Rheumatism early diagnosis and treatment system and method based on remote monitoring function

Abstract

The invention relates to the technical field of remote diagnosis and treatment of rheumatism, in particular to an early diagnosis and treatment system and method for rheumatism based on a remote monitoring function. And extracting characteristic indexes such as joint movement range variability, a stiffness diffusion mode, a local temperature abnormal region and a pressure asymmetry coefficient. And inputting the indexes into a risk assessment model to obtain individualized risk levels and pathological sign labels, and automatically generating a dynamic monitoring plan comprising monitoring items, frequency and early warning threshold adjustment schemes according to the individualized risk levels and pathological sign labels. And finally, based on plan execution feedback, generating a periodic risk assessment report and a diagnosis and treatment suggestion draft. According to the method, accurate identification of the early risk of rheumatism and dynamic personalized adjustment of the monitoring strategy are realized.

Description

Technical Field

[0001] This invention relates to the field of remote diagnosis and treatment technology for rheumatic diseases, and in particular to a system and method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function. Background Technology

[0002] Currently, the early detection and long-term management of rheumatic diseases mainly rely on patients' subjective symptom descriptions and regular in-hospital examinations. While conventional remote health monitoring technologies can achieve remote collection and transmission of physiological data, their application in early warning of rheumatic diseases has limitations. Existing technologies typically collect single or limited-dimensional vital sign data. These data are independent and lack correlation, failing to effectively reflect the complex synergistic changes between joint biomechanical state and local inflammatory response in the early progression of rheumatic diseases. The inherent connections between multidimensional data are severed, leading to insufficient ability to identify early atypical pathological changes.

[0003] Existing remote monitoring solutions mostly employ pre-set fixed monitoring frequencies and uniform alarm thresholds. This static monitoring logic cannot adapt to the dynamic evolution of early pathological states in rheumatic diseases. Monitoring strategies are disconnected from individual patients' real-time risk levels and specific physical signs, lacking the ability to self-adjust and optimize based on changes in the condition. This leads to missed detection of latent early risks or unnecessary monitoring redundancy, making it difficult to achieve accurate and efficient early risk stratification management and intervention resource allocation. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose an early diagnosis and treatment system and method for rheumatic diseases based on remote monitoring function.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an early diagnosis and treatment method for rheumatic diseases based on remote monitoring function, comprising: Acquire multimodal physiological data collected by the patient's wearable device, including periodic joint range of motion sequences, morning stiffness duration records, body surface temperature distribution maps, and joint pressure distribution data; The multimodal physiological data are processed to extract rheumatic pathological features, generating biomechanical and thermal imaging feature indicators related to early rheumatic diseases. The biomechanical and thermal imaging feature indicators include joint range of motion variability, stiffness diffusion pattern, local temperature anomaly area, and pressure asymmetry coefficient. The biomechanical and thermal imaging characteristics are input into a pre-constructed early risk assessment model for rheumatic diseases to generate individualized early risk level assessment results for rheumatic diseases and key pathological signs labels. Based on the early risk level assessment results of rheumatic diseases and key pathological signs, a dynamic monitoring plan is generated, which includes monitoring items, monitoring frequency and early warning threshold adjustment schemes. Based on the execution feedback data of the aforementioned dynamic monitoring plan, periodic risk assessment reports and draft treatment recommendations are generated for remote medical terminals to access.

[0006] As a further aspect of the present invention, the step of extracting rheumatic pathological features from the multimodal physiological data to generate biomechanical and thermal imaging feature indicators related to early rheumatic diseases includes: The periodic joint range of motion sequence is divided into multiple monitoring periods according to the time dimension, and the standard deviation of the maximum joint range of motion within each monitoring period is calculated to obtain the variability of the joint range of motion. The duration of morning stiffness is recorded and processed using pattern recognition. The differences in the onset time and the rate of regression of stiffness between symmetrical joints of the body are analyzed to generate the stiffness diffusion pattern that describes the migration law of stiffness. Identify continuous regions with a sustained temperature difference from healthy regions from the body surface temperature distribution map, extract the boundary morphology, temperature gradient, and diurnal temperature fluctuation value of the continuous regions, and construct the local temperature anomaly zone. Based on the joint pressure distribution data, the pressure peak ratio and pressure distribution center of gravity offset of the corresponding joints on the left and right sides of the body under the same movement mode are calculated to obtain the pressure asymmetry coefficient that reflects the uneven joint load. The pressure asymmetry coefficient is composed of the peak pressure ratio and the centroid offset of the pressure distribution, and is calculated in the following way: The peak pressure ratio of the target joints on the left and right sides of the body under the same standardized movement, extracted from the joint pressure distribution data, is nonlinearly weighted and fused with the pressure distribution center of gravity offset. The peak pressure ratio reflects the difference between the two joints when bearing the maximum load, and the pressure distribution center of gravity offset reflects the asymmetry between the two joints in the spatial position of the load distribution. After normalizing the peak pressure ratio and the centroid offset of the pressure distribution, the weighted sum of squares of the two is calculated and square rooted using a preset fusion function to obtain the pressure asymmetry coefficient. The numerical range of this coefficient is positively correlated with the degree of joint load imbalance.

[0007] As a further aspect of the present invention, the step of inputting the biomechanical and thermal imaging characteristic indicators into a pre-constructed early risk assessment model for rheumatic diseases to generate individualized early risk level assessment results and key pathological sign labels for rheumatic diseases includes: The variability of the joint range of motion is compared with the baseline data of healthy people of the same age group in a standardized manner to generate a joint function degeneration rate score. The standardized comparison of the joint function degeneration rate score adopts the Z-score method. The joint function degeneration rate score is the product of the absolute value of the Z-score value and the preset sensitivity coefficient. The rigid diffusion pattern is matched with pattern templates in the typical rheumatic disease pathological process library to generate a rigid pattern conformity score. The rigid pattern conformity score is obtained by calculating the cosine similarity between the feature vector of the rigid diffusion pattern to be evaluated and the feature vector of each predefined template in the typical rheumatic disease pathological process library. The highest similarity value is taken as the base of the rigid pattern conformity score, and then weighted and corrected according to the pattern duration and the number of joints involved. The characteristic parameters of the local temperature anomaly area are input into the inflammation activity prediction sub-model to generate a potential local inflammation activity intensity prediction value. The inflammation activity prediction sub-model is a regression model trained based on the gradient boosting decision tree algorithm. The characteristic parameters include the area of ​​the local temperature anomaly area, the average temperature difference, the maximum temperature gradient, and the diurnal fluctuation variance. The model output is a prediction value in the range of 0 to 1. The pressure asymmetry coefficient is compared with the joint injury risk threshold to generate a joint structural imbalance risk score. The score is calculated using a piecewise linear function. When the pressure asymmetry coefficient is below the first threshold, the score is 0. When it is between the first threshold and the second threshold, the score increases linearly. When it is above the second threshold, the score is full. By combining the joint function degeneration rate score, stiffness pattern conformity score, potential local inflammatory activity intensity prediction value, and joint structural imbalance risk score, a weighted decision algorithm is used to generate a quantitative early risk level assessment result for rheumatic diseases. The abnormal indicator with the highest contribution is output as the key pathological sign label. The contribution is determined by calculating the percentage of each input score multiplied by its corresponding weight coefficient relative to the comprehensive risk score. The indicator with the highest percentage is marked as the key pathological sign label.

[0008] As a further aspect of the present invention, the step of generating a dynamic monitoring plan based on the early risk level assessment results of rheumatic diseases and key pathological sign labels, including monitoring items, monitoring frequency, and early warning threshold adjustment schemes, includes: Based on the different intervals of the early risk level assessment results of rheumatic diseases, the combination of data collection items of the wearable device is dynamically adjusted. The high risk level corresponds to more comprehensive joint range of motion and pressure monitoring, while the medium risk level focuses on temperature and stiffness monitoring. Based on the specific body part indicated by the key pathological sign label, the monitoring frequency is set in a personalized manner. For parts where pathological sign labels have appeared, the data collection density is increased, while for parts where no signs have appeared, the conventional monitoring density is used. Based on the changing trends of the biomechanical and thermal imaging characteristic indicators in historical monitoring data, an individualized early warning threshold adjustment scheme is set for each characteristic indicator, and the early warning threshold adjustment scheme ensures a balance between early warning sensitivity and specificity.

