Intelligent rehabilitation digital therapy system for cerebral palsy

Abstract

The invention provides an intelligent rehabilitation digital therapy system for cerebral palsy. The system comprises a data processing and analyzing module; the data processing and analysis module comprises a data acquisition and transmission architecture unit, a data cleaning and preprocessing unit, a feature extraction and multi-modal fusion unit, a model training and prediction mechanism unit and an algorithm type selection unit; wherein the data acquisition and transmission architecture unit is used for acquiring physiological and motion data of a user; the data cleaning and preprocessing unit is used for carrying out anti-interference, complementation and standardization processing on the collected data; the feature extraction and multi-modal fusion unit is used for establishing a corresponding relation between the physiological data and the motion data and evaluating the training completion degree and the training tolerance degree of the user based on the physiological data and the motion data; the model training and prediction mechanism unit is used for constructing a training model of the cerebral palsy patient based on historical data of different types of cerebral palsy patients; by means of the system, the family type or linkage type precise rehabilitation effect can be achieved.

Description

Technical Field

[0001] The present invention relates to medical rehabilitation auxiliary system equipment, in particular to an intelligent rehabilitation digital therapy system for cerebral palsy. Background Art

[0002] Cerebral palsy (CP) is a group of persistent central motor and postural developmental disorders and activity restriction syndromes caused by non-progressive damage to the developing fetus or infant brain. The movement disorders of CP are often accompanied by sensory, perceptual, cognitive, communication and behavioral disorders, as well as epilepsy and secondary muscle and skeletal problems. Cerebral palsy is a developmental disorder that affects the lifelong developmental trajectory of children and their family life. Therefore, intervention measures must be considered from the perspective of promoting functional development and supporting family rehabilitation services. From the perspective of rehabilitation needs, the heterogeneity of clinical phenotypes, pathogenesis and etiology of CP patients poses a huge challenge to effective therapeutic intervention.

[0003] From the perspective of rehabilitation treatment programs, the early stages of cerebral palsy rehabilitation treatment mainly focused on traditional physical therapy and simple functional training. As the understanding of the pathological mechanisms of cerebral palsy deepened, a multidisciplinary comprehensive treatment model gradually developed. For example, modern rehabilitation concepts emphasize neurorehabilitation, using the brain's neuroplasticity to improve patient function. In terms of treatment methods, it has expanded from simple exercise training to include a variety of methods such as drug therapy, surgical intervention, and rehabilitation engineering. For example, functional orthopedic surgery can correct limb deformities and improve motor function; drug therapy is used to relieve symptoms such as spasticity. At the same time, rehabilitation treatment has begun to focus on long-term care and personalized services to meet the needs of patients at different stages. For example, personalized rehabilitation plans are developed for children with cerebral palsy with different MRI classifications to improve rehabilitation effects.

[0004] Looking at the current state of rehabilitation, there has been a trend toward earlier onset of rehabilitation treatment in recent years, but access to rehabilitation services varies across regions and economic sectors. For example, in South Korea, an analysis of rehabilitation treatment for children with cerebral palsy from 2003 to 2013 found that the age of initiation has been earlier in recent years, but the types of treatment vary across age groups. In Bangladesh, only 50.2% of children with cerebral palsy receive rehabilitation services, and factors such as being female, illiterate mothers, and low family income are associated with a lack of access.

[0005] With the development of information technology, digital therapies are gradually gaining popularity in the treatment of cerebral palsy. For example, interactive home-based training via the internet offers a new approach to addressing the challenges of limited medical resources and insufficient training intensity. Research has shown that this training approach allows children to receive more intensive and sustained training, improving functional recovery outcomes. The combination of robot-assisted therapy and digital technology has brought new breakthroughs to the treatment of cerebral palsy. For example, Armeo robotic therapy offers significant advantages over traditional therapies in improving upper limb motor quality in children with hemiplegia and cerebral palsy. However, it also suffers from high costs, limited accessibility, high professional requirements, limited flexibility, and wide variations in adaptability among children. Furthermore, virtual environment rehabilitation utilizes digital technology to create interactive scenarios to increase children's motivation and engagement in rehabilitation. For example, virtual environments based on Leap Motion and EEG sensors offer a new option for neuropsychomotor rehabilitation in children with cerebral palsy. However, their technical stability and accuracy need to be improved, and their comfort and safety require further verification. Therefore, while virtual reality rehabilitation shows promise, and home-based virtual reality rehabilitation can improve upper limb and gross motor function in children, traditional cerebral palsy rehabilitation training programs have the following drawbacks:

