ADHD evaluation and training system based on VR tasks

Abstract

The present application relates to the technical field of ADHD evaluation training, in particular to an ADHD evaluation and training system based on a VR task, which comprises a VR task generation unit, a multi-modal data acquisition unit, a dynamic coupling analysis unit and an adaptive modulation unit, in the present application, the VR task generation unit generates programmable virtual task scenes based on an attention load dynamic adjustment algorithm, the multi-modal data acquisition unit acquires eye movement and body movement data, the dynamic coupling analysis unit constructs a three-dimensional attention deviation index and correlates ADHD symptoms, and the adaptive modulation unit adjusts scene parameters, generates a personalized training strategy and outputs a three-dimensional evaluation atlas, solving the problem that traditional ADHD evaluation and training methods are difficult to accurately evaluate, through multi-modal data acquisition and analysis, dynamic task generation and personalized training strategy formulation, precise evaluation and personalized training of ADHD patients are realized, and evaluation accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of ADHD assessment and training technology, and more specifically, to an ADHD assessment and training system based on VR tasks. Background Technology

[0002] ADHD assessment and training is an important technique. Attention deficit hyperactivity disorder (ADHD) is a common neurodevelopmental disorder. Accurate assessment and effective training are crucial for the rehabilitation of patients. However, traditional assessment and training methods face many challenges in practical application and are difficult to meet clinical needs.

[0003] In terms of assessment, traditional methods mainly rely on doctors' subjective observation and scale evaluation. This makes it difficult to accurately determine the patient's core symptoms such as attention deficit, motor inhibition deficit, and decision delay, thus affecting the accuracy of diagnosis. For patients with different levels of attention deficit, using the same training intensity and content cannot effectively improve their attention levels. With the development of virtual reality (VR) technology, its application to ADHD assessment and training has become a new research direction. However, existing VR-based systems lack dynamic attention load adjustment algorithms, which makes it impossible for virtual task scenarios to be dynamically adjusted according to the patient's real-time performance. This makes it difficult to adapt to different patients' ability levels and training stages. It cannot change the visual interference density, target appearance time entropy value, and response time window dynamic compression rate in real time according to the patient's reaction, resulting in an inability to effectively stimulate the patient's attention. Consequently, it is difficult to accurately assess the patient's attention changes. It is also impossible to collect eye movement trajectory and limb movement data through eye-tracking modules and nine-axis inertial measurement modules respectively, thus making it impossible to accurately calculate the gaze point drift variance and limb micro-motion energy integral value. Therefore, it cannot comprehensively and accurately reflect the patient's behavioral characteristics, making it difficult to achieve accurate ADHD assessment. To solve this technical problem, we provide an ADHD assessment and training system based on VR tasks. Summary of the Invention

[0004] The purpose of this invention is to provide an ADHD assessment and training system based on VR tasks to solve the problems mentioned in the background art.

[0005] 1. Because traditional assessment methods rely on doctors' subjective observation and scale testing, it is difficult to accurately determine the core symptoms of ADHD patients, which affects the accuracy of diagnosis. Therefore, this case uses a multimodal data acquisition unit to collect eye movement and limb movement data, and a dynamic coupling analysis unit to construct a three-dimensional attention shift index and associate it with ADHD symptoms, which can comprehensively and accurately assess the patient's symptoms and improve the accuracy of diagnosis.

[0006] 2. Because existing VR-based systems lack dynamic attention load adjustment algorithms, virtual task scenarios cannot be adjusted according to the patient's real-time performance, making it difficult to stimulate the patient's attention and achieve accurate assessment. Therefore, this case uses a VR task generation unit to generate dynamically changing virtual task scenarios based on a dynamic attention load adjustment algorithm. This allows for the adjustment of scenario parameters according to the patient's response, effectively stimulating the patient's attention and achieving accurate assessment.

[0007] To achieve the above objectives, a VR task-based ADHD assessment and training system is provided, comprising the following units:

[0008] The VR task generation unit generates programmable virtual task scenes based on an attention load dynamic adjustment algorithm. Scene parameters include visual interference density, target appearance time entropy value, and response time window dynamic compression rate.