[0009] As a further aspect of the present invention, the step of generating periodic risk assessment reports and draft treatment recommendations for remote medical terminals based on the execution feedback data of the dynamic monitoring plan includes: All multimodal physiological data and calculated biomechanical and thermal imaging characteristic indicators generated during the execution of the dynamic monitoring plan are periodically summarized to form a structured time series dataset; The structured time series dataset of the current period is compared and analyzed longitudinally with the data of historical periods to calculate the rate of change and trend stability of each characteristic indicator. Based on the rate of change and trend stability, the early risk level assessment results of rheumatic diseases are updated, and the evolution of the risk level is presented in the form of a visual trend chart in the risk assessment report. Based on the updated risk level assessment results and the latest key pathological signs labels, combined with a pre-built diagnostic and treatment knowledge base, a draft of the diagnostic and treatment recommendations is automatically generated, which includes suggestions for further examinations, key points for lifestyle interventions, and a schedule for follow-up visits.

[0010] As a further aspect of the present invention, the step of calculating the ratio of peak pressure to the pressure distribution center of gravity shift of corresponding joints on the left and right sides of the body under the same movement mode based on the joint pressure distribution data includes: Extract the pressure data sequences of the target joints on the left and right sides of the body under a preset standardized movement from the joint pressure distribution data; Calculate the maximum pressure value of the pressure data sequence of the target joint on the left side of the body and the maximum pressure value of the pressure data sequence of the target joint on the right side of the body, respectively. The pressure peak ratio is obtained by calculating the ratio of the maximum pressure value of the target joint on the left side of the body to the maximum pressure value of the target joint on the right side of the body. The center of gravity analysis was performed on the pressure distribution data of the target joint on the left side of the body during the complete cycle of a standardized movement, and the coordinates of the center of gravity of the pressure distribution on the left side of the body were calculated. The center of gravity analysis was performed on the pressure distribution data of the target joint on the right side of the body during the complete cycle of a standardized movement, and the coordinates of the center of gravity of the pressure distribution on the right side of the body were calculated. Calculate the Euclidean distance between the coordinates of the center of gravity of the pressure distribution on the left side of the body and the coordinates of the center of gravity of the pressure distribution on the right side of the body, and normalize the Euclidean distance relative to the joint reference size to obtain the offset of the center of gravity of the pressure distribution.

[0011] As a further aspect of the present invention, the step of generating the quantified early risk level assessment result of rheumatic diseases through a weighted decision algorithm includes: A first weighting coefficient is assigned to the joint function degeneration rate score, a second weighting coefficient is assigned to the stiffness pattern conformity score, a third weighting coefficient is assigned to the predicted value of potential local inflammatory activity intensity, and a fourth weighting coefficient is assigned to the joint structural imbalance risk score. The joint function degeneration rate score is multiplied by the first weighting coefficient to obtain the first weighted score; The rigid pattern conformity score is multiplied by the second weighting coefficient to obtain the second weighted score; The predicted value of the potential local inflammatory activity intensity is multiplied by the third weighting coefficient to obtain the third weighted score; The joint structure imbalance risk score is multiplied by the fourth weighting coefficient to obtain the fourth weighted score; The first weighted score, the second weighted score, the third weighted score, and the fourth weighted score are summed to obtain a comprehensive risk score. The comprehensive risk score is mapped to a preset risk level division interval, which includes four levels: no risk, low risk, medium risk, and high risk. The corresponding level identifier is output as the quantitative early risk level assessment result of rheumatic diseases.

[0012] As a further aspect of the present invention, the method further includes: Receive and integrate subjective symptom descriptions and records of limitations in daily activities proactively reported by patients in the preset health log interface; Natural language processing is performed on the subjective symptom description text to extract symptom keywords, degree modifiers and occurrence frequency information, and then it is converted into a standardized symptom description vector. The standardized symptom description vector is fused with objective biomechanical and thermal imaging feature indicators collected through the wearable device to generate a rheumatic disease risk assessment correction result that integrates subjective and objective information. When the keywords in the standardized symptom description vector are highly consistent with the key pathological sign labels in terms of anatomical location and symptom nature, the weight of the key pathological sign labels in risk assessment is increased.

[0013] As a further aspect of the present invention, the step of performing natural language processing on the subjective symptom description text to extract symptom keywords, degree modifiers, and frequency information, and converting them into a standardized symptom description vector, includes: The subjective symptom description text was segmented and entity recognized using a pre-trained rheumatology medical dictionary, and specific words describing joint, muscle, and skin symptoms were identified as symptom keywords. Identify and quantify the adjectives and adverbs that modify the keywords of the symptoms, and map them to a preset severity level to obtain the quantified severity modifiers; The text is analyzed to extract descriptions of symptom onset time and duration, as well as frequency adverbs such as "frequently" and "occasionally," which are then converted into standardized frequency values ​​to form the occurrence frequency information. The symptom keywords, the degree modifiers, and the frequency of occurrence information are vectorized and encoded to generate the standardized symptom description vector.

[0014] As a further aspect of the present invention, the present invention also includes an early diagnosis and treatment system for rheumatic diseases based on remote monitoring function. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the above-mentioned early diagnosis and treatment method for rheumatic diseases based on remote monitoring function.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By performing temporal correlation analysis on periodic joint range of motion sequences and joint pressure distribution data, the pressure asymmetry coefficient is extracted. Furthermore, by spatiotemporally correlating records of morning stiffness duration with body surface temperature distribution maps, a stiffness diffusion pattern is defined. This data processing method achieves deep fusion of multi-dimensional physiological signals and precise characterization of specific pathological features. It can analyze the biomechanical and thermoimaging synergistic change patterns characterizing early lesions from previously fragmented data streams. This technical solution transforms subjective and vague clinical symptoms into objective and quantifiable composite characteristic indicators, providing risk assessment models with multi-dimensional discriminative criteria not available with traditional single indicators, thus enhancing early identification and differentiation capabilities.

[0016] Based on the key pathological signs labels output by the risk assessment model, a personalized monitoring plan is automatically generated and dynamically adjusted, including monitoring items, frequency, and warning thresholds. The system can adjust the monitoring intensity and alarm sensitivity for different data dimensions in real time according to the specific signs identified. This transforms monitoring behavior from a static, general mode to a dynamic, adaptive mode. The resource allocation of the monitoring system is optimized synchronously with the risk status, increasing monitoring density and warning sensitivity when the risk rises, and reducing the monitoring load when the status is stable, thereby improving the efficiency of remote monitoring and the accuracy of risk management. Attached Figure Description

[0017] Figure 1 This is a flowchart of the early diagnosis and treatment method for rheumatic diseases based on remote monitoring function according to the present invention; Figure 2 A flowchart for extracting and processing rheumatic pathological features; Figure 3 A flowchart for the operation of an early risk assessment model for rheumatic diseases; Figure 4 A normalized comparison bar chart of multiple feature indicators for remote monitoring of rheumatic diseases; Figure 5 A bar chart comparing subjective and objective data for early risk assessment of rheumatic diseases. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] Please see Figure 1The overall implementation scheme of the early diagnosis and treatment method for rheumatic diseases based on remote monitoring is as follows: Multimodal physiological data is acquired from wearable devices in patients. This multimodal physiological data includes periodic joint range of motion sequences, morning stiffness duration records, body surface temperature distribution maps, and joint pressure distribution data. Rheumatic pathological features are extracted from the multimodal physiological data to generate biomechanical and thermal imaging characteristic indicators related to early rheumatic diseases. These indicators include joint range of motion variability, stiffness diffusion patterns, local temperature anomalies, and pressure asymmetry coefficients. The biomechanical and thermal imaging characteristic indicators are input into a pre-constructed early risk assessment model for rheumatic diseases to generate individualized early risk level assessment results and key pathological sign labels. Based on the early risk level assessment results and key pathological sign labels, a dynamic monitoring plan is generated, including monitoring items, monitoring frequency, and early warning threshold adjustment schemes. Based on the execution feedback data of the dynamic monitoring plan, periodic risk assessment reports and draft treatment recommendations are generated for remote medical terminals.

