion of hyperechoic areas in the gluteus maximus, erector spinae, and pectoralis major muscles. A multi-frequency, multi-electrode body component detection module acquires total skeletal muscle mass, bioimpedance, and phase angle as evaluation benchmark parameters. An RGBD depth vision camera acquires thoracic morphology and pelvic tilt angle and quantifies postural deviation. A basic vital sign entry module inputs age, height, weight, and average daily sedentary time as auxiliary features. The AI ​​feature extraction and state evaluation unit incorporates an image segmentation network module, a multi-parameter fusion evaluation model, and a longitudinal comparison submodule. The image segmentation network uses a U-Net architecture with ResNet-34 as the backbone network to achieve semantic segmentation of the target muscle group, removes fat and skeletal interference signals, and extracts four-dimensional quantitative indicators: muscle thickness, mean echo grayscale value, proportion of hyperechoic areas, and postural deviation. The multi-parameter fusion evaluation model uses a random forest algorithm to fuse and analyze the four-dimensional indicators, classifying the degenerative state of the muscle group into 0, I, II, and III. Four levels; the longitudinal comparison submodule calculates the annual rate of change in muscle group thickness and predicts the trend of muscle group status changes in the next 1-3 years; the longitudinal comparison storage unit establishes a personal health and wellness electronic file, storing original ultrasound images, quantitative parameters, grading results, and trend curves, supporting cross-period data comparison, export, and follow-up records, realizing dynamic tracking of core muscle groups throughout their entire life cycle; the health and wellness plan intelligent matching output unit binds muscle group grading results and quantitative parameters, automatically pushes non-medical personalized dietary nutrition plans, home stretching and activation training guidance for the three core muscle groups, and automatically generates follow-up reminder plans, forming a closed loop of health and wellness services: detection—assessment—plan push—long-term follow-up.

2. The multimodal sensor fusion-based intelligent matching system for core muscle group status assessment and rehabilitation programs according to claim 1, characterized in that: The adaptive high-frequency musculoskeletal ultrasound probe is a broadband adaptive linear array probe with an adjustable frequency range of 7MHz to 15MHz and sub-millimeter spatial resolution.

3. The intelligent matching system for core muscle group status assessment and rehabilitation program based on multimodal sensor fusion as described in claim 1, characterized in that: The multi-frequency multi-electrode bulk component detection module is set with four detection frequency points: 5kHz, 50kHz, 250kHz, and 500kHz.

4. The intelligent matching system for core muscle group status assessment and rehabilitation program based on multimodal sensor fusion as described in claim 1, characterized in that: The RGBD depth vision camera has a depth resolution of 640×480@30fps, and the measurement accuracy of pelvic and thoracic angles is ≤2°.

5. The intelligent matching system for core muscle group status assessment and rehabilitation program based on multimodal sensor fusion as described in claim 1, characterized in that: The image segmentation network was trained using at least 3,000 annotated ultrasound images of the three core muscle groups, with a validation set Dice coefficient ≥ 0.

85. The multi-parameter fusion evaluation model was calibrated using MRI detection results, with an intra-group correlation coefficient (ICC) ≥ 0.80 for parameter extraction.

6. The intelligent matching system for core muscle group status assessment and rehabilitation program based on multimodal sensor fusion as described in claim 1, characterized in that: The method for calculating the proportion of hyperechoic regions in the muscle is as follows: perform pixel grayscale statistics on the effective muscle region after AI segmentation, use the mean grayscale value of the muscle region plus 1.5 times the standard deviation as a dynamic threshold, and divide the total area of ​​connected regions above the threshold by the overall area of ​​the muscle group to obtain the percentage of hyperechoic regions.

7. The intelligent matching system for core muscle group status assessment and rehabilitation program based on multimodal sensor fusion as described in claim 1, characterized in that: The sampling point for the gluteus maximus in the prone position is the midpoint of the line connecting the posterior superior iliac spine and the greater trochanter of the femur. The sampling point for the erector spinae in the prone position is 2 cm lateral to the spinous processes of L3-L4. The sampling point for the pectoralis major in the supine position is at the 4th intercostal space along the midclavicular line.

8. The intelligent matching system for core muscle group status assessment and rehabilitation program based on multimodal sensor fusion as described in claim 1, characterized in that: The entire set of equipment is a non-medical diagnostic and health care device. All output solutions are only for reference in daily health life and are not used for disease diagnosis, lesion identification and clinical treatment.

9. The intelligent matching system for core muscle group status assessment and rehabilitation program based on multimodal sensor fusion as described in claim 1, characterized in that: It can be disassembled into an independent desktop data acquisition terminal. After removing the hardware for health and wellness implementation, it retains only the functions of data acquisition, AI assessment, health and wellness plan push and electronic record management, and is suitable for batch deployment and use in community health and wellness centers and elderly care institutions.

10. A method for intelligent matching of core muscle group status assessment and rehabilitation program based on the system described in any one of claims 1 to 9, characterized in that, Includes the following steps: S1. Subjects enter their basic personal information through the basic vital signs entry module and then complete the ultrasound image acquisition of the three major muscle groups, multi-frequency body component impedance detection, and RGBD body posture three-dimensional scanning in sequence. S2. The system automatically segments the region of interest of the target muscle group through the U-Net deep learning network and extracts four standardized and quantified physical parameters; S3. The random forest multi-parameter fusion model outputs the muscle group degeneration classification level, calculates the annual thickness decay rate of the muscle group, and generates a medium- and long-term trend prediction report. S4. The system automatically matches dietary plans and home muscle group training exercise guidance based on the grading results and various quantitative indicators, and outputs them visually. S5. All test data, assessment conclusions, and health and wellness guidance plans will be automatically archived into your personal electronic file. The next review will retrieve historical data to iteratively update the assessment results and trends.

Images