[0006] 1. Lack of personalization in rehabilitation programs: Current cerebral palsy rehabilitation programs often rely on standardized training processes that fail to fully consider individual patient differences. Because the location and severity of brain damage, as well as physical function, vary from patient to patient, a standardized training program is unlikely to achieve optimal rehabilitation results. This results in some patients experiencing slow recovery progress and even missing the optimal time for recovery.

[0007] 2. Rehabilitation training lacks interest: Traditional rehabilitation training, which primarily involves repetitive physical exercises and physical therapy, can be tedious. Children, in particular, struggle to maintain focus and motivation for extended periods, leading to poor training compliance and an inability to maintain continuity and intensity, severely impacting rehabilitation outcomes.

[0008] 3. Uneven distribution of rehabilitation resources: High-quality cerebral palsy rehabilitation resources are concentrated in specialized medical institutions in major cities. Remote areas or grassroots communities lack specialized rehabilitation equipment and professional rehabilitation physicians. Patients and their families must travel to major cities, expending significant time and money, and struggle to obtain consistent and timely rehabilitation guidance, resulting in many patients being unable to receive effective rehabilitation treatment.

[0009] 4. Delayed monitoring and adjustment of the rehabilitation process: In the current rehabilitation process, doctors primarily monitor rehabilitation progress through regular face-to-face consultations and patient or family member descriptions. This information is not timely and comprehensive. Without real-time access to patient training data, it is difficult to adjust rehabilitation plans based on the patient's actual situation, resulting in reduced relevance and effectiveness of rehabilitation training.

[0010] 5. Heavy burden of rehabilitation on families: Cerebral palsy rehabilitation is a long process, requiring patients to frequently travel to hospitals for rehabilitation training, placing a heavy financial and time burden on families. Furthermore, families lack professional rehabilitation guidance, making it difficult for family members to properly assist patients with rehabilitation training, which not only affects the effectiveness of rehabilitation but also increases the emotional burden on the family.

[0011] 6. Limited application of new digital rehabilitation equipment: Currently, most digital rehabilitation equipment is expensive to purchase and maintain, and requires specialized technicians to operate. This makes it difficult for small and medium-sized rehabilitation institutions and remote areas to equip these devices, resulting in limited widespread adoption. Furthermore, the equipment has strict requirements for the operating environment and patient status. Some children resist training due to discomfort or technical errors, making it difficult for the equipment to fully function. This exacerbates the uneven distribution of rehabilitation resources and limits the full application of digital therapy in cerebral palsy rehabilitation. Summary of the Invention

[0012] The present invention aims to overcome the above-mentioned defects and, through multimodal data acquisition and analysis technology, accurately assess the individual conditions of patients, tailor-make exclusive rehabilitation training programs for them, and improve the rehabilitation effect.

[0013] The present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is characterized by comprising a data processing and analysis module;

[0014] The above-mentioned data processing and analysis module includes a data acquisition and transmission architecture unit, a data cleaning and preprocessing unit, a feature extraction and multimodal fusion unit, a model training and prediction mechanism unit, and an algorithm selection unit;

[0015] The data collection and transmission architecture unit collects the user's physiological and motion data;

[0016] The above-mentioned data cleaning and pre-processing unit performs anti-interference, completion and standardization processing on the collected data;

[0017] The feature extraction and multimodal fusion unit establishes a correspondence between physiological data and motion data, and evaluates the user's training completion and training tolerance based on the physiological data and motion data;

[0018] The above-mentioned model training and prediction mechanism unit constructs a training model for cerebral palsy patients based on historical data of different types of cerebral palsy patients;

[0019] The above algorithm selection unit,

[0020] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0021] The above-mentioned data acquisition and transmission architecture unit realizes the collection of the user's physiological data through an integrated wearable device; and realizes the collection of the user's motion data through a spatial motion capture camera.