[0021] See Figure 2In one embodiment of the present invention, rheumatic pathological features are extracted from multimodal physiological data to generate biomechanical and thermal imaging characteristic indicators related to early rheumatic diseases. This processing includes dividing the periodic joint range of motion sequence into multiple monitoring periods along the time dimension, calculating the standard deviation of the maximum joint range of motion within each monitoring period, and obtaining the variability of joint range of motion. Pattern recognition processing is performed on the duration of morning stiffness records to analyze the difference in the onset time and the difference in the rate of resolution of stiffness between symmetrical joints in the body, generating a stiffness diffusion pattern describing the migration pattern of stiffness. Continuous regions with a sustained temperature difference from healthy areas are identified from the body surface temperature distribution map, and the boundary morphology, temperature gradient, and diurnal temperature fluctuation values ​​of these continuous regions are extracted to construct local temperature anomaly zones. Based on joint pressure distribution data, the peak pressure ratio and the offset of the pressure distribution center of gravity of corresponding joints on the left and right sides of the body under the same movement pattern are calculated to obtain a pressure asymmetry coefficient reflecting uneven joint load. This process involves calculating the peak pressure ratio and pressure distribution center of gravity offset of corresponding joints on the left and right sides of the body under the same movement pattern. This includes extracting pressure data sequences generated by target joints on the left and right sides of the body under a preset standardized movement from joint pressure distribution data. The maximum pressure values ​​of the left and right target joint pressure data sequences are calculated separately. The ratio of the maximum pressure values ​​of the left and right target joints is then calculated to obtain the peak pressure ratio. Center of gravity analysis is performed on the pressure distribution data of the left target joint within the complete cycle of the standardized movement to obtain the coordinates of the left-side pressure distribution center of gravity. Similarly, center of gravity analysis is performed on the pressure distribution data of the right target joint within the complete cycle of the standardized movement to obtain the coordinates of the right-side pressure distribution center of gravity. The Euclidean distance between the coordinates of the left and right pressure distribution center of gravity is calculated and normalized relative to the joint reference dimensions to obtain the pressure distribution center of gravity offset.

[0022] The implementation involves extracting rheumatic pathological features from multimodal physiological data to generate biomechanical and thermal imaging indicators. This process is based on acquired periodic joint range of motion sequences, morning stiffness duration records, body surface temperature distribution maps, and joint pressure distribution data. Specifically, the periodic joint range of motion sequences are divided into multiple monitoring periods along the time dimension. The standard deviation of the maximum joint range of motion within each monitoring period is calculated, and the result is the joint range of motion variability. This calculation is achieved through a sliding time window, the size of which is set according to the data sampling frequency and clinical monitoring needs. Pattern recognition processing is performed on the morning stiffness duration records to analyze the onset time difference and resolution rate difference of stiffness between symmetrical joints, generating stiffness diffusion patterns describing the migration pattern of stiffness. The pattern recognition processing uses a temporal clustering algorithm to identify the propagation path and temporal dependency of stiffness between joint pairs. The continuous regions with a sustained temperature difference from healthy areas are identified from the body surface temperature distribution map. The boundary morphology, temperature gradient, and diurnal temperature fluctuation values ​​of the continuous regions are extracted to construct local temperature anomaly areas. The identification process is based on image segmentation technology, combined with thermal imaging threshold segmentation and region growing algorithm to locate abnormal temperature areas and quantify their morphological and thermal characteristics.

[0023] In some embodiments, the pressure peak ratio and pressure distribution center of gravity offset of corresponding joints on the left and right sides of the body under the same movement mode are calculated based on joint pressure distribution data to obtain a pressure asymmetry coefficient reflecting the uneven joint load. This pressure asymmetry coefficient is composed of the pressure peak ratio and the pressure distribution center of gravity offset. Specifically, it is calculated by nonlinearly weighting and fusing the pressure peak ratio of the target joints on the left and right sides of the body under the same standardized movement, extracted from the joint pressure distribution data, with the pressure distribution center of gravity offset. The pressure peak ratio reflects the difference in the degree of stress on both sides when bearing maximum load, while the pressure distribution center of gravity offset reflects the degree of asymmetry in the spatial position of the load distribution on both sides. After normalizing the pressure peak ratio and the pressure distribution center of gravity offset, the weighted sum of squares of the two is calculated using a preset fusion function, and the square root is taken to obtain the pressure asymmetry coefficient. The numerical range of this coefficient is positively correlated with the degree of uneven joint load. In the specific implementation, pressure data sequences generated by target joints on the left and right sides of the body under preset standardized movements are extracted from joint pressure distribution data. Standardized movements include walking, flexion / extension, or weight-bearing lifting, triggered synchronously by the wearable device's built-in motion recognition module. The maximum pressure values ​​of the left and right target joint pressure data sequences are calculated separately. The maximum pressure values ​​are obtained by applying a peak detection algorithm to the pressure data sequences to eliminate instantaneous noise interference. The ratio of the maximum pressure values ​​of the left and right target joints is calculated using division, and the result is expressed as a decimal or percentage. Centroid analysis is performed on the pressure distribution data of the left target joints within the complete cycle of the standardized movement to calculate the centroid coordinates of the left-side pressure distribution. Centroid analysis is achieved by calculating the weighted average position of the pressure distribution matrix, with the weights being the pressure values ​​at each sampling point. Centroid analysis is also performed on the pressure distribution data of the right target joints within the complete cycle of the standardized movement to calculate the centroid coordinates of the right-side pressure distribution. The calculation method is consistent with that of the left side to ensure coordinate system consistency. Calculate the Euclidean distance between the coordinates of the center of gravity of the pressure distribution on the left side of the body and the coordinates of the center of gravity of the pressure distribution on the right side of the body, and normalize the Euclidean distance relative to the joint reference size to obtain the offset of the center of gravity of the pressure distribution; the Euclidean distance is calculated by taking the square root of the sum of the squares of the differences between the two-dimensional or three-dimensional spatial coordinates, and the joint reference size is the calibration value of the anatomical width or length of the joint.

[0024] It can be understood that the formula for calculating the offset of the pressure distribution center of gravity is expressed as:

[0025] in: This indicates the coordinates of the center of gravity representing the pressure distribution on the left side of the body. This indicates the coordinates of the center of gravity representing the pressure distribution on the right side of the body. Indicates the joint reference dimensions; This represents the centroid offset of the normalized pressure distribution; the coordinate data comes from the spatial position mapping of the pressure sensor array, and the joint reference dimensions are obtained from the patient's individual anatomical parameter library or by using standardized average values.

[0026] In some embodiments, the extraction of local temperature anomaly regions further includes Fourier descriptor quantization of boundary morphology, radial basis function fitting of temperature gradient, and temporal variance calculation of diurnal temperature fluctuation values. These operations are integrated into the feature extraction pipeline and executed in parallel using a multi-threaded approach to improve processing efficiency. Optionally, the generation of rigid diffusion patterns can be combined with a hidden Markov model to model the state of the start time difference and decay rate difference, and the output state transition probability can be used as a pattern quantification index. Optionally, the calculation of joint range of motion variability can introduce a sliding window weight decay function to give higher weight to recent data in order to more sensitively capture dynamic changes in joint function. It is understood that the output of all feature extraction steps is stored in a structured data format for subsequent risk assessment model calls; the data format includes timestamps, feature values, confidence levels, and metadata fields to ensure data traceability and verifiability. In specific implementations, the entire feature extraction process is deployed on edge computing devices or cloud servers, and resources are allocated according to data volume and real-time requirements; the raw data collected by wearable devices is encrypted and transmitted to trigger the feature extraction process, and the process log records the input and output of each step of the operation for easy debugging and auditing.

[0027] See Figure 3In one embodiment of the present invention, biomechanical and thermal imaging characteristic indicators are input into a pre-constructed early risk assessment model for rheumatic diseases to generate individualized early risk level assessment results and key pathological sign labels for rheumatic diseases. This process includes: standardizing and comparing the variability of joint range of motion with benchmark data from healthy individuals of the same age group to generate a joint function degeneration rate score; matching the stiffness diffusion pattern with pattern templates in a typical rheumatic disease pathological process library to generate a stiffness pattern conformity score; inputting characteristic parameters of local temperature anomaly areas into an inflammation activity prediction sub-model to generate a predicted value for potential local inflammation activity intensity; comparing the pressure asymmetry coefficient with a joint injury risk threshold to generate a joint structural imbalance risk score; and combining the joint function degeneration rate score, stiffness pattern conformity score, predicted value for potential local inflammation activity intensity, and joint structural imbalance risk score, a weighted decision algorithm is used to generate a quantitative early risk level assessment result for rheumatic diseases, and the abnormal indicator with the highest contribution is output as a key pathological sign label. The algorithm generates a quantitative early risk assessment result for rheumatic diseases using a weighted decision-making process. This includes assigning a first weighting coefficient to the joint function degeneration rate score, a second weighting coefficient to the stiffness pattern conformity score, a third weighting coefficient to the predicted intensity of potential local inflammatory activity, and a fourth weighting coefficient to the joint structural imbalance risk score. The joint function degeneration rate score is multiplied by the first weighting coefficient to obtain a first weighted score. The stiffness pattern conformity score is multiplied by the second weighting coefficient to obtain a second weighted score. The predicted intensity of potential local inflammatory activity is multiplied by the third weighting coefficient to obtain a third weighted score. The joint structural imbalance risk score is multiplied by the fourth weighting coefficient to obtain a fourth weighted score. The first, second, third, and fourth weighted scores are summed to obtain a comprehensive risk score. This comprehensive risk score is mapped to a preset risk level division interval, which includes four levels: no risk, low risk, medium risk, and high risk. The corresponding level label is output as the quantitative early risk assessment result for rheumatic diseases.

