A multi-modal objective assessment device and method for fatigue symptoms in general practice outpatient clinics

Abstract

The present application relates to the technical field of medical devices and artificial intelligence, in particular to a multi-modal objective assessment device and method for fatigue symptoms in general outpatient clinics, which synchronously collects eyelid conjunctival images, surface electromyography signals, heart rate variability signals and voice signals of patients through a multi-modal physiological signal collection module, extracts multi-dimensional feature parameters after adaptive preprocessing, inputs a multi-modal fusion model with a dynamic weight adjustment mechanism, comprehensively generates fatigue risk stratification results and cause bias prompts, and finally outputs a structured assessment report and provides clinical decision support. The present application solves the problem of fatigue assessment in the prior art, which completely relies on subjective scales, has poor accuracy and low standardization, and realizes non-invasive, rapid and objective quantitative assessment of fatigue symptoms, which can effectively assist general practitioners in differential diagnosis, improve diagnosis and treatment efficiency and accuracy, and is suitable for wide application in general outpatient clinics of medical institutions at all levels.

Description

Technical Field

[0001] This invention relates to the field of medical device and artificial intelligence interdisciplinary technology, specifically a multimodal objective assessment device and method for fatigue symptoms in general practice outpatient clinics. Background Technology

[0002] Fatigue is one of the most common chief complaints in clinical practice. Statistics show that approximately 15%-30% of general practice outpatients present with fatigue as their primary reason for seeking medical attention. Fatigue can be caused by a variety of diseases, encompassing hematologic disorders, neuromuscular diseases, endocrine and metabolic disorders, cardiovascular diseases, infectious diseases, and functional / psychological disorders. In general practice, early identification and accurate assessment of fatigue symptoms are crucial for early diagnosis, differential diagnosis, and evaluation of treatment effectiveness.

[0003] Currently, clinical assessment of fatigue symptoms mainly relies on subjective scales and physician interviews. Commonly used subjective scales include the Fatigue Assessment Scale (FAS), the Fatigue Severity Scale (FSS), and the Multidimensional Fatigue Scale-20 (MFI-20). While these scales are easy to use, they have significant limitations: First, the assessment results are highly dependent on the patient's subjective feelings and expressive abilities; different patients have significant differences in their understanding and description of "fatigue," leading to highly subjective and poor repeatability of the assessment results. Second, existing scales cannot distinguish between fatigue caused by different etiologies, making it difficult to provide targeted differential diagnostic clues. Third, scale assessments cannot achieve quantitative longitudinal tracking, making it difficult to objectively evaluate treatment effects and disease progression.

[0004] In recent years, some studies have attempted to use physiological signal measurement techniques to objectively assess fatigue symptoms. For example, some studies have used surface electromyography (sEMG) to assess muscle fatigue levels, quantifying muscle strength decline by analyzing the time and frequency domain characteristics of electromyographic signals; others have used heart rate variability (HRV) to analyze autonomic nervous system function, indirectly assessing psychogenic or functional fatigue. However, these single-modal assessment methods can only reflect one aspect of the pathophysiological mechanism of fatigue and cannot comprehensively cover the multiple causes of fatigue. For example, a simple sEMG examination cannot identify fatigue caused by anemia, and a simple HRV analysis cannot distinguish fatigue caused by neuromuscular diseases.

[0005] To address the limitations of single-modal assessment, some scholars have proposed the idea of ​​multimodal data fusion. For example, some studies have integrated multimodal data such as heart rate, skin temperature, and acceleration to assess physical fatigue. However, this method is mainly for assessing exercise fatigue in healthy individuals and is not suitable for assessing pathological fatigue in clinical patients. Some patent literature discloses chronic fatigue syndrome (CFS) monitoring devices based on multimodal physiological signals, but these devices are mainly designed for long-term home monitoring. They are complex and cumbersome to operate, making them unsuitable for the needs of rapid screening in general practice clinics.

[0006] In addition, existing technologies also have the following problems: First, it has high requirements for the acquisition environment. For example, the acquisition of eyelid conjunctival images requires strict standard lighting conditions, but the lighting in the actual outpatient environment is complex and variable, which seriously affects the accuracy of the measurement. Second, the signal acquisition process is easily affected by motion artifacts, especially when the patient performs standardized movements, the artifacts generated by limb movements will significantly reduce the quality of electromyographic signals. Third, existing multimodal fusion models typically use fixed weights, which cannot adapt to the physiological differences of patients of different ages, genders, and body mass indexes, resulting in large fluctuations in assessment accuracy across different populations. Fourth, most existing equipment requires integration with hospital information systems (HIS / LIS), which results in high deployment costs and long cycles, limiting its widespread application in primary healthcare institutions.

[0007] Therefore, there is an urgent need to develop a device and method that can rapidly, non-invasively, objectively, and comprehensively assess fatigue symptoms in a general practice setting, to solve the problems of strong subjectivity, low accuracy, poor standardization, and difficult deployment in existing technologies, and to provide general practitioners with an effective clinical decision support tool. Summary of the Invention

[0008] To address the problems of existing technologies that rely entirely on subjective scales for fatigue assessment, resulting in poor accuracy and low standardization; single-modal assessment methods cannot comprehensively cover multiple causes of fatigue; multimodal assessment methods have high environmental requirements, poor anti-interference ability, and cannot adapt to differences among different populations; and complex equipment deployment is unsuitable for use in primary care clinics, this invention provides a multimodal objective assessment device and method for fatigue symptoms in general practice clinics.

[0009] The technical solution adopted by this invention to solve its technical problem is: a multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics, comprising the following steps: S1: Acquire images of the patient's lower eyelid conjunctiva under non-standard lighting conditions, perform adaptive lighting compensation and precise segmentation of vascular regions on the images, extract the reflectance changes of hemoglobin characteristic spectrum, and calculate and generate an anemia risk index; S2: The wearable wireless sensor synchronously collects the surface electromyography (EMG) signal and six-axis inertial motion signal of the patient during the standardized movement process. Based on the inertial motion signal, the surface EMG signal is adaptively eliminated to remove motion artifacts, and the time domain and frequency domain feature parameters are extracted to calculate and generate the muscle weakness index. S3: Collect photoelectric pulse wave signals and voice signals of patients in a resting state, extract heart rate variability parameters and voice emotion feature parameters respectively, and fuse them to generate an autonomic nervous system function state index; S4: Obtain the patient's demographic characteristics and dynamically adjust the weight coefficients of each feature dimension in the multimodal fusion model based on this information; S5: Input the anemia risk index, muscle weakness index, and autonomic nervous system function status index into the multimodal fusion model after weight adjustment to generate weakness risk stratification results and etiological bias suggestions; S6: Generates a structured assessment report that includes comparison charts of various indicators, risk stratification results, confidence levels, and clinical decision recommendations, and provides report interpretation in natural language and auxiliary content for doctor-patient communication.