[0022] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0023] The above data cleaning and preprocessing unit,

[0024] To address the electromagnetic interference in electromyographic signals, wavelet transform is used for frequency domain filtering;

[0025] For motion trajectory data, Kalman filtering is used to eliminate jitter errors;

[0026] Bidirectional linear interpolation is used to fill short-term missing data, and long-term missing data are marked as invalid segments; heterogeneous data collected by different devices are converted into a unified dimension through Z-Score standardization.

[0027] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0028] The feature extraction and multimodal fusion unit uses CNN to analyze the spatiotemporal characteristics of electromyographic signals to determine the muscle movement state; and uses RNN to process continuous action sequences to evaluate movement coordination.

[0029] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0030] The evaluation method of the training completion of the above-mentioned user is based on video data, intercepting the motion image of each frame / set frame, comparing it with the standard motion image, calculating the deviation value, and obtaining the completion evaluation result according to different deviation values;

[0031] The training tolerance evaluation method of the above-mentioned user is to determine whether the physiological detection data exceeds the limit of normal rehabilitation exercise based on the real-time or specific time physiological detection data, and obtain the training tolerance evaluation result according to the comparison result.

[0032] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0033] It also includes a rehabilitation plan generation module;

[0034] The above-mentioned rehabilitation program generation module includes a rehabilitation program generation unit;

[0035] The above-mentioned rehabilitation program generation unit uses the patient's basic data as a benchmark, matches the closest model in the training model, and obtains the optimal training combination.

[0036] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0037] The above-mentioned rehabilitation plan generation module also includes a rule engine and machine learning fusion decision-making unit;

[0038] The above-mentioned rule engine and machine learning fusion decision unit compare the optimal training combination with the preset rules. When the optimal training combination is not higher than the clinical rules, the optimal training combination is adopted, otherwise the optimal training combination is excluded.

[0039] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0040] The above-mentioned rule engine is integrated with the machine learning decision-making unit to evaluate the user's training progress based on training completion and training tolerance, determine whether the difficulty needs to be increased or reduced, and maintain or adjust the training combination based on the judgment results.

[0041] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0042] The above-mentioned rehabilitation program generation unit uses the hierarchical analysis method to construct a multi-dimensional evaluation system, quantify the weights of various indicators, and finally generate a comprehensive rehabilitation needs index. Based on the comprehensive rehabilitation needs index, the closest model in the training model is matched to obtain the optimal training combination.

[0043] Furthermore, the present invention provides a cerebral palsy intelligent rehabilitation digital therapy system, which is also characterized by:

[0044] The rehabilitation program generated by the rehabilitation program generating unit is based on a VR motion image game.

[0045] Function and effect of the present invention:

[0046] This system uses digital technologies such as virtual reality and gamification to integrate rehabilitation training into fun interactive scenarios, stimulate patients' enthusiasm for active participation in training, and ensure the continuity and effectiveness of training.

[0047] This system builds a remote rehabilitation platform to enable patients to receive expert guidance remotely, break geographical restrictions, and achieve optimal allocation and sharing of rehabilitation resources.

[0048] This system uses wearable devices and smart sensors to collect patient training data in real time, combines artificial intelligence algorithms to analyze rehabilitation effects, provides data support for doctors, and realizes dynamic optimization of rehabilitation plans.

[0049] This system can reduce the frequency of patients going to the hospital for rehabilitation to a certain extent, thereby lowering rehabilitation costs; at the same time, it can provide family rehabilitation guidance functions, allowing family members to better assist patients in rehabilitation training and reduce the burden on the family. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 , the system architecture diagram provided by this embodiment;

[0051] Figure 2 , a diagram showing the system usage process provided in this embodiment. DETAILED DESCRIPTION

[0052] The present invention is susceptible to various modifications and embodiments, and thus specific embodiments are illustrated and described in the accompanying drawings. However, this is not intended to limit the present invention to specific embodiments, but rather should be understood to encompass all modifications, equivalents, and even substitutes that fall within the spirit and technical scope of the present invention.

[0053] This cerebral palsy intelligent rehabilitation digital therapy system adopts a modular design, integrating existing mature technologies and emerging technologies to ensure clinical feasibility and practical application value. This cerebral palsy intelligent rehabilitation digital therapy system mainly consists of two parts: hardware equipment and software system.