[0028] The implementation involves inputting biomechanical and thermal imaging characteristics into a pre-constructed early risk assessment model for rheumatic diseases and generating individualized assessment results. The early risk assessment model is built upon a fusion of machine learning classification algorithms and a clinical expert knowledge rule base. In practice, the variability of joint range of motion is standardized and compared with benchmark data from healthy individuals of the same age group to generate a joint function degeneration rate score. The standardization comparison uses the Z-score method; the joint function degeneration rate score is the product of the absolute value of the Z-score and a preset sensitivity coefficient. A higher value indicates a greater deviation of the joint function degeneration rate from the healthy baseline. The stiffness diffusion pattern is matched with pattern templates in a typical rheumatic disease pathological process database to generate a stiffness pattern conformity score. The matching process calculates the cosine similarity between the feature vector of the stiffness diffusion pattern to be assessed and the feature vector of each predefined template in the database. The highest similarity value is taken as the base for the stiffness pattern conformity score, and then weighted and corrected based on the pattern duration and the number of joints involved.

[0029] In some embodiments, the feature parameters of the local temperature anomaly area are input into the inflammation activity prediction sub-model to generate a predicted value of the potential local inflammation activity intensity. The inflammation activity prediction sub-model is a regression model trained based on the gradient boosting decision tree algorithm. The feature parameters include the area of ​​the local temperature anomaly area, the average temperature difference, the maximum temperature gradient, and the diurnal fluctuation variance. The model output is a predicted value in the range of 0 to 1. The pressure asymmetry coefficient is compared with the joint injury risk threshold to generate a joint structural imbalance risk score. The joint injury risk threshold is determined based on biomechanical simulation and clinical retrospective data. The score is calculated using a piecewise linear function. When the pressure asymmetry coefficient is below the first threshold, the score is 0. When it is between the first and second thresholds, it increases linearly. When it is above the second threshold, it reaches the full score.

[0030] It is understood that a quantitative early risk assessment of rheumatic diseases is generated through a weighted decision algorithm, which integrates the joint function degeneration rate score, stiffness pattern conformity score, predicted intensity of potential local inflammatory activity, and joint structural imbalance risk score. The weighted decision algorithm assigns an independent weight coefficient to each score and performs a linear weighted summation. In specific implementation, a first weight coefficient is assigned to the joint function degeneration rate score, a second weight coefficient to the stiffness pattern conformity score, a third weight coefficient to the predicted intensity of potential local inflammatory activity, and a fourth weight coefficient to the joint structural imbalance risk score. The values ​​of the first, second, third, and fourth weight coefficients are jointly determined through feature importance analysis and the Delphi method used by clinical experts, and the sum of all coefficients is 1. The joint function degeneration rate score is multiplied by the first weight coefficient to obtain the first weighted score; the stiffness pattern conformity score is multiplied by the second weight coefficient to obtain the second weighted score; the predicted intensity of potential local inflammatory activity is multiplied by the third weight coefficient to obtain the third weighted score; and the joint structural imbalance risk score is multiplied by the fourth weight coefficient to obtain the fourth weighted score.

[0031] Optionally, the calculation formula for the weighted decision algorithm is expressed as follows:

[0032] in: This represents the overall risk score; Indicates the first weighting coefficient; A score indicating the rate of joint function deterioration; This represents the second weighting coefficient; Indicates the rigid pattern conformity score; This represents the third weighting coefficient; This represents a predicted value for the intensity of potential local inflammatory activity. This represents the fourth weighting coefficient; This indicates a risk score for joint structural imbalance.

[0033] In one specific implementation of the weighted decision algorithm, the weight coefficients , , , The method for setting up the model is as follows: First, a retrospective database containing multiple confirmed rheumatic disease patients and healthy controls is constructed. This database stores each individual's joint function degeneration rate score, stiffness pattern conformity score, predicted value of potential local inflammatory activity intensity, joint structural imbalance risk score, and final pathological stage label. In the model building phase, a tree-based ensemble learning algorithm, such as gradient boosting decision tree, is used. The four scores are used as input features, and the pathological stage label is used as the prediction target for model training. After training, the algorithm's built-in feature importance evaluation function is used to calculate the total contribution of each input feature to reducing impurity during decision tree splitting. The contributions of the four features are normalized and used as initial weight coefficients. A clinical consultation group composed of multiple rheumatology experts is organized to review and fine-tune the initial weight coefficients based on clinical pathology knowledge. For example, in the early stages of the disease, experts may believe that the predicted value of local inflammatory activity intensity, representing inflammatory activity, should be given a higher weight, while for patients who have entered the structural damage stage, the weight of the joint structural imbalance risk score needs to be increased accordingly. The final weight coefficients... , , , The weighted average of the results of the aforementioned objective feature importance analysis and expert knowledge correction is set and embedded in the risk assessment model. During use, the model can be periodically retrained based on newly accumulated clinical data to achieve dynamic iterative optimization of the weighting coefficients.

[0034] The specific calculation methods for each scoring item are as follows: The joint function degeneration rate score is obtained by standardizing and comparing the variability of joint range of motion within the current monitoring period with the corresponding indicators in a pre-constructed benchmark database of healthy individuals of the same age and gender. Specifically, the difference between the current patient's joint range of motion variability and the mean of the healthy benchmark database is calculated, and then divided by the standard deviation of the healthy benchmark database to obtain a standard score. This standard score is then mapped to the interval of 0 to 1 using a preset S-shaped function to obtain the final joint function degeneration rate score. A higher score indicates a faster rate of functional degeneration. The stiffness pattern conformity score is determined by calculating the dynamic time-normalized distance between the temporal feature vector extracted from the patient's morning stiffness duration record and multiple pattern templates in a typical rheumatic disease pathological process database. The reciprocal of the minimum distance is taken and normalized to obtain a conformity score between 0 and 1. A larger value indicates a higher degree of matching between the patient's stiffness diffusion pattern and the typical pathological process. The potential local inflammatory activity intensity prediction value is generated by an inflammatory activity prediction sub-model, which is a multi-input single-output deep neural network. Its input layer receives a feature parameter vector of the local temperature anomaly region, which includes four neurons: the area of ​​the anomaly region, the average temperature difference, the maximum temperature gradient, and the diurnal temperature fluctuation value. The network structure contains two fully connected hidden layers, each with 16 neurons, and the activation function is a modified linear unit. The output layer is a single neuron using a sigmoid activation function to compress the output value between 0 and 1, directly serving as a prediction of the potential intensity of local inflammatory activity. The joint structural imbalance risk score is calculated by comparing the pressure asymmetry coefficient with a joint injury risk threshold. This score is defined using a piecewise linear function: when the pressure asymmetry coefficient is lower than a preset first threshold, the score is 0; when the pressure asymmetry coefficient is higher than a preset second threshold, the score is 1; when the coefficient is between the first and second thresholds, the score is linearly positively correlated with the coefficient value.

[0035] In practice, the first, second, third, and fourth weighted scores are summed to obtain a comprehensive risk score. This summation is performed after all multiplications are completed, and the result is rounded to two decimal places. The comprehensive risk score is mapped to a preset risk level range, which includes four levels: no risk, low risk, medium risk, and high risk. The corresponding level identifier is output as a quantitative assessment result of the early risk level of rheumatic diseases. The mapping rule is a predefined segmented range; for example, a comprehensive risk score in the range [0, 0.2) corresponds to a no-risk level, in [0.2, 0.5) to a low-risk level, in [0.5, 0.8) to a medium-risk level, and in [0.8, 1.0] to a high-risk level. The abnormal indicator with the highest contribution is output as a key pathological sign label. The contribution is determined by calculating the percentage of each input score multiplied by its corresponding weight coefficient relative to the comprehensive risk score; the indicator with the highest percentage is marked as the key pathological sign label.