[0009] The multimodal data acquisition unit acquires the gaze point coordinate sequence of the user's eye movement trajectory and the frequency domain energy spectrum of limb movement through the eye tracking module and the nine-axis inertial measurement module, respectively. The eye tracking data is used to calculate the gaze point drift variance, and the nine-axis inertial measurement data is used to extract the limb micro-motion energy integral value.

[0010] The dynamic coupling analysis unit receives scene parameters, gaze point drift variance, and frequency domain energy spectrum of limb movements, and constructs a three-dimensional attention shift index. Then, it associates the attentional distraction, motor inhibition deficit, and decision delay of ADHD symptoms through nonlinear mapping.

[0011] The adaptive modulation unit adjusts scene parameters based on the three-dimensional attention shift index and simultaneously generates a personalized training strategy based on the covariance matrix of the gaze point drift variance and the frequency domain energy spectrum of limb movement. Finally, it integrates the three-dimensional attention shift index curve, scene parameter adjustment records, and user behavior improvement rate to output a three-dimensional evaluation map.

[0012] As a further improvement to this technical solution, the steps of the VR task generation unit in generating programmable virtual task scenes are as follows:

[0013] Initialize the basic parameters of the attention load dynamic adjustment algorithm, and set the initial range of visual interference density, target appearance time entropy value and response time window dynamic compression rate;

[0014] Based on the preset task type, basic scene elements are arranged in the virtual scene to build an initial task framework. According to the attention load dynamic adjustment algorithm, user feedback data is monitored in real time during task execution. The visual interference density, target appearance time entropy value and response time window dynamic compression rate are dynamically adjusted according to preset rules to generate a dynamically changing virtual task scene.

[0015] As a further improvement to this technical solution, the steps for the multimodal data acquisition unit to acquire data are as follows:

[0016] The eye-tracking module is calibrated, and during task execution, the gaze coordinate sequence of the user's eye movement trajectory is collected at a preset frequency, and the collected data is denoised.

[0017] The nine-axis inertial measurement module is initialized and calibrated. When the user performs VR tasks, limb motion data is continuously collected. The frequency domain energy spectrum of limb motion is extracted by frequency domain analysis method. At the same time, the gaze point drift variance and the limb micro-motion energy integral value are calculated.

[0018] As a further improvement to this technical solution, the steps for constructing the three-dimensional attention shift index by the dynamic coupling analysis unit are as follows:

[0019] The received scene parameters, gaze point drift variance, and frequency domain energy spectrum of limb movement are normalized to unify the data scale. Based on the normalized data, sub-indices are constructed from three dimensions: visual interference response, eye movement stability, and limb movement control. The sub-indices are then weighted and combined to form a three-dimensional attention shift index.

[0020] As a further improvement to this technical solution, the dynamic coupling analysis unit associates ADHD symptoms through a nonlinear mapping relationship as follows:

[0021] An ADHD symptom database was established, containing typical data samples of different degrees of attention deficit, motor inhibition deficit, and decision delay. A machine learning method was used, with the three-dimensional attention shift index as input and ADHD symptom data as output, to train a nonlinear mapping model.

[0022] The real-time calculated three-dimensional attention shift index is input into a trained nonlinear mapping model to obtain correlation results with ADHD symptoms such as attention distractibility, motor inhibition deficit, and decision delay.

[0023] As a further improvement to this technical solution, the dynamic coupling analysis unit also performs the following steps when training the nonlinear mapping model:

[0024] The data in the ADHD symptom database is divided into training, validation, and test sets according to a certain ratio. During the training process, the validation set is used to periodically evaluate the model and adjust the model hyperparameters. The test set is used to perform final performance validation on the trained model.

[0025] As a further improvement to this technical solution, the steps for the adaptive modulation unit to adjust the scene parameters are as follows:

[0026] Set different threshold ranges for the 3D attention shift index, with each range corresponding to a different scene parameter adjustment strategy. Monitor the 3D attention shift index in real time and determine the threshold range in which it falls.