[0054] 1. Hardware

[0055] 1. Multimodal data collection equipment: This includes wearable physiological data collection devices (such as wristbands with integrated heart rate, blood oxygen, and body temperature sensors, as well as myoelectric sensors that adhere to muscle groups), a spatial motion capture camera array (distributed around the training area to capture the patient's limb movement trajectory), and an intelligent voice interaction microphone and touchscreen all-in-one device. These devices are connected to the central processing unit via wireless communication technologies such as Bluetooth and Wi-Fi, enabling real-time data transmission.

[0056] 2. User Interaction Terminal: A high-performance VR headset, combined with force feedback gloves, foot pedals, and other peripherals, creates an immersive interactive environment. The VR headset, equipped with a high-resolution display, gyroscope, accelerometer, and other sensors, senses the user's head posture and movements in real time, presenting a virtual rehabilitation scenario for the patient. Force feedback gloves simulate the resistance and tactile sensation of grasping objects, enhancing the realism of training.

[0057] 3. Remote communication equipment: Video call terminals equipped with high-definition cameras, microphones, and speakers, as well as 5G / 4G network communication modules, ensure the smoothness and stability of remote communication between doctors and patients.

[0058] 4. Wearable monitoring equipment: In addition to the above-mentioned electromyographic sensors, it also includes joint angle sensors (worn at the joints of the limbs) and pressure sensor insoles (embedded in rehabilitation training shoes) to monitor the patient's motion parameters in real time during training.

[0059] 2. Software system, including data processing and analysis module and rehabilitation plan generation module;

[0060] The data processing and analysis module, deployed on cloud servers, integrates deep learning algorithm libraries (such as TensorFlow and PyTorch) with a big data analysis platform. This module provides data cleaning, feature extraction, model training, and prediction capabilities, enabling unified processing of heterogeneous, multi-source data.

[0061] Rehabilitation Plan Generation Module: Based on a rules engine and machine learning model, a personalized rehabilitation plan is automatically generated based on patient assessment results. The plan includes parameters such as training items (such as balance training and gait training), training intensity (such as range of motion and number of repetitions), and training duration.

[0062] Specifically: The data processing and analysis module adopts an intelligent processing flow for multi-source heterogeneous data, which is divided into a data acquisition and transmission architecture unit, a data cleaning and preprocessing unit, a feature extraction and multimodal fusion unit, and a model training and prediction mechanism unit.

[0063] The data acquisition and transmission architecture utilizes multiple data sources, specifically: physiological data (heart rate, electromyographic signals), motion data (limb trajectory, joint angles), and speech interaction data (voice commands, training feedback) collected by integrated wearable devices (electromyography sensors, joint angle sensors), spatial motion capture cameras, and voice interaction devices. This data is transmitted to a cloud server in real time via wireless technologies such as Bluetooth and Wi-Fi, using encrypted channels to ensure data security and transmission latency of ≤50ms to support real-time analysis.

[0064] The data cleaning and preprocessing unit uses wavelet transforms for frequency-domain filtering to address electromagnetic interference in EMG signals. A Kalman filter is used to eliminate jitter errors in motion trajectory data. Short-term missing data is filled using bidirectional linear interpolation, and long-term missing data is marked as invalid to minimize impact on analysis accuracy. Heterogeneous data collected by different devices (e.g., EMG signals in μV, joint angles in degrees) is converted to a unified dimension using Z-score standardization to facilitate subsequent fusion analysis.

[0065] The feature extraction and multimodal fusion unit extracts time-domain features from EMG signals, including metrics such as root mean square (RMS) and zero-crossing rate (ZCR) that reflect muscle activation intensity. It also extracts dynamic parameters such as velocity and acceleration from motion trajectories. Frequency-domain features analyze the power spectral density of EMG signals using fast Fourier transforms to identify frequency band changes associated with muscle fatigue. Spatiotemporal feature fusion utilizes convolutional neural networks to extract features from the spatiotemporal matrix of EMG signals. It also incorporates long-short-term memory (LSTM) networks to process temporal dependencies within motion sequences, enabling spatiotemporal correlation analysis of multimodal data. For example, a CNN-LSTM model is used to correlate hand EMG signals with grasping motion trajectories in VR scenarios, while an RNN is used to process continuous motion sequences to assess movement coordination. CNNs analyze the spatiotemporal features of EMG signals to determine muscle movement status. Afterwards, based on the video data, the motion image of each frame / set frame is captured and compared with the standard motion image, and the deviation value is calculated. The evaluation results of the user's training completion are obtained according to the different deviation values; based on the physiological detection data in real time or at a specific time, it is judged whether the physiological detection data exceeds the limit of normal rehabilitation exercise, and the evaluation results of the user's training tolerance are obtained according to the comparison results.