[0036] The early diagnosis and treatment method for rheumatic diseases based on remote monitoring proposed in this invention includes the following complete technical path from raw data collection to comprehensive risk scoring: Raw multimodal physiological data is acquired from wearable devices, specifically: for joint range of motion sensors, periodic joint range of motion sequences are acquired; for temperature sensor arrays, body surface temperature distribution maps are acquired; for pressure sensor arrays, joint pressure distribution data are acquired; and the duration of morning stiffness is recorded through patient logs.

[0037] The raw multimodal physiological data undergoes feature extraction processing, transforming it into composite feature indicators, specifically: The variability of joint range of motion is obtained by calculating the standard deviation of the maximum range of motion for each monitoring cycle of the periodic joint range of motion sequence; a stiffness diffusion pattern is generated by analyzing the difference in the onset and resolution time of morning stiffness between symmetrical joints; a local temperature anomaly zone is constructed by segmenting continuous regions with a sustained temperature difference from healthy regions from the body surface temperature distribution map and extracting their boundary morphology and temperature gradient; and a pressure asymmetry coefficient is obtained by calculating the ratio of peak pressure to the left and right corresponding joints under the same movement and the offset of the pressure distribution center of gravity.

[0038] The composite characteristic indicators are input into different sub-modules of the risk assessment model and converted into quantitative scores, specifically: The following methods are used to obtain a joint function degeneration rate score by comparing the variability of joint range of motion with a healthy baseline; a stiffness pattern conformity score by matching the stiffness diffusion pattern with a pathological database; a predicted value of potential local inflammatory activity intensity by inputting the characteristic parameters of local temperature anomaly areas into the inflammation activity prediction sub-model; and a joint structural imbalance risk score by comparing the pressure asymmetry coefficient with a risk threshold. These four scores are used as input to a weighted decision algorithm, which is a linear weighted summation model. The input layer receives four scores, and the corresponding weight coefficients are determined jointly by pre-training and expert correction. The summation layer adds the four weighted scores to obtain a comprehensive risk score. The output layer maps this comprehensive risk score to four risk levels—no risk, low risk, medium risk, and high risk—through a pre-defined segmented interval and outputs the result.

[0039] In some embodiments, the pattern templates in the typical rheumatic disease pathological process library include templates showing morning stiffness spreading from distal small joints to proximal large joints, templates showing symmetrical synchronous stiffness in multiple joints, and templates showing migratory asymmetric stiffness; each template is generated by clustering stiffness record data from historically diagnosed patients. Optionally, the joint injury risk threshold can be differentiated according to different joint types, for example, the second threshold for the knee joint is higher than the second threshold for the interphalangeal joints of the hand. It is understood that the process of the weighted decision algorithm generating a comprehensive risk score and key pathological sign labels is fully automated. The calculation module completes the calculation and mapping within milliseconds after receiving four input scores and writes the level identifier and label into the assessment report database. In specific implementations, the early risk assessment model for rheumatic diseases is periodically retrained using newly added clinically diagnosed data and corresponding multimodal feature data to dynamically update the weight coefficients and risk threshold division intervals, thereby achieving iterative optimization of the model.

[0040] In one embodiment of the invention, a dynamic monitoring plan is generated based on the early risk level assessment results of rheumatic diseases and key pathological sign labels. This plan includes monitoring items, monitoring frequency, and early warning threshold adjustment schemes. The generation process involves dynamically adjusting the data collection item combination of the wearable device according to the different risk level intervals of the early risk level assessment results. High-risk levels correspond to more comprehensive joint range of motion and pressure monitoring, while medium-risk levels focus on temperature and stiffness monitoring. The monitoring frequency is personalized according to the specific body parts indicated by the key pathological sign labels. Data collection density is increased for areas with existing pathological sign labels, while conventional monitoring density is used for areas without signs. Based on the changing trends of biomechanical and thermal imaging characteristic indicators in historical monitoring data, an individualized early warning threshold adjustment scheme is set for each characteristic indicator. This scheme ensures a balance between early warning sensitivity and specificity.

[0041] The implementation involves generating a dynamic monitoring plan based on the early risk level assessment results of rheumatic diseases and key pathological signs. This plan includes monitoring items, monitoring frequency, and adjustment schemes for warning thresholds. In practice, the data collection items of the wearable device are dynamically adjusted according to the different risk levels identified in the early risk assessment. High-risk levels correspond to more comprehensive joint range of motion and pressure monitoring, while medium-risk levels focus on temperature and stiffness monitoring. Risk levels are divided into four categories: high-risk, medium-risk, low-risk, and no-risk. Each level is mapped to a preset monitoring item configuration template. The system automatically loads and sends corresponding configuration instructions to the wearable device on the patient's end based on the assessment results. The monitoring frequency is personalized based on the specific body part indicated by the key pathological sign tags. Data collection density is increased for parts with existing pathological sign tags, while a standard monitoring density is used for parts without signs. The monitoring frequency is set based on a preset frequency matrix, which defines the recommended collection interval for different body parts under different risk signals. For example, for the wrist joint marked with a key pathological sign tag, the data collection density can be increased from the usual once every 4 hours to once every 1 hour.

[0042] Understandably, based on the changing trends of biomechanical and thermal imaging characteristic indicators in historical monitoring data, an individualized warning threshold adjustment scheme is set for each characteristic indicator. This scheme ensures a balance between warning sensitivity and specificity. In practice, the warning threshold adjustment scheme is established by analyzing the time-series data of each characteristic indicator for a specific patient within a past time window. The time-series data includes historical records of joint range of motion variability, stiffness diffusion pattern quantification, characteristic parameters of local temperature anomalies, and pressure asymmetry coefficient.

[0043] In some embodiments, dynamically adjusting the data acquisition items of the wearable device specifically manifests as follows: for high-risk assessment results, the system command simultaneously activates the joint range of motion sensor, pressure sensor array, temperature sensor, and stiffness recording module for synchronous acquisition of all items; for medium-risk assessment results, the system command only activates the temperature sensor and stiffness recording module, while reducing the acquisition frequency of the joint range of motion and pressure sensors or putting them into standby mode. The process of personalized monitoring frequency settings takes into account device battery life and patient compliance. While increasing the monitoring frequency of key areas, the system can correspondingly reduce the monitoring frequency of non-key areas to maintain overall energy balance. The monitoring frequency adjustment command is synchronized to the wearable device in real time via a wireless network.

[0044] Optionally, the early warning threshold adjustment scheme is generated using an adaptive algorithm. This algorithm dynamically calculates the upper and lower limits of the early warning for the next monitoring cycle based on the mean and variance of historical data of the feature indicators. The calculation formula for the early warning threshold adjustment scheme can be understood as follows:

[0045] in: This indicates the warning threshold set for the new monitoring cycle; This represents the arithmetic mean of the characteristic indicator within a historical time window; This represents the standard deviation of the characteristic indicator within the historical time window; This represents the coefficient determined from the lookup table based on the target warning specificity. This formula is used to calculate the upper and lower warning thresholds, respectively, by adjusting the coefficients. The size is used to balance the sensitivity and specificity of the early warning system.

[0046] In some embodiments, for body parts with existing pathological signs, increasing data acquisition density not only means shortening the acquisition interval but also includes initiating high-frequency acquisition at specific triggering events, such as patient-reported pain events or abnormal movement patterns detected by the device. After setting an individualized warning threshold adjustment scheme for each feature indicator, the system encapsulates the new threshold parameters as configuration information and pushes them to the wearable device and cloud monitoring backend via a secure communication link, enabling subsequent data streams to be compared with the latest threshold in real time and trigger warnings. In specific implementation, the complete configuration of the dynamic monitoring plan, including the monitoring item combination, monitoring frequency table, and warning threshold set, is serialized into a structured configuration file. This file has a unique version identifier and is associated with and stored with the early risk level assessment results of rheumatic diseases and key pathological signs of a specific patient. A new configuration file version is generated and device synchronization is triggered after each assessment update. Optionally, the warning threshold adjustment scheme undergoes a logical verification step before application. This step checks whether the new threshold significantly deviates from the group baseline range or the patient's own historical range. Adjustments exceeding a reasonable range are marked and submitted for manual review. In practice, the generation, issuance, and execution status of dynamic monitoring plans are logged and monitored by the central management platform. The platform ensures the traceability of plan changes and the confirmation receipt of equipment execution instructions, thus forming a closed-loop management process.