[0010] Specifically, the adaptive illumination compensation in step S1 is as follows: using a multi-scale illumination estimation algorithm based on Retinex theory, the illumination component and the reflection component in the image are separated, and the illumination component is nonlinearly corrected and then re-fused with the reflection component to obtain a conjunctival image with uniform illumination.

[0011] Specifically, the precise segmentation of the vascular region in step S1 involves: using an improved U-Net neural network model, applying the vascular segmentation technology of fundus OCT images to eyelid conjunctival images, automatically identifying and segmenting the conjunctival vascular region, and calculating the hemoglobin spectral reflectance only within the vascular region.

[0012] Specifically, the adaptive elimination of motion artifacts in step S2 involves: establishing a correlation model between inertial motion signals and surface electromyography (EMG) signals, and using an adaptive noise cancellation algorithm to filter out EMG signal interference components caused by limb movement in real time.

[0013] Specifically, the speech emotion feature parameters mentioned in step S3 include the fundamental frequency mean, fundamental frequency range, energy mean, energy variance, speech rate, and pause frequency. These parameters are extracted by a pre-trained speech emotion recognition model and quantified into a psychological state score.

[0014] Specifically, the demographic information mentioned in step S4 includes age, gender, and body mass index. Based on the physiological differences among different groups, a weight coefficient adjustment matrix is ​​pre-established to achieve dynamic adaptation of model weights.

[0015] Specifically, the multimodal fusion model in step S5 adopts a deep neural network structure with an attention mechanism, which can automatically learn the contribution of different feature dimensions to the weakness assessment and output the weakness risk stratification results of low risk, medium risk and high risk.

[0016] A multimodal objective assessment device for fatigue symptoms in general practice outpatient clinics includes: The multimodal physiological signal acquisition module is used to acquire images of the patient's eyelid conjunctiva, surface electromyography signals, inertial motion signals, photoelectric pulse wave signals, and voice signals. The signal preprocessing module is connected to the multimodal physiological signal acquisition module and is used to perform adaptive preprocessing on the acquired signals. The feature extraction and fusion module, connected to the signal preprocessing module, is used to extract feature parameters of various signals and perform multimodal fusion. The risk stratification assessment module, connected to the feature extraction and fusion module, is used to generate weak risk stratification results and etiological bias indications. The report generation and output module, connected to the risk stratification assessment module, is used to generate and output structured assessment reports.

[0017] Specifically, the multimodal physiological signal acquisition module includes: The multispectral imaging unit, consisting of a high-resolution camera and a multi-wavelength LED light source array, is used to acquire images of the eyelid conjunctiva; The electromyography-inertial fusion acquisition unit is composed of a surface electromyography sensor and a six-axis inertial measurement unit, and is used to simultaneously acquire surface electromyography signals and inertial motion signals; The photoelectric pulse wave acquisition unit is used to acquire photoelectric pulse wave signals; The voice acquisition unit, consisting of a high-fidelity microphone, is used to acquire voice signals.

[0018] Specifically, it also includes a data storage and management module, which stores patient assessment data from each assessment, establishes an individualized baseline database, supports longitudinal tracking of changes in fatigue levels, and enables follow-up comparison of assessment results.

[0019] The beneficial effects of this invention are: It achieves multi-dimensional objective quantitative assessment of fatigue symptoms, transforming patients' subjective feelings of fatigue into a quantifiable and comparable objective indicator system, thus solving the problems of strong subjectivity and poor repeatability in existing fatigue assessment technologies.

[0020] It covers the three most common causes of fatigue in multiple dimensions: anemia, neuromuscular diseases, and functional / psychological disorders, providing general practitioners with clear clues for differential diagnosis and improving the efficiency and accuracy of diagnosis and treatment.

[0021] By employing adaptive illumination compensation and precise vascular segmentation technology, the challenge of acquiring eyelid and conjunctival images under non-standard lighting conditions was solved, thus improving the accuracy of anemia risk assessment.

[0022] By employing electromyography-inertial fusion acquisition and motion artifact adaptive elimination technology, the problem of motion interference in electromyography signals during standardized movements is effectively solved, thus improving the reliability of muscle strength decline assessment.

[0023] By integrating voice emotion features with heart rate variability parameters, a more comprehensive assessment of autonomic nervous system function and psychological state is achieved, improving the accuracy of recognizing functional / psychological weakness.

[0024] By introducing a dynamic weight adjustment mechanism based on demographic characteristics, the model can adapt to the physiological differences among different populations, significantly improving the accuracy and generalization ability of the evaluation results.

[0025] It adopts a completely independent hardware design with zero data interface dependency, is simple to deploy and easy to use, and is suitable for widespread application in general practice clinics of medical institutions at all levels.

[0026] It provides structured assessment reports and natural language interpretation, as well as doctor-patient communication assistance functions, which can effectively reduce doctors' workload and improve the efficiency of doctor-patient communication. Attached Figure Description

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] Figure 1 This invention provides a system architecture diagram of a multimodal objective assessment device for fatigue symptoms in general practice outpatient clinics. Figure 2 A flowchart of a multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics provided by this invention; Figure 3 This is an architectural diagram of the multimodal physiological signal acquisition module in this invention; Figure 4 This is a schematic diagram of the multimodal fusion model network in this invention. Detailed Implementation

[0029] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0030] like Figures 1-4 As shown, the present invention provides a multimodal objective assessment device and method for fatigue symptoms in general practice outpatient clinics. Through multimodal physiological signal synchronous acquisition, adaptive signal preprocessing, and dynamic weighted multimodal fusion, it achieves objective, comprehensive, and accurate assessment of fatigue symptoms.

[0031] A multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics includes the following steps: S1: Acquire images of the patient's lower eyelid conjunctiva under non-standard lighting conditions, perform adaptive lighting compensation and precise segmentation of vascular regions on the images, extract the reflectance changes of hemoglobin characteristic spectra, and calculate and generate an anemia risk index.