[0066] Model training and prediction mechanism unit, training data construction: Collect rehabilitation training data of patients of different age groups (2-18 years old), different physiological / medical indicators, and different cerebral palsy types (spastic quadriplegia, spastic diplegia, spastic hemiplegia, involuntary movement type, ataxia type, mixed type, etc.), annotate training effects (GMFM-88 score changes, Peabody Motor Assessment Scale), and construct a labeled dataset containing more than 100,000 samples.

[0067] In order to optimize the effect, an algorithm selection unit can also be included.

[0068] Transfer learning is used to optimize model training efficiency. Based on the pre-trained ResNet network, parameters are fine-tuned for cerebral palsy rehabilitation data. Because there may be slight differences between the training and actual cerebral palsy rehabilitation data, including data, task objectives, and optimized training plans, fine-tuning may be required. The fine-tuning strategy can be designed with a loss function that freezes layered parameters and adapts to training and tasks. Through parameter adjustment, the pre-trained ResNet model is transformed from a "general feature extractor" to a "cerebral palsy rehabilitation-specific model", which is specifically manifested in the following ways: ① Focusing on key features: Strengthening the ability to identify abnormal movements of cerebral palsy patients (such as gait disorders and abnormal muscle tone). ② Improving prediction accuracy: Ensure that the error rate of evaluating training adaptability and movement status meets clinical needs (≤8%). ③ Enhancing generalization ability: With limited rehabilitation data, reduce the model's dependence on specific samples and improve adaptability to different patients and different training scenarios.

[0069] A multi-classification model is constructed using a gradient boosting tree to predict a patient's adaptability to different training programs (core training, stability training, strength training, balance training, gait training, transfer training, etc.). Based on the trained model, the system infers the real-time training data and outputs the current training status (such as movement completion, movement accuracy, and muscle fatigue level), with a prediction error rate of ≤8%.

[0070] The rehabilitation plan generation module realizes intelligent mapping from assessment to personalized plan, which includes a rehabilitation plan generation unit, a rule engine and a machine learning fusion decision unit.

[0071] The rehabilitation plan generation unit uses the patient's basic data as a benchmark, matching the closest model within the training model to obtain the optimal training combination. It uses the analytic hierarchy process to construct a multidimensional evaluation system, quantifying the weights of various indicators and ultimately generating a comprehensive rehabilitation needs index. Based on this comprehensive rehabilitation needs index, it matches the closest model within the training model to obtain the optimal training combination.

[0072] The rule engine and machine learning fusion decision-making unit compare the optimal training combination with the preset rules. If the optimal training combination does not exceed the clinical rules, it is adopted; otherwise, it is excluded. The user's training progress is evaluated based on training completion and training tolerance, and whether the difficulty needs to be increased or decreased. Based on the judgment results, the training combination is maintained or adjusted.

[0073] Specifically, the rehabilitation program generation unit processes input data, including static parameters such as basic patient information such as age, gender, cerebral palsy classification (GMFCS classification), hand function classification (MCS), muscle strength, muscle tone, joint mobility, abnormal reflexes, balance function, cognitive level, results of conventional assessment scales, and cooperation; clinical examination and test information such as MRI imaging features (brain injury location, lesion, conclusion, etc.); and dynamic parameters, including real-time assessment results output by the data processing module (such as joint mobility and motor coordination scores) and historical training data (accumulated training time and improvement trend). Then, a hierarchical analysis method is used to construct a multidimensional evaluation system, quantifying the weights of various indicators (for example, motor function accounts for 40% and cognitive engagement accounts for 30%), ultimately generating a comprehensive rehabilitation needs index.

[0074] Program parameter generation and dynamic adjustment. Training project matching: Based on cluster analysis, patients are divided into different subtypes (such as "upper limb dysfunction type" and "gait abnormality type"), and matched with the corresponding training project library:

[0075] ① Upper limb dysfunction: VR grasping game (combined with force feedback gloves) and electromyography biofeedback training.