[0027] Based on the threshold range, and according to the preset adjustment strategy, the visual interference density, target appearance temporal entropy value, and response time window dynamic compression rate in the VR task generation unit are adjusted accordingly.

[0028] As a further improvement to this technical solution, the steps for the adaptive modulation unit to generate a personalized training strategy are as follows:

[0029] Calculate the covariance matrix of gaze drift variance and the frequency domain energy spectrum of limb movement, analyze the correlation between the two, and formulate targeted training objectives based on the correlation characteristics reflected by the covariance matrix, combined with the user's individual characteristics and historical training data.

[0030] Based on the training objectives and the current task scenario, we design personalized training strategies that include adjusting task difficulty, planning training duration, and feedback methods.

[0031] As a further improvement to this technical solution, the adaptive modulation unit also performs the following steps when designing personalized training strategies:

[0032] A reinforcement learning mechanism is introduced to reward or punish users based on their real-time performance during training. Based on the feedback from rewards or punishments, the task difficulty, training duration, and feedback methods in the personalized training strategy are dynamically adjusted, and the training strategy is continuously optimized until the maximum number of optimizations is reached.

[0033] As a further improvement to this technical solution, the steps for the adaptive modulation unit to output the three-dimensional evaluation spectrum are as follows:

[0034] Collect the 3D attention shift index curve, record its changing trend during the training process, organize the scene parameter adjustment records, clarify the time, content and basis of each adjustment, calculate the user's behavior improvement rate during the training process, evaluate the training effect, and visualize the 3D attention shift index curve, scene parameter adjustment records and user behavior improvement rate to generate a 3D evaluation map.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] In a VR-based ADHD assessment and training system, a multimodal data acquisition unit collects eye-tracking and limb movement data. This data, combined with a dynamic coupling analysis unit, constructs a three-dimensional attention shift index and correlates it with ADHD symptoms. This overcomes the limitations of traditional assessment methods, enabling precise quantitative assessment of core symptoms such as attention deficit, motor inhibition deficits, and decision-making delays. This improves diagnostic accuracy and provides a reliable basis for subsequent treatment. The VR task generation unit generates dynamically changing virtual task scenarios based on an attention load dynamic adjustment algorithm. An adaptive modulation unit generates personalized training strategies based on patient data and incorporates reinforcement learning mechanisms to optimize these strategies. This allows training to closely align with individual patient differences, effectively stimulating patient participation and attention, enhancing training effectiveness, and enabling patients to gradually improve their symptoms under appropriate training intensity and content. This promotes the development of precision in ADHD patient assessment and training. Attached Figure Description

[0037] Figure 1 This is an overall block diagram of the present invention.

[0038] The meanings of the labels in the diagram are as follows:

[0039] 1. VR task generation unit; 2. Multimodal data acquisition unit; 3. Dynamic coupling analysis unit; 4. Adaptive modulation unit. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] This invention provides an ADHD assessment and training system based on VR tasks. Please refer to [link / reference]. Figure 1 As shown, it includes the following units:

[0042] VR task generation unit 1 generates programmable virtual task scenes based on a dynamic attention load adjustment algorithm. Scene parameters include visual interference density, target appearance time entropy value, and response time window dynamic compression rate.

[0043] The steps for VR task generation unit 1 to generate a programmable virtual task scene are as follows:

[0044] The basic parameters of the attention load dynamic adjustment algorithm are initialized, and the initial ranges of visual interference density, target appearance time sequence entropy value and response time window dynamic compression rate are set. This provides clear starting conditions for subsequent task scene generation and adjustment, making the operation of the entire system predictable and controllable. Among them, the target appearance time sequence entropy value is used to quantify the randomness of the target appearance time sequence, and the frequency domain energy spectrum is the energy distribution of motion data converted into the frequency domain through Fourier transform.

[0045] Based on the preset task type, basic scene elements are arranged in the virtual scene to construct an initial task framework, so that the virtual task scene is closely matched with the task type, which improves the effectiveness and relevance of the task. It can more accurately assess or train the user's attention ability. According to the attention load dynamic adjustment algorithm, during the task execution, user feedback data is monitored in real time, and the visual interference density, target appearance time entropy value and response time window dynamic compression rate are dynamically adjusted according to preset rules to generate a dynamically changing virtual task scene. This can more accurately assess the user's attention level. At the same time, by gradually increasing or decreasing the task difficulty, the user's attention ability is effectively trained to achieve better training or evaluation results.