[0032] The adaptive illumination compensation employs a multi-scale illumination estimation algorithm based on Retinex theory, which effectively eliminates interference from uneven ambient lighting, strong light reflection, and shadows, resulting in a uniformly illuminated conjunctival image. Precise vascular region segmentation utilizes an improved U-Net neural network model, transferring mature fundus OCT image vascular segmentation technology from the medical imaging field to eyelid conjunctival images. This automatically identifies and precisely segments the conjunctival vascular region, calculating hemoglobin spectral reflectance only within the vascular region, avoiding interference from non-vascular tissues and significantly improving the accuracy of hemoglobin estimation.

[0033] S2: The surface electromyography (EMG) signal and six-axis inertial motion signal of the patient are collected synchronously through wearable wireless sensors during the standardized movement process. Based on the inertial motion signal, the surface EMG signal is adaptively eliminated to remove motion artifacts, and the time domain and frequency domain feature parameters are extracted to calculate and generate the muscle weakness index.

[0034] Among these techniques, adaptive motion artifact elimination establishes a correlation model between inertial motion signals and surface electromyography (EMG) signals, employing an adaptive noise cancellation algorithm to filter out EMG signal interference components caused by limb movements in real time. This technology draws inspiration from inertial navigation and signal fusion in the aerospace field, effectively addressing motion artifact problems arising when patients perform standardized movements such as clenching their fists and raising their legs, thereby improving the signal-to-noise ratio of EMG signals and the accuracy of muscle strength assessment.

[0035] S3: Collect photoelectric pulse wave signals and voice signals from the patient at rest, extract heart rate variability parameters and voice emotion feature parameters respectively, and fuse them to generate an autonomic nervous system function state index.

[0036] The speech emotion feature parameters include the fundamental frequency mean, fundamental frequency range, energy mean, energy variance, speech rate, and pause frequency. These parameters are extracted and quantified into a psychological state score using a pre-trained speech emotion recognition model. Integrating speech emotion features with heart rate variability parameters allows for a more comprehensive and accurate assessment of the patient's autonomic nervous system function and psychological state, improving the accuracy of identifying functional / psychological fatigue.

[0037] S4: Obtain demographic information such as the patient's age, gender, and body mass index, and dynamically adjust the weight coefficients of each feature dimension in the multimodal fusion model based on this information.

[0038] By pre-establishing weight coefficient adjustment matrices for different population groups, the model weights can be dynamically adapted. For example, for elderly patients, the weight of the muscle weakness index is appropriately increased; for female patients, the weight of the anemia risk index is appropriately increased; and for patients with a high body mass index, the weight of the autonomic nervous system function index is appropriately adjusted. This dynamic weight adjustment mechanism can effectively improve the model's generalization ability and assessment accuracy across different populations.

[0039] S5: Input the anemia risk index, muscle weakness index, and autonomic nervous system function status index into the multimodal fusion model after weight adjustment to generate weakness risk stratification results and etiological bias indications.

[0040] The multimodal fusion model employs a deep neural network structure based on an attention mechanism, which can automatically learn the contribution of different feature dimensions to fatigue assessment and output fatigue risk stratification results of low risk, medium risk, and high risk. At the same time, it provides etiological bias suggestions, such as "high probability of anemia-related fatigue", "neuromuscular dysfunction", and "autonomic nervous system dysfunction", providing doctors with clear directions for differential diagnosis.

[0041] S6: Generates a structured assessment report that includes comparison charts of various indicators, risk stratification results, confidence levels, and clinical decision recommendations, and provides report interpretation in natural language and auxiliary content for doctor-patient communication.

[0042] Clinical decision-making recommendations include directions for further investigation, referral suggestions, and lifestyle intervention recommendations. The report interpretation uses plain, easy-to-understand natural language to help doctors quickly comprehend the assessment results; the doctor-patient communication support content provides doctors with standardized scripts for communicating with patients, improving communication efficiency and patient compliance.

[0043] This invention also provides a multimodal objective assessment device for fatigue symptoms in general practice outpatient clinics, comprising: The multimodal physiological signal acquisition module is used to acquire images of the patient's eyelid conjunctiva, surface electromyography signals, inertial motion signals, photoelectric pulse wave signals, and voice signals. The signal preprocessing module is connected to the multimodal physiological signal acquisition module and is used to perform adaptive preprocessing on the acquired signals. The feature extraction and fusion module, connected to the signal preprocessing module, is used to extract feature parameters of various signals and perform multimodal fusion. The risk stratification assessment module, connected to the feature extraction and fusion module, is used to generate weak risk stratification results and etiological bias indications. The report generation and output module, connected to the risk stratification assessment module, is used to generate and output structured assessment reports; The data storage and management module is used to store patients' assessment data from previous assessments, establish an individualized baseline database, support longitudinal tracking of changes in fatigue levels, and enable follow-up comparison of assessment results.

[0044] The multimodal physiological signal acquisition module includes: The multispectral imaging unit, consisting of a high-resolution camera and a multi-wavelength LED light source array, is used to acquire images of the eyelid conjunctiva; The electromyography-inertial fusion acquisition unit is composed of a surface electromyography sensor and a six-axis inertial measurement unit, and is used to simultaneously acquire surface electromyography signals and inertial motion signals; The photoelectric pulse wave acquisition unit is used to acquire photoelectric pulse wave signals; The voice acquisition unit, consisting of a high-fidelity microphone, is used to acquire voice signals.

[0045] The device of this invention adopts a completely independent hardware design, does not need to be connected to any hospital information system, and can complete the entire evaluation process on a standalone machine. It is simple to deploy and easy to use, and is particularly suitable for promotion and application in primary general practice clinics.

[0046] Example 1 describes in detail the overall structure and workflow of the multimodal objective assessment device for fatigue symptoms for general practice outpatient clinics of the present invention.

[0047] The assessment device of this invention is designed as a portable all-in-one machine, measuring approximately 35cm × 25cm × 20cm and weighing no more than 5kg, making it easy to move and use between general practice clinics. The device features a touchscreen interface, a built-in embedded processing system, and a high-capacity battery, supporting continuous operation for over 8 hours, meeting the daily needs of outpatient use.