[0076] ② Abnormal gait: virtual scene walking training (monitoring plantar force through pressure insoles).

[0077] Intensity and duration calculation: Range of motion: Calculated based on the difference between measured range of motion (ROM) and normal range of motion (ROM) using the formula: Target range of motion = Measured range of motion + (Normal range of motion - Measured range of motion) × Training phase coefficient (0.3 for initial training, 0.6 for mid-training, 0.9 for late training). Repetitions: Based on fatigue prediction models (e.g., the MUAP frequency decay rate in electromyographic signals), a safety upper limit is set: Maximum repetitions = Baseline repetitions (20) × Fatigue coefficient (0.8-1.0, dynamically adjusted based on real-time electromyographic data). A "time-efficiency" optimization model is used to find the optimal duration through Bayesian optimization (e.g., 15-25 minutes per session yields the highest rate of improvement in GMFM scores).

[0078] The rule engine is integrated with the machine learning decision-making unit. The core logic of the rule engine is based on the basic rules preset in clinical guidelines, including that low-intensity resistance training (≤1kg resistance) is recommended for patients with spastic cerebral palsy to avoid excessive muscle activation; ataxia patients focus on balance training, and a single training session should not exceed 20 minutes to prevent fatigue; children with involuntary cerebral palsy perform low-frequency rhythmic exercises, such as buoyancy training simulated by hydrotherapy (1-2 limb swings per second), slow grasping games in VR scenes (target movement speed ≤0.5m / s), and continuous resistance (3-5N) provided by force feedback gloves to assist in controlling involuntary hand movements. The rule engine automatically matches the "tactile feedback + visual guidance" combined training mode.

[0079] To resolve rule conflicts: a priority mechanism is adopted (clinical rules > data statistical rules). For example, when the machine learning model recommends high-intensity training, but the clinical rules limit the upper limit of intensity, the rule conflict check is automatically triggered and the parameters are adjusted. In terms of dynamic optimization of reinforcement learning, a deep Q network is used to build a decision-making model. The state space includes patient evaluation indicators and training history data, the action space is the training program parameters (such as training items, intensity, and duration), and the reward function uses the improvement in GMFM score as the core indicator. By simulating more than 100,000 training processes, the optimal strategy π*(s)→a is learned. For example, when the patient's balance training achievement rate is greater than 80% for three consecutive times, the training difficulty is automatically increased (such as reducing the support surface).

[0080] Each user's usage data and structure are stored on this platform, enabling solution verification and iteration. A newly generated solution is first simulated 500 times in a simulation environment to verify safety (e.g., whether the range of motion exceeds the safe range of joint mobility) and effectiveness (predicting an improvement of GMFM score by ≥5 points). Feedback data from doctors and patients (e.g., solution completion rate, subjective fatigue) is collected, and model parameters are updated quarterly to ensure that the solution evolves in tandem with clinical practice.

[0081] The system also includes a user interaction module: a dedicated rehabilitation training app has been developed that runs on VR headsets and touch screen terminals. The app interface adopts a gamification design and includes functional modules such as task guidance, progress tracking, and achievement rewards to enhance patient training enthusiasm.

[0082] The system also includes a remote rehabilitation platform: an online service platform built on a B / S architecture. Doctors log in via a web browser and can view patient training data and video surveillance footage in real time, providing remote guidance and adjusting plans. The platform supports data visualization, presenting patient rehabilitation progress in charts and graphs.

[0083] The system also includes a family rehabilitation guidance module: a mobile app designed to provide family members with graphic and video tutorials, as well as an online consultation service portal. The tutorials cover basic rehabilitation knowledge, family training techniques, and troubleshooting common problems.

[0084] The core working principle of this system is:

[0085] The principle of multimodal data acquisition and analysis is to collect multi-dimensional data such as physiological, motor, and language data from patients through wearable devices and environmental sensors. Signal processing algorithms (such as filtering and noise reduction) are used to pre-process the raw data. Deep learning models (such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs)) are then used to extract data features. For example, CNNs can analyze the spatiotemporal characteristics of electromyographic signals to determine muscle movement status, while RNNs can be used to process continuous motion sequences to assess movement coordination.