[0047] In one embodiment of the present invention, based on the execution feedback data of the dynamic monitoring plan, a periodic risk assessment report and draft treatment recommendations are generated for access by a remote medical terminal. This generation includes a periodic summary of all multimodal physiological data generated during the execution of the dynamic monitoring plan, as well as calculated biomechanical and thermal imaging characteristic indicators, forming a structured time series dataset. The current period's structured time series dataset is compared longitudinally with historical period data to calculate the rate of change and trend stability of each characteristic indicator. Based on the rate of change and trend stability, the early risk level assessment results for rheumatic diseases are updated, and the evolution of the risk level is presented in the risk assessment report as a visual trend chart. Based on the updated risk level assessment results and the latest key pathological signs, combined with a pre-built treatment knowledge base, a draft treatment recommendation is automatically generated, including suggestions for further examinations, key points for lifestyle interventions, and follow-up appointment schedules.

[0048] The implementation involves generating periodic risk assessment reports and draft treatment recommendations based on the execution feedback data of the dynamic monitoring plan. The execution feedback data includes all multimodal physiological data and biomechanical and thermal imaging indicators collected by wearable devices and processed through feature extraction during the plan's execution cycle. In practice, all multimodal physiological data and calculated biomechanical and thermal imaging indicators generated during the execution of the dynamic monitoring plan are periodically summarized to form a structured time-series dataset. The summary operation is automatically triggered according to a preset reporting cycle, set weekly or monthly. The system extracts all data records belonging to the same patient within the corresponding time period from distributed storage and cleans and reassembles them according to a unified timestamp, data source, indicator type, and numerical format.

[0049] In some embodiments, the structured time series dataset of the current period is compared with the data of historical periods to calculate the rate of change and trend stability of each characteristic indicator. The longitudinal comparison analysis selects the statistical value of the same characteristic indicator of the current reporting period and one or more previous historical reporting periods for comparison, and the rate of change is calculated as the relative percentage change between the average value of the current period and the average value of the previous period.

[0050] It is understandable that calculating the rate of change and trend stability of various characteristic indicators involves specific mathematical operations. Trend stability reflects the degree of fluctuation of the characteristic indicator over multiple consecutive periods. The formula for calculating the rate of change of the characteristic indicator is expressed as:

[0051] in: This represents the rate of change of the m-th characteristic indicator; This represents the arithmetic mean of the m-th feature indicator within the current reporting period; This represents the arithmetic mean of the m-th feature indicator over the immediately preceding reporting period.

[0052] Optionally, trend stability can be assessed by calculating the ratio of the standard deviation to the mean of the feature indicators over N consecutive reporting periods. This ratio is defined as the coefficient of variation; a smaller coefficient of variation indicates higher trend stability. See Table 1 for a simplified example of longitudinal comparative analysis of some feature indicators in a structured time series dataset.

[0053] Table 1: Vertical Comparison Analysis of Feature Indicators Characteristic index name Historical period average (P) Current period average (C) Rate of change (ΔV) Trend stability (coefficient of variation for the last 3 periods) Joint range of motion variability (left wrist) 4.2 degrees 5.1 degrees +21.4% 0.18 Average temperature difference in local temperature abnormality area (right knee joint) 1.8°C 1.5°C -16.7% 0.12 Pressure asymmetry coefficient (ankle joint) 1.25 1.35 +8.0% 0.25 In practice, the early risk level assessment results for rheumatic diseases are updated based on the rate of change and trend stability. The evolution of the risk level is then presented in the risk assessment report as a visual trend chart. When updating the assessment results, the latest calculated characteristic indicator values, along with their rate of change and trend stability indicators, are input into the early risk assessment model for rheumatic diseases. The scoring and weighted decision-making process is then re-executed to generate a new risk level label. The visual trend chart is generated using a chart library, transforming the risk level assessment results at different time points into time-series line charts or regional charts. Each time point in the chart is labeled with the specific assessment result date and risk level, and the main contributing indicators at that time point can be viewed interactively.

[0054] Based on the updated risk level assessment results and the latest key pathological signs labels, combined with a pre-built clinical knowledge base, a draft clinical recommendation is automatically generated, including suggestions for further examinations, key points of lifestyle interventions, and follow-up appointment schedules. The pre-built clinical knowledge base exists in the form of a rule engine, storing a large number of mapping rules in the form of "IF (condition) THEN (recommendation)," where the condition part is associated with the risk level, key pathological signs labels, and the rate of change of specific characteristic indicators. In some embodiments, periodic risk assessment reports are generated in a structured document format, with fixed report chapters including data summaries, characteristic indicator trend analysis, updated risk level assessments, visualization charts, and draft clinical recommendations. The report is automatically pushed to the report center of authorized remote medical terminals via a secure application programming interface for doctors to access and review. Optionally, the generation of the draft clinical recommendation is logically divided into three parallel branches: one branch handles laboratory or imaging examination recommendations, one branch handles lifestyle intervention recommendations such as exercise and diet, and one branch handles follow-up appointment schedule recommendations. The outputs of the three branches are finally merged into a coherent draft text. In practice, the entire process of generating reports and constructing draft recommendations is encapsulated as an automated service. This service is scheduled to be executed at the end of the reporting cycle. The execution process involves multiple steps, including data querying, calculation, model reasoning, rule matching, and document synthesis. All steps are logged to ensure auditability.

[0055] See Figure 4 This is a normalized comparison bar chart of multiple characteristic indicators used for remote monitoring of rheumatic diseases. Joint activity variability generally showed an upward trend, peaking in cycle 4, indicating that joint function degeneration was most pronounced in the middle stage, with slight relief in the later stage. Abnormal temperature difference showed a continuous downward trend, indicating that local joint inflammation gradually decreased during the monitoring period. Pressure asymmetry coefficient showed a continuous upward trend, stabilizing after cycle 4, reflecting the persistent and stabilizing problem of uneven joint load. This chart clearly shows the separation trend between "inflammatory relief" and "joint function degeneration and structural imbalance," indicating that even if inflammation is controlled, biomechanical abnormalities of the joint may persist, requiring continuous monitoring and intervention. Cycle 4 is a turning point for multiple indicators and can serve as a key clinical focus point. Combined with the patient's subjective symptoms and risk level during the same period, subsequent dynamic monitoring plans can be optimized.

[0056] In one embodiment of the present invention, the method further includes receiving and integrating subjective symptom descriptions and records of daily activity limitations actively reported by patients in a preset health log interface. Natural language processing is performed on the subjective symptom description text to extract symptom keywords, severity modifiers, and frequency information, and it is converted into a standardized symptom description vector. The standardized symptom description vector is then fused with objective biomechanical and thermal imaging feature indicators collected through a wearable device to generate a rheumatic disease risk assessment correction result that integrates subjective and objective information. When the keywords in the standardized symptom description vector are highly consistent with key pathological sign labels in terms of anatomical location and symptom nature, the weight of the key pathological sign labels in the risk assessment is increased. The natural language processing of the subjective symptom description text to extract symptom keywords, severity modifiers, and frequency information, and its conversion into a standardized symptom description vector, includes using a pre-trained rheumatic disease medical dictionary to segment and recognize entities in the subjective symptom description text, identifying specific words describing joint, muscle, and skin symptoms as symptom keywords. Adjectives and adverbs modifying the symptom keywords are identified and quantified, mapped to preset severity levels to obtain quantified severity modifiers. The text is analyzed to extract descriptions of symptom onset time and duration, as well as frequency adverbs, which are then converted into standardized frequency values ​​to form occurrence frequency information. Symptom keywords, quantified degree modifiers, and standardized occurrence frequency information are then vectorized and combined to generate standardized symptom description vectors.