[0048] The core components of the device include: Multimodal physiological signal acquisition module Multispectral imaging unit: Composed of a 12-megapixel high-resolution CMOS camera and an LED light source array with four wavelengths: 450nm, 520nm, 630nm, and 850nm. The camera is equipped with autofocus and auto exposure functions, enabling it to quickly and clearly capture images of the patient's lower eyelid conjunctiva. The multi-wavelength LED light source array is evenly distributed around the camera, providing stable multispectral illumination.

[0049] Electromyography-Inertial Fusion Acquisition Unit: Employing a wireless wearable design, this unit integrates a surface electromyography (EMG) sensor and a six-axis inertial measurement unit (accelerometer + gyroscope) into a small patch sensor. The sensor is attached to the surface of the patient's forearm flexor muscles and quadriceps femoris muscles using medical conductive adhesive, enabling simultaneous acquisition of surface EMG signals and six-axis inertial motion signals. The sampling frequency is set to 1000Hz, and the wireless transmission distance is no less than 10 meters.

[0050] Photoelectric pulse wave acquisition unit: Adopting a finger clip design, it incorporates a green LED and photodiode, acquiring the photoelectric pulse wave signal from the patient's fingertip using photoplethysmography. The sampling frequency is set to 200Hz.

[0051] Voice acquisition unit: Built-in high-fidelity digital microphone with a sampling frequency of 16kHz, which can clearly acquire the patient's voice signal.

[0052] Signal preprocessing module The signal preprocessing module consists of multiple parallel preprocessing subunits, each preprocessing different types of signals: Image preprocessing subunit: responsible for adaptive illumination compensation, noise reduction, and vascular region segmentation of eyelid conjunctival images; Electromyography signal preprocessing subunit: responsible for filtering, amplifying and eliminating motion artifacts in surface electromyography signals; Pulse wave preprocessing subunit: responsible for filtering, baseline drift correction and peak detection of photoelectric pulse wave signals; Speech preprocessing subunit: responsible for pre-emphasizing, framing, windowing, and endpoint detection of speech signals.

[0053] Feature extraction and fusion module The feature extraction and fusion module is responsible for extracting effective feature parameters from the preprocessed signal and performing preliminary feature fusion. Specifically, this includes: Extracting hemoglobin spectral reflectance features at different wavelengths from eyelid conjunctival images; Time-domain and frequency-domain features such as root mean square (RMS), average rectified value (ARV), median frequency (MNF), and average power frequency (MPF) are extracted from surface electromyography signals. Normal sinus interval (NN interval) is extracted from photoelectric pulse wave signal, and heart rate variability parameters in time domain (SDNN, RMSSD, etc.) and frequency domain (LF, HF, LF / HF, etc.) are calculated. Extracting emotional feature parameters such as fundamental frequency, energy, speech rate, and pause frequency from speech signals; By fusing heart rate variability parameters with voice emotion feature parameters, an autonomic nervous system function state index is generated.

[0054] Risk stratification assessment module The risk stratification assessment module employs a deep neural network model based on an attention mechanism. Inputs include anemia risk index, muscle weakness index, and autonomic nervous system function status index, as well as demographic characteristics such as the patient's age, gender, and body mass index. The model first dynamically adjusts the weight coefficients of each feature dimension based on the demographic information, then performs feature fusion and classification through a multi-layer neural network, ultimately outputting the weakness risk stratification result (low risk / medium risk / high risk) and etiological bias indications.

[0055] Report generation and output module The report generation and output module is responsible for generating a structured evaluation report with illustrations and text. The report content includes: Patient basic information and assessment time; Comparison chart of measured values ​​of anemia risk index, muscle weakness index, and autonomic nervous system function status index with normal reference ranges; Results and confidence levels of risk stratification for weakness; Etiological bias suggests; Clinical decision-making recommendations, including directions for further investigation, referral recommendations, and lifestyle intervention recommendations; Report interpretation in natural language format and supplementary content for doctor-patient communication.

[0056] The report can be printed directly or exported as a PDF file.

[0057] Data storage and management module The data storage and management module has a built-in high-capacity solid-state drive for storing patients' basic information, previous assessment data, and assessment reports. The system can automatically establish an individualized baseline database for each patient, recording the indicators from their initial assessment as baseline values. When patients return for follow-up visits, the system can automatically compare the current assessment results with the baseline values ​​and previous assessment results, generating a trend chart of changes in fatigue levels to objectively evaluate the intervention effect.

[0058] The working process of the device in this embodiment is as follows: Patient registration: Doctors enter the patient's basic information on the touchscreen, including name, gender, age, weight, height, chief complaint, etc.

[0059] Signal acquisition: (1) Eyelid conjunctival image acquisition: The doctor instructs the patient to look down, gently flips the patient's lower eyelid with a cotton swab, aligns the multispectral imaging unit with the patient's lower eyelid conjunctiva, presses the acquisition button, and the system automatically captures multispectral images.

[0060] (2) Surface electromyography signal acquisition: The electromyography-inertial fusion acquisition unit is attached to the surface of the patient's forearm flexor muscles and quadriceps femoris muscles respectively. The patient is guided to complete the standardized fist clenching and leg raising movements in sequence. Each movement is repeated 3 times. The system simultaneously acquires surface electromyography signals and inertial motion signals.

[0061] (3) Photoelectric pulse wave and voice signal acquisition: Instruct the patient to sit and rest for 5 minutes, clip the finger clip photoelectric pulse wave acquisition unit on the patient's index finger, and instruct the patient to read a standard text. The system simultaneously acquires photoelectric pulse wave signal and voice signal.

[0062] Signal processing and evaluation: The system automatically preprocesses and extracts features from various acquired signals, inputs them into a multimodal fusion model for analysis, and generates weakness risk stratification results and evaluation reports.

[0063] Report output: Doctors review the assessment report and make further diagnoses and treatments based on the recommendations provided.

[0064] Data storage: The system automatically stores patient assessment data and reports into the database to establish patient health records.

[0065] The entire assessment process takes about 5-8 minutes, is easy to operate, and can be completed without the need for professional technicians, making it very suitable for the rapid screening needs of general practice clinics.

[0066] Example 2 describes in detail the specific methods for multimodal physiological signal acquisition and preprocessing in this invention, focusing on key technologies such as adaptive illumination compensation, precise segmentation of vascular regions, and adaptive elimination of motion artifacts.