[0086] How personalized rehabilitation plans are generated: A reinforcement learning algorithm dynamically optimizes the rehabilitation plan based on basic patient information (age, gender, diagnosis), assessment data, and rehabilitation goals. The algorithm simulates the rehabilitation effects of different training strategies, selects the optimal combination, and adjusts parameters in real time based on patient feedback.

[0087] Virtual reality interaction principles: Computer graphics technology is used to generate a three-dimensional virtual scene, combined with the positioning and tracking capabilities of VR devices, to achieve real-time interaction between the scene and the patient's movements. When a patient wears a VR headset for training, the headset's built-in sensors detect head rotation and body movements. The system updates the virtual scene based on the movement commands, while also providing tactile feedback through force feedback devices, enhancing immersion and training effectiveness.

[0088] Telerehabilitation Principle: It utilizes real-time audio and video transmission and data synchronization technology to enable remote interaction between doctors and patients. Patient training data is uploaded to the cloud via an encrypted channel. Doctors access the data through the telerehabilitation platform, conduct real-time analysis and guidance, and push the adjusted rehabilitation plan to the patient's terminal.

[0089] The specific usage of this system is as follows:

[0090] S1. Data Collection and Processing: The patient enters the training area wearing the wearable device. The multimodal data acquisition device begins operating, collecting data in real time and transmitting it to the data processing and analysis module. This module preprocesses the data, extracts features, and analyzes them, generating a patient rehabilitation status assessment report that is then passed to the rehabilitation plan generation module.

[0091] S2. Rehabilitation Training Execution Process: The Rehabilitation Plan Generation Module generates a personalized training plan based on the assessment report and pushes it to the User Interaction Module. Patients receive training tasks via a VR headset or touchscreen terminal and perform rehabilitation training in a virtual environment. During training, wearable monitoring devices collect training data in real time and feed it back to the Data Processing and Analysis Module for evaluating training effectiveness and adjusting subsequent plans.

[0092] S3. Telerehabilitation Interaction Process: During patient training, the remote communication device activates video monitoring and data transmission, synchronizing training footage and data to the telerehabilitation platform. Doctors use the platform to monitor the patient's progress, provide real-time guidance, and adjust the rehabilitation plan. Updates are instantly pushed to the patient's device.

[0093] S4. Home Rehabilitation Assistance Process: Family members access home rehabilitation guidance content through a mobile app and follow the tutorial to assist the patient in home training. During training, family members can upload training photos, videos, or ask questions through the app to receive professional guidance and advice.

[0094] The effects and functions of this embodiment are as follows:

[0095] 1. Precise and personalized rehabilitation: To address the problem of lack of personalization in existing rehabilitation programs, this invention customizes exclusive rehabilitation programs for patients through multimodal data collection and in-depth analysis, significantly improving the pertinence and effectiveness of rehabilitation training, and enabling patients to achieve better rehabilitation results.

[0096] 2. Improve compliance with rehabilitation training: To address the shortcomings of traditional rehabilitation training being boring, this invention uses virtual reality and gamification technology to combine rehabilitation training with interesting scenes, stimulate patients' enthusiasm to actively participate in training, ensure the continuity and intensity of training, and thus improve the rehabilitation effect.

[0097] 3. Optimize the allocation of rehabilitation resources: The remote rehabilitation platform of the present invention breaks the geographical restrictions, allowing patients in remote areas to obtain high-quality rehabilitation resources, solving the problem of uneven distribution of rehabilitation resources, enabling more patients to benefit from professional rehabilitation services, and improving the utilization rate of rehabilitation resources.

[0098] 4. Realize dynamic rehabilitation management: With the help of wearable devices and smart sensors to collect data in real time, combined with artificial intelligence algorithm analysis, doctors can timely and comprehensively grasp the patient's rehabilitation progress, realize dynamic optimization and adjustment of rehabilitation plans, and improve the scientificity and effectiveness of rehabilitation training.

[0099] 5. Reduce the burden of family rehabilitation: reduce the frequency of patients going to the hospital and reduce rehabilitation costs; at the same time, provide professional rehabilitation guidance to families so that family members can better assist patients in training, which not only reduces the family's financial and time pressure, but also improves the effect of family rehabilitation.