[0046] The multimodal data acquisition unit 2 acquires the gaze point coordinate sequence of the user's eye movement trajectory and the frequency domain energy spectrum of limb movement through the eye tracking module and the nine-axis inertial measurement module, respectively. The eye tracking data is used to calculate the gaze point drift variance, and the nine-axis inertial measurement data is used to extract the limb micro-motion energy integral value.

[0047] The steps for data acquisition by the multimodal data acquisition unit 2 are as follows:

[0048] Before the user starts the VR task, the eye-tracking module's calibration program is initiated to calibrate it. Multiple calibration points are displayed, and the user is required to gaze at each point in sequence. The system records the deviation data between the user's actual gaze point and the calibration point, and corrects the output of the hardware sensor through interpolation or extrapolation algorithms to complete the calibration process. During task execution, the eye-tracking module continuously collects the gaze point coordinate sequence of the user's eye movement trajectory at a preset frequency of 30-120 times per second. The collected coordinate sequence is processed using a median filtering algorithm. By setting the filter window size, abnormal data points caused by blinking and nystagmus are removed, the eye movement trajectory curve is smoothed, and the quality of eye movement data is improved, providing effective data support for assessing the stability of user attention through gaze point drift variance.

[0049] The nine-axis inertial measurement module (INS) is initialized and calibrated. After the user wears the INS, it is placed in a static, horizontal state for initialization, resetting the internal parameters of the sensors. The accelerometer, gyroscope, and magnetometer are calibrated using calibration tools or standard posture samples. While the user performs VR tasks, limb motion data is continuously collected. The frequency domain energy spectrum of the limb motion is extracted using frequency domain analysis methods. Simultaneously, the gaze point drift variance and the integral value of limb micro-motion energy are calculated. The INS continuously collects limb motion data at a frequency of 50-200Hz. Fourier transforms are performed on the collected acceleration and angular velocity data to convert the time-domain data into frequency-domain data. The energy distribution of different frequency bands is calculated to obtain the frequency domain energy spectrum of the limb motion. Simultaneously, the gaze point drift variance is calculated based on the gaze point coordinate sequence. By integrating the amplitude and duration of the limb motion data, the integral value of limb micro-motion energy is calculated. This provides rich and accurate limb motion data for the dynamic coupling analysis unit 3, helping to more comprehensively and deeply assess the user's attention and motor control abilities, and improving the effectiveness of the ADHD assessment and training system.

[0050] The dynamic coupling analysis unit 3 receives scene parameters, gaze point drift variance, and frequency domain energy spectrum of limb movement, and constructs a three-dimensional attention shift index. Then, it associates the attentional distraction, motor inhibition deficit, and decision delay of ADHD symptoms through nonlinear mapping.

[0051] The steps for constructing the three-dimensional attention shift index in dynamic coupling analysis unit 3 are as follows:

[0052] The received scene parameters, gaze point drift variance, and frequency domain energy spectrum of limb movements are normalized to eliminate the influence of data scale and dimensions, ensuring that each data feature has equal importance in subsequent calculations. This avoids result bias caused by differences in data range and unifies the data scale. Based on the normalized data, sub-indices are constructed from three dimensions: visual interference response, eye-tracking stability, and limb movement control. These sub-indices provide a more accurate assessment of user attention and help identify attention problems in different aspects. The weights of each sub-indicator are determined based on historical experimental data, and the sub-indices are weighted and combined to form a three-dimensional attention shift index. This facilitates a comprehensive assessment of the user's attention state and provides a clear reference for the adaptive modulation unit 4 to adjust scene parameters and formulate personalized training strategies.