[0067] I. Image Acquisition and Preprocessing of the Eyelid Conjunctiva Image acquisition of the palpebral conjunctiva is a crucial step in assessing anemia-related fatigue. Hemoglobin exhibits different absorption characteristics to different wavelengths of light, with the greatest differences in absorption to 450 nm blue light and 630 nm red light. Therefore, hemoglobin levels can be estimated by analyzing the spectral reflectance at these two wavelengths. However, lighting conditions in actual outpatient environments are complex and variable, and ambient light interference can significantly affect the accuracy of spectral reflectance measurements. Furthermore, palpebral conjunctival images include not only vascular areas but also non-vascular tissues such as conjunctival epithelium and meibomian glands. The spectral characteristics of these tissues differ from those of blood vessels, and failure to distinguish them can also affect the accuracy of hemoglobin estimation.

[0068] To address the aforementioned issues, this invention employs adaptive illumination compensation and precise segmentation techniques for blood vessel regions.

[0069] Adaptive illumination compensation This invention employs a multi-scale illumination estimation algorithm based on Retinex theory to perform illumination compensation on eyelid conjunctival images. Retinex theory posits that the color of an object is determined by its ability to reflect light of different wavelengths, independent of the intensity of the incident light. An image can be decomposed into the product of an illumination component and a reflection component: ; in, It is the original image. It is the light component. It is the reflection component.

[0070] The specific steps of adaptive illumination compensation are as follows: (1) Gaussian filtering is applied to the original image to obtain the illumination component estimates at different scales; (2) Weighted fusion of illumination component estimates at different scales is performed to obtain the final illumination components. ; (3) Regarding the light component Perform nonlinear correction to compress its dynamic range; (4) Transfer the original image Divide by the corrected illumination component The reflection component image R(x,y) with uniform illumination is obtained.

[0071] The above processing effectively eliminates interference from uneven ambient lighting, strong light reflection, and shadows, resulting in eyelid conjunctival images with uniform lighting and appropriate contrast, laying the foundation for subsequent vascular segmentation and spectral analysis.

[0072] Precise segmentation of vascular regions This invention employs an improved U-Net neural network model for accurate segmentation of the conjunctival vascular region of the eyelid. U-Net is a classic medical image segmentation model, consisting of an encoder and a decoder, and has advantages such as simple structure, small training data requirements, and high segmentation accuracy.

[0073] To improve segmentation accuracy, this invention makes the following improvements to the traditional U-Net model: (1) Introduce residual connections in the encoder part to solve the gradient vanishing problem in the training process of deep networks; (2) An attention mechanism is introduced into the decoder part to enable the model to pay more attention to the features of the vascular region; (3) A multi-scale input strategy is adopted to input images of different resolutions into the model, thereby improving the model's ability to segment blood vessels of different thicknesses.

[0074] The model training employs transfer learning. First, it is pre-trained on a large-scale fundus OCT image vascular segmentation dataset to learn the general features of blood vessels. Then, it is fine-tuned on the eyelid conjunctival image dataset collected in this invention to adapt the model to the characteristics of eyelid conjunctival blood vessels. After training, the model can automatically identify and accurately segment vascular regions in eyelid conjunctival images, achieving a segmentation accuracy of over 95%.

[0075] Hemoglobin spectral reflectance calculation and anemia risk index generation After segmenting the vascular region, the spectral reflectance at different wavelengths is calculated only within the segmented vascular region. Specifically, the average grayscale value of the vascular region at four wavelengths (450nm, 520nm, 630nm, and 850nm) is calculated, and then the hemoglobin index is calculated based on the spectral absorption characteristics of hemoglobin. ; in, It is the hemoglobin index (dimensionless). and These are the average reflectances at wavelengths of 630nm and 450nm, respectively.

[0076] Finally, the hemoglobin index is compared with the pre-established calibration curve to calculate the anemia risk index, which ranges from 0 to 100. The higher the value, the greater the risk of anemia.

[0077] II. Acquisition and Preprocessing of Surface Electromyography Signals Surface electromyography (EMG) signals are electrical activities generated during muscle contraction and can reflect the functional state of muscles. In weakness assessment, the degree of muscle weakness can be quantified by analyzing the EMG signal characteristics when patients perform standardized movements. However, when patients perform movements such as clenching their fists or raising their legs, the movement of the limbs and the deformation of the muscles produce motion artifacts. These artifacts are superimposed on the EMG signals, seriously affecting signal quality and the accuracy of feature extraction.

[0078] To address the motion artifact problem, this invention employs electromyography-inertial fusion acquisition and adaptive motion artifact elimination technology.

[0079] Electromyography-Inertial Fusion Acquisition This invention integrates a surface electromyography (EMG) sensor and a six-axis inertial measurement unit into a small patch sensor, enabling simultaneous acquisition of EMG signals and six-axis inertial motion signals (three-axis acceleration and three-axis angular velocity). Since motion artifacts are directly related to the movement state of a limb, inertial motion signals can be used to estimate and eliminate motion artifacts in EMG signals.

[0080] Adaptive elimination of motion artifacts This invention employs an adaptive noise cancellation algorithm to eliminate motion artifacts in surface electromyography (EMG) signals. The basic idea of ​​the adaptive noise cancellation algorithm is as follows: the noisy signal is used as the main input, and the noise-related signal is used as the reference input. An adaptive filter is used to adjust the weights of the reference input so that the output of the filter approximates the noise component in the main input as closely as possible. Then, the output of the filter is subtracted from the main input to obtain the denoised signal.

[0081] The specific steps for adaptive motion artifact removal are as follows: (1) Use surface electromyography signal as the main input Using the six-axis inertial motion signal as the reference input ; (2) Initialize the weight coefficients of the adaptive filter ; (3) Calculate the output of the filter ; (4) Calculate the error signal This error signal is the denoised electromyographic signal; (5) Update the weight coefficients according to the Least Mean Square (LMS) algorithm: , where μ is the step size factor; (6) Repeat steps (3)-(5) until the weight coefficients converge.

[0082] The above processing effectively filters out interference components in electromyographic (EMG) signals caused by limb movement, significantly improving the signal-to-noise ratio (SNR) of EMG signals. Experimental results show that this method can improve the SNR of EMG signals by more than 15 dB, effectively improving the accuracy of muscle strength decline assessment.

[0083] Electromyographic feature extraction and muscle weakness index generation Feature extraction was performed on the denoised surface electromyography (EMG) signal. The extracted time-domain features included root mean square (RMS), average rectified value (ARV), and integrated electromyography (iEMG), while the frequency-domain features included median frequency (MNF), average power frequency (MPF), and peak frequency (PF).