[0100] While the above description focuses on the embodiments, this is merely illustrative and does not limit the present invention. Persons skilled in the art will readily appreciate that various modifications and applications not illustrated above are possible without departing from the essential characteristics of the embodiments. For example, the various components specifically illustrated in the embodiments may be implemented with modifications. Furthermore, any differences associated with such modifications and applications should be construed as being within the scope of the present invention as defined in the appended claims.

Claims

1. A cerebral palsy intelligent rehabilitation digital therapy system, characterized by: It includes a data processing and analysis module; the data processing and analysis module includes a data acquisition and transmission architecture unit, a data cleaning and preprocessing unit, a feature extraction and multimodal fusion unit, and a model training and prediction mechanism unit; The data collection and transmission architecture unit collects the user's physiological and motion data; The data cleaning and preprocessing unit performs anti-interference, complementation and standardization processing on the collected data; the feature extraction and multimodal fusion unit establishes a correspondence between physiological data and motion data, and evaluates the user's training completion and training tolerance based on the physiological data and motion data; The model training and prediction mechanism unit constructs a training model for cerebral palsy patients based on historical data of different types of cerebral palsy patients.

2. The cerebral palsy intelligent rehabilitation digital therapy system according to claim 1, characterized in that: The data acquisition and transmission architecture unit collects the user's physiological data through an integrated wearable device and collects the user's motion data through a spatial motion capture camera.

3. The intelligent digital rehabilitation therapy system for cerebral palsy according to claim 1, characterized in that: The data cleaning and preprocessing unit, To address the electromagnetic interference in electromyographic signals, wavelet transform is used for frequency domain filtering; For motion trajectory data, Kalman filtering is used to eliminate jitter errors; Bidirectional linear interpolation is used to fill short-term missing data, and long-term missing data are marked as invalid segments; heterogeneous data collected by different devices are converted into a unified dimension through Z-Score standardization.

4. The intelligent digital rehabilitation therapy system for cerebral palsy according to claim 1, characterized in that: The feature extraction and multimodal fusion unit analyzes the spatiotemporal characteristics of electromyographic signals through CNN to determine the muscle movement state; and uses RNN to process continuous action sequences to evaluate movement coordination.

5. The cerebral palsy intelligent rehabilitation digital therapy system according to claim 1, characterized in that: The evaluation method of the user's training completion is based on video data, intercepting the motion image of each frame / set frame, comparing it with the standard motion image, calculating the deviation value, and obtaining the completion evaluation result according to different deviation values; The training tolerance evaluation method of the user is to determine whether the physiological detection data exceeds the limit of normal rehabilitation exercise based on the real-time or specific time physiological detection data, and obtain the evaluation result of the training tolerance according to the comparison result.

6. The intelligent digital rehabilitation therapy system for cerebral palsy according to claim 1, characterized in that: It also includes a rehabilitation plan generation module; The rehabilitation program generation module includes a rehabilitation program generation unit; The rehabilitation program generating unit uses the basic data of the patient as a benchmark, matches the closest model in the training model, and obtains the optimal training combination.

7. The intelligent digital rehabilitation therapy system for cerebral palsy according to claim 1, characterized in that: The rehabilitation program generation module further includes a rule engine and a machine learning fusion decision unit; The rule engine and the machine learning fusion decision unit compare the optimal training combination with the preset rules. When the optimal training combination is not higher than the clinical rules, the optimal training combination is adopted; otherwise, the optimal training combination is excluded.

8. The intelligent digital rehabilitation therapy system for cerebral palsy according to claim 7, characterized in that: The rule engine is integrated with the machine learning decision-making unit to evaluate the user's training progress based on training completion and training tolerance, determine whether the difficulty needs to be increased or decreased, and maintain or adjust the training combination based on the judgment results.

9. The intelligent digital rehabilitation therapy system for cerebral palsy according to claim 1, characterized in that: The rehabilitation program generation unit uses the hierarchical analysis method to construct a multi-dimensional evaluation system, quantifies the weights of various indicators, and finally generates a comprehensive rehabilitation demand index. Based on the comprehensive rehabilitation demand index, the closest model in the training model is matched to obtain the optimal training combination.

10. The intelligent digital rehabilitation therapy system for cerebral palsy according to claim 1, characterized in that: The rehabilitation program generated by the rehabilitation program generating unit is based on a VR motion image game.

Images