[0053] The steps of the dynamic coupling analysis unit 3 in associating ADHD symptoms through nonlinear mapping are as follows:

[0054] An ADHD symptom database was established, containing typical data samples of varying degrees of attention deficit, motor inhibition deficit, and decision delay. This improved the accuracy and generalization ability of the subsequently trained nonlinear mapping model, enabling the model to more accurately correlate the 3D attention shift index with ADHD symptoms. Using machine learning methods, the nonlinear mapping model was trained with the 3D attention shift index as input and ADHD symptom data as output. Backpropagation and an optimizer were used to adjust the model's weights and biases, minimizing the error between the model's predicted output and the actual labels. Cross-validation was employed during training to prevent overfitting, resulting in a model that accurately correlates the input 3D attention shift index with the output ADHD symptoms, providing an effective tool for subsequent real-time correlation analysis.

[0055] The real-time calculated 3D attention shift index is input into the trained nonlinear mapping model. The model calculates and predicts based on the input index to obtain specific values ​​of attention distractibility, motor inhibition deficit, and decision delay related to ADHD symptoms. The correlation results are fed back to the adaptive modulation unit 4 to adjust scene parameters and formulate personalized training strategies. It also provides a reference for professionals to assess users' ADHD symptoms, improving the real-time performance and effectiveness of the system. This allows the system to dynamically adjust training and assessment strategies according to the user's actual situation, better meeting user needs.

[0056] When training the nonlinear mapping model, the dynamic coupling analysis unit 3 also performs the following steps:

[0057] A stratified sampling method was employed, dividing the data in the database proportionally according to the severity of ADHD symptoms and the category of healthy controls. The ADHD symptom database was further divided into training, validation, and test sets to ensure that the proportion of each category in each set remained consistent with the original database. This prevented an imbalance in the proportion of any one category, which could negatively impact model training and evaluation. An independent data storage structure was established for each set, recording data indexes and labels for easy future use. During training, the validation set was used to periodically evaluate the model. After each training cycle, validation set data was input into the model, and the model's evaluation metrics on the validation set were calculated. The trends of these metrics across different training cycles were compared. If the metrics continuously decreased on the training set but began to rise on the validation set, overfitting was suspected. In this case, the hyperparameters were adjusted. Within the preset hyperparameter value range, different combinations of hyperparameters were tried, and the model was retrained after each adjustment. The model is trained and evaluated on a validation set. The optimal combination of hyperparameters is selected to optimize the model during training, improving its ability to learn data features and ultimately achieving better prediction and association results. A test set is used to validate the trained model's performance. After training on the training set and adjusting the hyperparameters on the validation set, the test set data is fully input into the trained model. Multiple evaluation metrics are calculated on the test set: mean squared error is used to evaluate the accuracy of ADHD symptom prediction, and classification accuracy is used to evaluate the correctness of symptom severity level judgment. Based on the test set evaluation results, the model's performance is comprehensively analyzed to determine if it meets the application requirements of a VR-based ADHD assessment and training system. If performance is substandard, the data partitioning, model structure, or training process is re-examined and improved to ensure the deployed model has good performance, accurately associating the 3D attention shift index with ADHD symptoms, thus improving the overall assessment and training effect of the system.

[0058] The adaptive modulation unit 4 adjusts the scene parameters according to the three-dimensional attention shift index, and at the same time generates a personalized training strategy based on the covariance matrix of the gaze point drift variance and the frequency domain energy spectrum of limb movement. Finally, it integrates the three-dimensional attention shift index curve, scene parameter adjustment records and user behavior improvement rate to output a three-dimensional evaluation map.

[0059] The steps for the adaptive modulation unit 4 to adjust the scene parameters are as follows:

[0060] Based on historical data, different thresholds for the 3D attention shift index are determined, and different threshold ranges for the 3D attention shift index are set. Each range corresponds to a different scene parameter adjustment strategy. For each threshold range, a corresponding scene parameter adjustment strategy is formulated. In the low shift range, it indicates that the user's attention is relatively concentrated, so the visual interference density can be increased, the target object appearance time entropy value can be increased, and the dynamic compression rate of the response time window can be compressed. In the medium shift range, the current scene parameters are maintained. In the high shift range, it indicates that the user's attention is severely scattered, so the visual interference density can be reduced, the target object appearance time entropy value can be decreased, and the dynamic compression rate of the response time window can be expanded to improve the targeting and effectiveness of training and evaluation. The 3D attention shift index is obtained in real time from the dynamic coupling analysis unit 3. The system uses real-time monitoring of the 3D attention shift index to determine its threshold range. Based on this range, it employs appropriate adjustment strategies to modify the visual interference density, target appearance temporal entropy, and response time window dynamic compression rate in VR task generation unit 1. If the target is in a low-shift range, the visual interference density can be increased by a certain percentage, the target appearance temporal entropy can be increased by a fixed value, and the response time window dynamic compression rate can be reduced to 90% of its original value. The adjusted scene parameters are then passed to VR task generation unit 1, which regenerates or updates the virtual task scene based on the new parameters. This allows users to train or evaluate in a scene that suits their attention level, improving training effectiveness and evaluation accuracy.

[0061] The steps of the adaptive modulation unit 4 in generating the personalized training strategy are as follows:

[0062] The covariance matrix of gaze drift variance and limb movement frequency domain energy spectrum is calculated, and the correlation between the two is analyzed. The user's gaze drift variance and limb movement frequency domain energy spectrum data are obtained from the multimodal data acquisition unit 2 to ensure that the time range of the data is consistent to ensure the accuracy of the analysis. Based on the correlation characteristics of the covariance matrix, the training strategy can be more accurately adapted to the user's actual situation and improve the effectiveness of training. Combined with the user's individual characteristics and historical training data, targeted training goals are formulated. The user's historical training data, including previous training results and attention improvement, can be viewed. Analyze user performance at different training stages to identify strengths and weaknesses. Based on the correlation analysis results of the covariance matrix, individual characteristics, and historical training data, formulate targeted training goals. For example, if the covariance shows a high positive correlation between eye movement and limb movement and the user's attention is improving slowly, the training goal can be set to reduce the correlation between the two within a certain period of time, improve the stability of attention, make training more directional, and help the user achieve the expected training effect more quickly. Adjust the difficulty of the task according to the training goal and the user's current ability level so that the user can train in a more suitable training environment, improve the efficiency and quality of training, and promote the improvement of the user's attention and behavior.

[0063] When designing personalized training strategies, the adaptive modulation unit 4 also performs the following steps:

[0064] A reinforcement learning mechanism is introduced, providing corresponding rewards or penalties based on the user's real-time performance during training. Clear behavioral judgment rules are set; for example, a user receives a +10 reward for completing the target task correctly 3 times consecutively, and a -5 penalty is imposed if attention is lost and the task times out. The user's current three-dimensional attention shift index, task completion progress, and historical training data are integrated into a state space. Adjustable task difficulty levels, training duration options, and feedback mode types constitute the action space. Reward and penalty values ​​calculated in real-time according to behavioral rules serve as reward signals, forming a complete reinforcement learning model. This makes the training strategy adaptive, proactively guiding users to improve their attention and behavioral performance. Based on reward or penalty feedback, the task difficulty, training duration, and feedback methods in the personalized training strategy are dynamically adjusted, and the training strategy is continuously optimized until the maximum number of optimizations is reached. If a reward signal is obtained, the task difficulty is adjusted according to a gradient increase rule based on the current difficulty level, such as increasing visual interference density by 5% or adjusting the response time window. Compress the training time by 0.5 seconds. When receiving continuous rewards, appropriately extend the duration of each training session. When receiving frequent penalties, shorten the training time and increase rest intervals to prevent user fatigue or resistance. When a certain feedback method is accompanied by multiple rewards, strengthen that feedback method. If the user performs poorly under a specific feedback, switch the feedback type, for example, from visual feedback to tactile feedback. Under appropriate training intensity, the user gradually improves their attention level, significantly improving training efficiency. Set an optimization counter with an initial value of 0. Each time the strategy is adjusted based on reward and penalty feedback, the counter is incremented by 1. After each adjustment, continue to monitor user performance to obtain new reward and penalty feedback, and adjust the strategy parameters again, forming a closed-loop iteration of "monitoring-feedback-adjustment". When the counter reaches the preset maximum number of optimizations, stop the strategy optimization process and output the final training strategy. If the maximum number of optimizations has not been reached but the user has stably achieved the training goal, the process can also be terminated early. Through a systematic iterative optimization mechanism, fully tap the user's potential and ensure that the training strategy reaches the best fit.