[0084] Then, the extracted feature parameters are input into a pre-trained muscle strength assessment model to calculate and generate a muscle strength decline index, with a value ranging from 0 to 100. A higher value indicates more significant muscle strength decline. The muscle strength assessment model uses a support vector machine (SVM) algorithm and is trained on a large amount of clinical data, enabling it to accurately distinguish between normal muscle strength and different degrees of muscle strength decline.

[0085] III. Acquisition and Preprocessing of Photoelectric Pulse Waves and Speech Signals Photoelectric pulse wave signal acquisition and preprocessing A finger-clip photoelectric pulse wave acquisition unit was used to acquire photoelectric pulse wave signals from patients at rest for 5 minutes. The acquired signals were preprocessed, including 50Hz power frequency filtering, 0.5-40Hz bandpass filtering, baseline drift correction, and peak detection, to extract the normal sinus interval (NN interval) sequence.

[0086] Then, the time-domain and frequency-domain parameters of heart rate variability are calculated: Time-domain parameters: SDNN (standard deviation of all NN intervals), RMSSD (root mean square of the difference between adjacent NN intervals), NN50 (number of adjacent NN intervals with a difference greater than 50ms), pNN50 (percentage of NN50 to the total number of NN intervals), etc. Frequency domain parameters: LF (low-frequency power, 0.04-0.15Hz), HF (high-frequency power, 0.15-0.4Hz), LF / HF (ratio of low-frequency power to high-frequency power), etc.

[0087] Speech signal acquisition and preprocessing The system guides the patient to read aloud a standard text of approximately 30 seconds, and the speech signal is captured through a built-in microphone. The captured speech signal is preprocessed, including pre-emphasis, framing, windowing, and endpoint detection, to remove silent and noisy segments.

[0088] Then, the emotional feature parameters of the speech are extracted, including: Fundamental frequency characteristics: fundamental frequency mean, fundamental frequency range, fundamental frequency standard deviation; Energy characteristics: mean energy, variance energy, energy range; Speech rate characteristics: average speech rate, maximum speech rate, minimum speech rate; Pause characteristics: number of pauses, average pause duration, and longest pause duration.

[0089] The extracted speech emotion feature parameters are input into a pre-trained speech emotion recognition model to obtain a psychological state score, with a value range of 0-100. The higher the value, the worse the psychological state, and the greater the possibility of emotional problems such as anxiety and depression.

[0090] Generation of Autonomic Nervous Functional State Index Heart rate variability parameters and psychological state scores were normalized, and then a weighted fusion method was used to generate an autonomic nervous system function index, with a value range of 0-100. Higher values ​​indicate more severe autonomic nervous system dysfunction. The weight of the heart rate variability parameter was 0.7, and the weight of the psychological state score was 0.3.

[0091] Example 3 details the construction method of the multimodal fusion model, the dynamic weight adjustment mechanism, and the specific process of the weak risk stratification assessment in this invention, and verifies the effectiveness of this invention through clinical trials.

[0092] I. Construction of Multimodal Fusion Model This invention employs a deep neural network model based on an attention mechanism for multimodal feature fusion and fatigue risk stratification assessment. The model's input includes three dimensions of features: anemia risk index, muscle weakness index, and autonomic nervous system function status index, as well as three demographic characteristics: patient age, gender, and body mass index. The model's output is the fatigue risk stratification result (low risk / medium risk / high risk) and etiological bias indication.

[0093] The model structure is as follows: Input layer: Receives 6 input features, namely anemia risk index, muscle weakness index, autonomic nervous system function status index, age, gender, and body mass index.

[0094] Feature preprocessing layer: Normalizes the input features, mapping all features to the range of 0-1, thus eliminating the influence of different feature dimensions.

[0095] Dynamic weight adjustment layer: The weight coefficients of the three physiological characteristic dimensions are dynamically adjusted based on the patient's demographic characteristics (age, gender, body mass index).

[0096] Attention mechanism layer: Employs a self-attention mechanism to automatically learn the contribution of different feature dimensions to the weak evaluation and generate a weighted feature vector.

[0097] Hidden layers: consist of three fully connected layers, each containing 64 neurons, using the ReLU activation function to extract higher-order features.

[0098] Output layer: consists of two branches: Risk stratification branch: Using the Softmax activation function, output the probabilities of three categories: low risk, medium risk, and high risk; Etiological bias branch: Using the Sigmoid activation function, output the probabilities of three etiological dimensions: anemia-related, neuromuscular-related, and autonomic-related.

[0099] The model was trained using supervised learning. The training dataset contained 1000 clinical samples, including 700 patients with fatigue and 300 healthy controls. All samples were diagnosed by clinical experts and labeled with their fatigue risk level and primary cause. The Adam optimization algorithm was used during training, with a learning rate of 0.001, a batch size of 32, and 100 training epochs. Dropout and L2 regularization were employed to prevent overfitting.

[0100] II. Dynamic Weight Adjustment Mechanism The physiological characteristics and common causes of fatigue vary significantly among people of different ages, sexes, and body mass indexes. For example, elderly patients are more prone to fatigue caused by neuromuscular dysfunction; female patients have a significantly higher incidence of anemia than male patients due to physiological reasons; and obese patients are more likely to experience autonomic nervous system dysfunction. Therefore, fixed-weight multimodal fusion models cannot adapt to the differences among different populations, leading to large fluctuations in assessment accuracy across different groups.

[0101] To address this issue, this invention introduces a dynamic weight adjustment mechanism based on demographic characteristics. The specific method is as follows: First, based on a large amount of clinical data, we analyzed the distribution of various causes of fatigue in people of different ages, genders, and body mass indexes; Then, a three-dimensional weight coefficient adjustment matrix is ​​established, with the three dimensions corresponding to age, gender, and body mass index, respectively. During model runtime, the corresponding weight coefficients are retrieved from the weight coefficient adjustment matrix based on the input patient demographic characteristics. Finally, the anemia risk index, muscle weakness index, and autonomic nervous system function index are weighted using the retrieved weight coefficients and then input into subsequent network layers for processing.

[0102] The weighting coefficient adjustment matrix was established using a data-driven approach, obtained through statistical analysis of the training dataset. For example, for male patients over 60 years of age, the weight of the muscle weakness index was set to 0.5, the weight of the anemia risk index was set to 0.25, and the weight of the autonomic nervous system function index was set to 0.25; for female patients aged 30-50 years of age, the weight of the anemia risk index was set to 0.5, the weight of the muscle weakness index was set to 0.25, and the weight of the autonomic nervous system function index was set to 0.25.