[0057] The implementation involves receiving and integrating subjective information proactively reported by patients and fusing it with objective monitoring data, thus expanding the information dimensions of risk assessment. Specifically, the implementation receives and integrates subjective symptom descriptions and records of limitations in daily activities proactively reported by patients through a pre-defined health log interface. This health log interface, provided as a mobile application or web form, includes structured input fields and unstructured text input boxes. Patient-submitted records and data collected by wearable devices are linked and stored in a unified database using a unique patient identifier and timestamp. Natural language processing (NLP) is applied to the subjective symptom descriptions, extracting symptom keywords, degree modifiers, and frequency information, and converting them into standardized symptom description vectors. The NLP workflow is optimized for medical texts in the rheumatology field, including text preprocessing, medical entity recognition, relation extraction, and numerical encoding steps.

[0058] In some embodiments, standardized symptom description vectors are fused with objective biomechanical and thermal imaging feature indicators collected through wearable devices to generate a rheumatic disease risk assessment correction result that integrates subjective and objective information. Multimodal data fusion is performed in a feature-level fusion module, which concatenates the numerical symptom description vector with objective feature indicator vectors from the same time period to form a new extended feature vector. This extended feature vector is then input into a retrained or adjusted risk assessment model. When keywords in the standardized symptom description vector are highly consistent with key pathological sign labels in terms of anatomical location and symptom nature, the weight of the key pathological sign labels in the risk assessment is increased. Consistency judgment is performed by calculating the semantic similarity between the anatomical entities extracted from the symptom description vector and the locations indicated by the key pathological sign labels, as well as the matching degree between the symptom nature description and the pathological nature of the labels. If both similarities exceed a preset threshold, the weight increase logic is triggered.

[0059] It is understandable that performing natural language processing on subjective symptom description text to generate standardized symptom description vectors involves a series of specific operational steps. A pre-trained rheumatology medical dictionary is used to segment and recognize entities in the subjective symptom description text, identifying specific words describing joint, muscle, and skin symptoms as symptom keywords. The rheumatology medical dictionary contains standard medical terms such as "joint swelling," "morning stiffness," and "burning sensation," along with their common synonyms and colloquial expressions. Entity recognition employs a method combining dictionary-based and conditional random field models. Adjectives and adverbs modifying symptom keywords are identified and quantified, mapping them to preset severity levels to obtain quantified severity modifiers. The preset severity levels are a numerical scale from 0 to 10; for example, the adjective "mild" maps to level 2, "moderate" to level 5, and "severe" to level 9, while the adverb "very" is weighted and multiplied according to the severity level.

[0060] Optionally, in natural language processing, descriptions of symptom occurrence time and duration, as well as frequency adverbs such as "frequently" and "occasionally," are parsed from the text and converted into standardized frequency values ​​to form occurrence frequency information. The conversion rules are based on a predefined mapping table; for example, "several times a day" is mapped to a value of 7 (times / day), "several times a week" is mapped to a value of 3.5 (times / week), and "occasionally" is mapped to a value of 0.5 (times / day). Symptom keywords, quantified degree modifiers, and standardized occurrence frequency information are vectorized and encoded to generate a standardized symptom description vector. The vectorization encoding assigns a unique dimension to each identified symptom keyword, and the value of this dimension is filled by the product of the degree modifier level and the occurrence frequency value. The dimension of symptom keywords that do not appear is filled with zero, thus forming a fixed-length and numerical vector representation.

[0061] In some embodiments, the calculation process for generating revised risk assessment results through multimodal data fusion can incorporate a weighted adjustment mechanism. The calculation formulas for fusion and weight adjustment are expressed as follows:

[0062] in: This indicates the revised risk score; This indicates the original risk score based solely on objective characteristic indicators; This represents a preset fusion adjustment coefficient; Represents a standardized symptom description vector Signature characteristics of objective key pathological signs Consistency and similarity between them; This represents the severity-weighted value indicated by the symptom description vector; This represents the basic weight.

[0063] In practice, the health log interface, in addition to text input, provides a visual body diagram for patients to select areas of pain or discomfort. The selected data is converted into structured site codes and integrated with the text description. The generation of symptom description vectors and the multimodal fusion process are fully automated. Upon receiving the log submitted by the patient, the system automatically triggers the natural language processing pipeline and fusion calculation process, updating the corrected results to the patient's risk assessment file. When the keywords and key pathological signs in the standardized symptom description vector are highly consistent in anatomical location and symptom nature, the system multiplies the contribution weight of the key pathological signs in the risk assessment model by a strengthening factor greater than 1. The specific value of this strengthening factor is linearly derived from the consistency similarity score. In practice, the entire subjective and objective information fusion processing module serves as an optional enhancement branch of the risk assessment process. Its activation and execution do not interfere with the original core assessment process based on objective data. The results of both processes are displayed in parallel or a final decision is made based on confidence level. The processing log records the data source, intermediate vectors, and weight changes in detail for auditing purposes.

[0064] See Figure 5 This is a bar chart comparing subjective and objective data fusion for early risk assessment of rheumatic diseases. It clearly shows the difference between the original risk score based solely on objective data from wearable devices and the revised risk score after incorporating patient subjective symptom reports. Both scores increased with monitoring time, indicating that the patient's risk of rheumatic diseases gradually increased over time. The revised risk score was consistently higher than the original risk score, and the gap between the two gradually increased over time, reaching its maximum at week 12 (approximately 1.2 points). This indicates that patient subjective symptom reports significantly improved the sensitivity of risk assessment and also suggests that objective data may underestimate the true pathological progression. This result validates the necessity of subjective and objective data fusion: relying solely on objective physiological data may overlook the patient's subjective discomfort, leading to a conservative risk assessment.

[0065] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for early diagnosis and treatment of rheumatic diseases based on remote monitoring, characterized in that: The method includes: Acquire multimodal physiological data collected by the patient's wearable device, including periodic joint range of motion sequences, morning stiffness duration records, body surface temperature distribution maps, and joint pressure distribution data; The multimodal physiological data are processed to extract rheumatic pathological features, generating biomechanical and thermal imaging feature indicators related to early rheumatic diseases. The biomechanical and thermal imaging feature indicators include joint range of motion variability, stiffness diffusion pattern, local temperature anomaly area, and pressure asymmetry coefficient. The biomechanical and thermal imaging characteristics are input into a pre-constructed early risk assessment model for rheumatic diseases to generate individualized early risk level assessment results for rheumatic diseases and key pathological signs labels. Based on the early risk level assessment results of rheumatic diseases and key pathological signs, a dynamic monitoring plan is generated, which includes monitoring items, monitoring frequency and early warning threshold adjustment schemes. Based on the execution feedback data of the aforementioned dynamic monitoring plan, periodic risk assessment reports and draft treatment recommendations are generated for remote medical terminals to access.

2. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function according to claim 1, characterized in that, The process of extracting rheumatic pathological features from the multimodal physiological data to generate biomechanical and thermal imaging feature indicators related to early rheumatic diseases includes: The periodic joint range of motion sequence is divided into multiple monitoring periods according to the time dimension, and the standard deviation of the maximum joint range of motion within each monitoring period is calculated to obtain the variability of the joint range of motion. The duration of morning stiffness is recorded and processed using pattern recognition. The differences in the onset time and the rate of regression of stiffness between symmetrical joints of the body are analyzed to generate the stiffness diffusion pattern that describes the migration law of stiffness. Identify continuous regions with a sustained temperature difference from healthy regions from the body surface temperature distribution map, extract the boundary morphology, temperature gradient, and diurnal temperature fluctuation value of the continuous regions, and construct the local temperature anomaly zone. Based on the joint pressure distribution data, the pressure peak ratio and pressure distribution center of gravity offset of the corresponding joints on the left and right sides of the body under the same movement mode are calculated to obtain the pressure asymmetry coefficient that reflects the uneven joint load. The pressure asymmetry coefficient is composed of the peak pressure ratio and the centroid offset of the pressure distribution, and is calculated in the following way: The peak pressure ratio of the target joints on the left and right sides of the body under the same standardized movement, extracted from the joint pressure distribution data, is nonlinearly weighted and fused with the pressure distribution center of gravity offset. The peak pressure ratio reflects the difference between the two joints when bearing the maximum load, and the pressure distribution center of gravity offset reflects the asymmetry between the two joints in the spatial position of the load distribution. After normalizing the peak pressure ratio and the centroid offset of the pressure distribution, the weighted sum of squares of the two is calculated and square rooted using a preset fusion function to obtain the pressure asymmetry coefficient. The numerical range of this coefficient is positively correlated with the degree of joint load imbalance.

3. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function according to claim 2, characterized in that, The process involves inputting the biomechanical and thermal imaging characteristics into a pre-constructed early risk assessment model for rheumatic diseases to generate individualized early risk level assessment results and key pathological sign labels for rheumatic diseases, including: The variability of the joint range of motion is compared with the baseline data of healthy people of the same age group in a standardized manner to generate a joint function degeneration rate score. The standardized comparison of the joint function degeneration rate score adopts the Z-score method. The joint function degeneration rate score is the product of the absolute value of the Z-score value and the preset sensitivity coefficient. The rigid diffusion pattern is matched with pattern templates in the typical rheumatic disease pathological process library to generate a rigid pattern conformity score. The rigid pattern conformity score is obtained by calculating the cosine similarity between the feature vector of the rigid diffusion pattern to be evaluated and the feature vector of each predefined template in the typical rheumatic disease pathological process library. The highest similarity value is taken as the base of the rigid pattern conformity score, and then weighted and corrected according to the pattern duration and the number of joints involved. The characteristic parameters of the local temperature anomaly area are input into the inflammation activity prediction sub-model to generate a potential local inflammation activity intensity prediction value. The inflammation activity prediction sub-model is a regression model trained based on the gradient boosting decision tree algorithm. The characteristic parameters include the area of ​​the local temperature anomaly area, the average temperature difference, the maximum temperature gradient, and the diurnal fluctuation variance. The model output is a prediction value in the range of 0 to 1. The pressure asymmetry coefficient is compared with the joint injury risk threshold to generate a joint structural imbalance risk score. The score is calculated using a piecewise linear function. When the pressure asymmetry coefficient is below the first threshold, the score is 0. When it is between the first threshold and the second threshold, the score increases linearly. When it is above the second threshold, the score is full. By combining the joint function degeneration rate score, stiffness pattern conformity score, potential local inflammatory activity intensity prediction value, and joint structural imbalance risk score, a weighted decision algorithm is used to generate a quantitative early risk level assessment result for rheumatic diseases. The abnormal indicator with the highest contribution is output as the key pathological sign label. The contribution is determined by calculating the percentage of each input score multiplied by its corresponding weight coefficient relative to the comprehensive risk score. The indicator with the highest percentage is marked as the key pathological sign label.

4. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function according to claim 1, characterized in that, The dynamic monitoring plan, which includes monitoring items, monitoring frequency, and early warning threshold adjustment schemes, is generated based on the early risk level assessment results of rheumatic diseases and key pathological signs. Based on the different intervals of the early risk level assessment results of rheumatic diseases, the combination of data collection items of the wearable device is dynamically adjusted. The high risk level corresponds to more comprehensive joint range of motion and pressure monitoring, while the medium risk level focuses on temperature and stiffness monitoring. Based on the specific body part indicated by the key pathological sign label, the monitoring frequency is set in a personalized manner. For parts where pathological sign labels have appeared, the data collection density is increased, while for parts where no signs have appeared, the conventional monitoring density is used. Based on the changing trends of the biomechanical and thermal imaging characteristic indicators in historical monitoring data, an individualized early warning threshold adjustment scheme is set for each characteristic indicator, and the early warning threshold adjustment scheme ensures a balance between early warning sensitivity and specificity.

5. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function according to claim 4, characterized in that, Based on the execution feedback data of the dynamic monitoring plan, periodic risk assessment reports and draft treatment recommendations are generated for remote medical terminals to access, including: All multimodal physiological data and calculated biomechanical and thermal imaging characteristic indicators generated during the execution of the dynamic monitoring plan are periodically summarized to form a structured time series dataset; The structured time series dataset of the current period is compared and analyzed longitudinally with the data of historical periods to calculate the rate of change and trend stability of each characteristic indicator. Based on the rate of change and trend stability, the early risk level assessment results of rheumatic diseases are updated, and the evolution of the risk level is presented in the form of a visual trend chart in the risk assessment report. Based on the updated risk level assessment results and the latest key pathological signs labels, combined with a pre-built diagnostic and treatment knowledge base, a draft of the diagnostic and treatment recommendations is automatically generated, which includes suggestions for further examinations, key points for lifestyle interventions, and a schedule for follow-up visits.

6. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function as described in claim 2, characterized in that, The calculation of the pressure peak ratio and pressure distribution center of gravity offset of corresponding joints on the left and right sides of the body under the same movement mode based on the joint pressure distribution data includes: Extract the pressure data sequences of the target joints on the left and right sides of the body under a preset standardized movement from the joint pressure distribution data; Calculate the maximum pressure value of the pressure data sequence of the target joint on the left side of the body and the maximum pressure value of the pressure data sequence of the target joint on the right side of the body, respectively. The pressure peak ratio is obtained by calculating the ratio of the maximum pressure value of the target joint on the left side of the body to the maximum pressure value of the target joint on the right side of the body. The center of gravity analysis was performed on the pressure distribution data of the target joint on the left side of the body during the complete cycle of a standardized movement, and the coordinates of the center of gravity of the pressure distribution on the left side of the body were calculated. The center of gravity analysis was performed on the pressure distribution data of the target joint on the right side of the body during the complete cycle of a standardized movement, and the coordinates of the center of gravity of the pressure distribution on the right side of the body were calculated. Calculate the Euclidean distance between the coordinates of the center of gravity of the pressure distribution on the left side of the body and the coordinates of the center of gravity of the pressure distribution on the right side of the body, and normalize the Euclidean distance relative to the joint reference size to obtain the offset of the center of gravity of the pressure distribution.

7. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function as described in claim 3, characterized in that, The quantitative assessment results of the early risk level of rheumatic diseases generated by the weighted decision algorithm include: A first weighting coefficient is assigned to the joint function degeneration rate score, a second weighting coefficient is assigned to the stiffness pattern conformity score, a third weighting coefficient is assigned to the predicted value of potential local inflammatory activity intensity, and a fourth weighting coefficient is assigned to the joint structural imbalance risk score. The joint function degeneration rate score is multiplied by the first weighting coefficient to obtain the first weighted score; The rigid pattern conformity score is multiplied by the second weighting coefficient to obtain the second weighted score; The predicted value of the potential local inflammatory activity intensity is multiplied by the third weighting coefficient to obtain the third weighted score; The joint structure imbalance risk score is multiplied by the fourth weighting coefficient to obtain the fourth weighted score; The first weighted score, the second weighted score, the third weighted score, and the fourth weighted score are summed to obtain a comprehensive risk score. The comprehensive risk score is mapped to a preset risk level division interval, which includes four levels: no risk, low risk, medium risk, and high risk. The corresponding level identifier is output as the quantitative assessment result of the early risk level of rheumatic disease.

8. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function according to claim 1, characterized in that, The method further includes: Receive and integrate subjective symptom descriptions and records of limitations in daily activities proactively reported by patients in the preset health log interface; Natural language processing is performed on the subjective symptom description text to extract symptom keywords, degree modifiers and occurrence frequency information, and then it is converted into a standardized symptom description vector. The standardized symptom description vector is fused with objective biomechanical and thermal imaging feature indicators collected through the wearable device to generate a rheumatic disease risk assessment correction result that integrates subjective and objective information. When the keywords in the standardized symptom description vector are highly consistent with the key pathological sign labels in terms of anatomical location and symptom nature, the weight of the key pathological sign labels in risk assessment is increased.

9. The method for early diagnosis and treatment of rheumatic diseases based on remote monitoring function according to claim 8, characterized in that, The process of performing natural language processing on the subjective symptom description text, extracting symptom keywords, degree modifiers, and frequency information, and converting them into standardized symptom description vectors includes: The subjective symptom description text was segmented and entity recognized using a pre-trained rheumatology medical dictionary, and specific words describing joint, muscle, and skin symptoms were identified as symptom keywords. Identify and quantify the adjectives and adverbs that modify the keywords of the symptoms, and map them to a preset severity level to obtain the quantified severity modifiers; The text is analyzed to extract descriptions of symptom onset time and duration, as well as frequency adverbs such as "frequently" and "occasionally," which are then converted into standardized frequency values ​​to form the occurrence frequency information. The symptom keywords, the degree modifiers, and the frequency of occurrence information are vectorized and encoded to generate the standardized symptom description vector.

10. A rheumatic disease early diagnosis and treatment system based on remote monitoring, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the early diagnosis and treatment method for rheumatic diseases based on remote monitoring function as described in any one of claims 1 to 9.

Images