[0103] Through this dynamic weight adjustment mechanism, the model can automatically adjust the importance of each feature dimension according to the characteristics of different patients, thereby significantly improving the accuracy and generalization ability of the assessment results.

[0104] III. Risk Stratification of Fatigue and Clinical Decision-Making Recommendations The output of the multimodal fusion model includes weakness risk stratification results and etiological bias indications.

[0105] Weak risk stratification Based on the probabilities of the three categories output by the risk stratification branch, patients are divided into three levels: low risk, medium risk, and high risk. Low risk: A low risk probability ≥0.7 indicates that the patient has mild fatigue symptoms and no obvious signs of organic lesions; Medium risk: A medium risk probability ≥ 0.7, or a low risk probability < 0.7 and a high risk probability < 0.7, indicates that the patient's fatigue symptoms are moderate and further examination is needed to determine the cause. High risk: A high risk probability ≥0.7 indicates that the patient's fatigue symptoms are severe and there is a greater possibility of obvious organic lesions, requiring timely further examination or referral.

[0106] Etiological bias suggestion Based on the probabilities of the three etiological dimensions output by the etiological bias branch, etiological bias hints are given: If the probability of anemia correlation is ≥0.7, it suggests that "the possibility of anemia correlation is high, and it is recommended to complete a complete blood count." If the neuromuscular correlation probability is ≥0.7, it indicates "significantly abnormal muscle weakness index, and further investigation of nervous system or muscle diseases is recommended"; If the autonomic nervous system correlation probability is ≥0.7, it indicates "HRV suggests autonomic nervous system dysfunction, and it is recommended to pay attention to sleep and psychological factors"; If the probabilities of multiple dimensions are all between 0.4 and 0.7, it suggests that "fatigue may be caused by a combination of factors, and a comprehensive examination is recommended."

[0107] Clinical decision recommendations Based on the results of fatigue risk stratification and etiological bias indications, the system automatically generates corresponding clinical decision-making suggestions, including: Further examinations may be needed, such as complete blood count, electrolytes, thyroid function tests, and electromyography. Referral recommendations: such as referring to hematology, neurology, psychology, etc. Lifestyle intervention recommendations include: a balanced diet, regular sleep patterns, moderate exercise, and psychological adjustment.

[0108] To verify the effectiveness of this invention, we conducted a clinical trial at the general practice clinic of a tertiary hospital, including 200 patients with fatigue as the chief complaint and 50 healthy volunteers as controls. All subjects were evaluated by the device of this invention, and two experienced general practitioners conducted subjective assessments using the Fatigue Assessment Scale (FAS), with the final clinical diagnosis used as the gold standard.

[0109] The experimental results are shown in the table below:

[0110] The experimental results show that the accuracy, sensitivity and specificity of the method of the present invention are significantly higher than those of the traditional subjective scale assessment method, with a Kappa value of 0.82, indicating a high degree of consistency with the clinical gold standard.

[0111] In addition, we analyzed the assessment accuracy of different populations. The results showed that the assessment accuracy of the method of the present invention remained above 88% in different age groups, genders and body mass index groups, which was significantly better than the fixed-weight multimodal fusion method (accuracy fluctuated between 75% and 85%), verifying the effectiveness of the dynamic weight adjustment mechanism.

[0112] Comparison with Example 1: A comparison with traditional subjective scale assessments Traditional fatigue assessments primarily rely on subjective scales, such as the Fatigue Assessment Scale (FAS) and the Fatigue Severity Scale (FSS). These scales are typically rated by patients based on their own subjective feelings, and while convenient, they have significant limitations.

[0113] To compare the differences between the present invention and traditional subjective scale assessments, we selected 50 patients from the aforementioned clinical trial and assessed them using the method of the present invention and the FAS scale, with the final clinical diagnosis as the gold standard.

[0114] The results showed that the FAS scale had an accuracy rate of 70%, with 12 patients being misdiagnosed. Further analysis revealed that the misdiagnoses mainly occurred in the following situations: Elderly patients: Due to cognitive decline, their understanding and responses to scale questions may be biased, leading to inaccurate scores; Patients significantly affected by psychological factors: Some patients may exaggerate or underestimate their fatigue symptoms due to psychological factors such as anxiety and depression; Early-stage mild fatigue patients: The patients do not have obvious subjective feelings, but there are already abnormalities in objective physiological indicators that cannot be identified by scales.

[0115] The method of this invention, by objectively measuring the patient's physiological signals, is unaffected by the patient's subjective feelings and cognitive abilities, and can accurately identify patients with early-stage mild fatigue and differentiate fatigue caused by different etiologies. In these 50 patients, the assessment accuracy of the method of this invention reached 92%, with only 4 misdiagnoses, all of which were borderline cases.

[0116] Furthermore, the method of this invention can provide clear clues to etiological bias, offering doctors clues for differential diagnosis, whereas traditional scales can only provide an overall fatigue score and cannot provide etiological information.

[0117] Comparison with Example 2: A comparison with single-modal evaluation methods Single-modal assessment methods can only reflect one aspect of the pathophysiological mechanism of weakness and cannot comprehensively cover the multiple causes of weakness. To compare the differences between the present invention and single-modal assessment methods, we evaluated the above 200 patients using single anemia risk assessment, single muscle weakness assessment, and single autonomic nerve function assessment, and compared the results with the multimodal fusion assessment of the present invention.

[0118] The results are shown in the table below:

[0119] The results show that while the single-modal assessment method has a high recognition rate for its corresponding etiological type, it has a very low recognition rate for other types of weakness, with an overall accuracy rate of no more than 50%. In contrast, the multimodal fusion assessment method of this invention integrates information from three dimensions, achieving a recognition rate of over 90% for all types of weakness and an overall accuracy rate of 91.5%, which is significantly better than the single-modal assessment method.

[0120] Comparison with Example 3: A comparison with existing multimodal evaluation methods While some existing multimodal assessment methods integrate multiple physiological signals, they suffer from drawbacks such as high environmental requirements, poor anti-interference capabilities, and inability to adapt to differences among different populations. To compare the differences between this invention and existing multimodal assessment methods, we selected an existing multimodal physical fatigue assessment method based on heart rate, skin temperature, and acceleration, and conducted comparative tests under the same clinical trial conditions.

[0121] The results show that the overall accuracy of existing multimodal evaluation methods is 76.5%, significantly lower than the 91.5% of this invention. Further analysis reveals that the errors in existing methods mainly originate from the following aspects: Ambient lighting interference: The accuracy of physiological signal acquisition by existing methods is significantly reduced under non-standard lighting conditions; Motion artifact interference: Current methods lack effective measures to eliminate motion artifacts, resulting in a significant decrease in signal quality when patients perform standardized movements; Fixed-weight model: Existing methods use a fixed-weight fusion model, which cannot adapt to the physiological differences of different populations, and the assessment accuracy is significantly reduced in elderly patients and female patients.

[0122] The present invention effectively solves the above problems by employing techniques such as adaptive illumination compensation, adaptive elimination of motion artifacts, and dynamic weight adjustment, and significantly improves the accuracy and robustness of the evaluation.

[0123] The multimodal objective assessment device and method for fatigue symptoms in general practice clinics described in this invention can be industrially manufactured into a portable all-in-one device, which can be widely used in general practice clinics, community health service centers, and physical examination centers at all levels of medical institutions. This invention's device is easy to operate, provides rapid assessment, and yields accurate results. It can effectively assist general practitioners in the early identification and differential diagnosis of fatigue symptoms, improving diagnostic efficiency and accuracy, and has significant clinical application value and broad market prospects.

[0124] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics, characterized in that, Includes the following steps: S1: Acquire images of the patient's lower eyelid conjunctiva under non-standard lighting conditions, perform adaptive lighting compensation and precise segmentation of vascular regions on the images, extract the reflectance changes of hemoglobin characteristic spectrum, and calculate and generate an anemia risk index; S2: The wearable wireless sensor synchronously collects the surface electromyography (EMG) signal and six-axis inertial motion signal of the patient during the standardized movement process. Based on the inertial motion signal, the surface EMG signal is adaptively eliminated to remove motion artifacts, and the time domain and frequency domain feature parameters are extracted to calculate and generate the muscle weakness index. S3: Collect photoelectric pulse wave signals and voice signals of patients in a resting state, extract heart rate variability parameters and voice emotion feature parameters respectively, and fuse them to generate an autonomic nervous system function state index; S4: Obtain the patient's demographic characteristics and dynamically adjust the weight coefficients of each feature dimension in the multimodal fusion model based on this information; S5: Input the anemia risk index, muscle weakness index, and autonomic nervous system function status index into the multimodal fusion model after weight adjustment to generate weakness risk stratification results and etiological bias suggestions; S6: Generates a structured assessment report that includes comparison charts of various indicators, risk stratification results, confidence levels, and clinical decision recommendations, and provides report interpretation in natural language and auxiliary content for doctor-patient communication.

2. The multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics according to claim 1, characterized in that: The adaptive illumination compensation in step S1 specifically involves: using a multi-scale illumination estimation algorithm based on Retinex theory to separate the illumination component and the reflection component in the image; performing nonlinear correction on the illumination component and then re-fusing it with the reflection component to obtain a conjunctival image with uniform illumination.

3. The multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics according to claim 1, characterized in that: The precise segmentation of the vascular region in step S1 specifically involves: using an improved U-Net neural network model, transferring the vascular segmentation technology of fundus OCT images to eyelid conjunctival images, automatically identifying and segmenting the conjunctival vascular region, and calculating the hemoglobin spectral reflectance only within the vascular region.

4. The multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics according to claim 1, characterized in that: The adaptive elimination of motion artifacts in step S2 specifically involves: establishing a correlation model between inertial motion signals and surface electromyography (EMG) signals, and using an adaptive noise cancellation algorithm to filter out EMG signal interference components caused by limb movement in real time.

5. The multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics according to claim 1, characterized in that: The speech emotion feature parameters mentioned in step S3 include the fundamental frequency mean, fundamental frequency range, energy mean, energy variance, speech rate, and pause frequency. These parameters are extracted by a pre-trained speech emotion recognition model and quantified into a psychological state score.

6. The multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics according to claim 1, characterized in that: The demographic information mentioned in step S4 includes age, gender, and body mass index. Based on the physiological differences among different groups, a weight coefficient adjustment matrix is ​​pre-established to achieve dynamic adaptation of model weights.

7. The multimodal objective assessment method for fatigue symptoms in general practice outpatient clinics according to claim 1, characterized in that: The multimodal fusion model described in step S5 adopts a deep neural network structure with an attention mechanism, which can automatically learn the contribution of different feature dimensions to the weakness assessment and output the weakness risk stratification results of low risk, medium risk and high risk.

8. A multimodal objective assessment device for fatigue symptoms in general practice outpatient clinics, characterized in that, include: The multimodal physiological signal acquisition module is used to acquire images of the patient's eyelid conjunctiva, surface electromyography signals, inertial motion signals, photoelectric pulse wave signals, and voice signals. The signal preprocessing module is connected to the multimodal physiological signal acquisition module and is used to perform adaptive preprocessing on the acquired signals. The feature extraction and fusion module, connected to the signal preprocessing module, is used to extract feature parameters of various signals and perform multimodal fusion. The risk stratification assessment module, connected to the feature extraction and fusion module, is used to generate weak risk stratification results and etiological bias indications. The report generation and output module, connected to the risk stratification assessment module, is used to generate and output structured assessment reports.

9. A multimodal objective assessment device for fatigue symptoms in general practice outpatient clinics according to claim 8, characterized in that: The multimodal physiological signal acquisition module includes: The multispectral imaging unit, consisting of a high-resolution camera and a multi-wavelength LED light source array, is used to acquire images of the eyelid conjunctiva; The electromyography-inertial fusion acquisition unit is composed of a surface electromyography sensor and a six-axis inertial measurement unit, and is used to simultaneously acquire surface electromyography signals and inertial motion signals; The photoelectric pulse wave acquisition unit is used to acquire photoelectric pulse wave signals; The voice acquisition unit, consisting of a high-fidelity microphone, is used to acquire voice signals.

10. A multimodal objective assessment device for fatigue symptoms in general practice outpatient clinics according to claim 8, characterized in that: It also includes a data storage and management module, which stores patient assessment data from each assessment, establishes an individualized baseline database, supports longitudinal tracking of changes in fatigue levels, and enables follow-up comparison of assessment results.

Images