[0065] The steps for the adaptive modulation unit 4 to output the three-dimensional evaluation spectrum are as follows:

[0066] The three-dimensional attention shift index curve was collected, and its changing trend during the training process was recorded. Before training began, the data recording module was activated, a fixed sampling interval was set, and the three-dimensional attention shift index output by the dynamic coupling analysis unit 3 was collected in real time. The collected index values ​​were stored as a data sequence in chronological order, and a continuous curve was plotted using a data visualization library. The horizontal axis of the curve represents the training time, and the vertical axis represents the three-dimensional attention shift index value. Key time nodes were marked, and scene parameter adjustment records were compiled, clarifying the time, content, and basis of each adjustment. A dedicated log recording system was established. Each time the adaptive modulation unit 4 adjusted the scene parameters of the VR task generation unit 1, the precise time of the adjustment and the specific value of the adjusted parameters were automatically recorded. Combined with the three-dimensional attention shift index curve, the basis for each adjustment was analyzed. For example, when the index was in the high shift range, the decision logic for reducing visual interference density was recorded and added to the log in the form of text notes. The user's behavior improvement rate during the training process was statistically analyzed to evaluate the training effect. Behavioral indicators related to ADHD assessment were defined, and these indicators were measured and recorded before and after training. The improvement rate of each behavioral indicator is calculated as follows: Improvement rate = (Post-training indicator value - Pre-training indicator value) / Pre-training indicator value × 100%. The overall user behavior improvement rate is obtained by combining the improvement rates of multiple behavioral indicators, providing a clear quantitative standard for judging whether the training has achieved the expected goals. The 3D attention shift index curve, scene parameter adjustment records, and user behavior improvement rate are visualized to generate a 3D evaluation map. A 3D spatial coordinate system is constructed using 3D visualization technology, with time as the X-axis, the 3D attention shift index as the Y-axis, and the user behavior improvement rate as the Z-axis. In 3D space, the 3D attention shift index curve is used as the basic data and drawn as lines. Scene parameter adjustment records are represented by cubes of different colors and sizes, with the cube position corresponding to the adjustment time, color distinguishing parameter types, and size indicating the adjustment magnitude. On the Z-axis, the user behavior improvement rate values ​​at different time points are displayed through bar charts or scatter plots. Annotations and labels are added to indicate key data and analysis conclusions, integrating multi-dimensional training data in an intuitive and three-dimensional way, reducing the difficulty of data understanding, improving information transmission efficiency, and providing a comprehensive reference for subsequent personalized training program development.

[0067] In this invention, the VR task generation unit 1 generates programmable virtual task scenes based on an attention load dynamic adjustment algorithm, the multimodal data acquisition unit 2 collects eye movement and limb movement data, the dynamic coupling analysis unit 3 constructs a three-dimensional attention shift index and associates it with ADHD symptoms, and the adaptive modulation unit 4 adjusts scene parameters, generates personalized training strategies, and outputs a three-dimensional assessment map. This solves the problem of inaccurate assessment in traditional ADHD assessment and training methods. Through multimodal data acquisition and analysis, dynamic task generation, and personalized training strategy formulation, accurate assessment and personalized training for ADHD patients are achieved, improving assessment accuracy.

[0068] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A VR task-based ADHD assessment and training system, characterized in that, Comprise the following units: The VR task generation unit (1) generates a programmable virtual task scene based on an attention load dynamic adjustment algorithm, and the scene parameters include visual interference density, target object appearance timing entropy value, and response time window dynamic compression rate; The multi-modal data acquisition unit (2) acquires the gaze point coordinate sequence of the eye movement trajectory and the frequency energy spectrum of the limb movement through the eye movement tracking module and the nine-axis inertial measurement module, respectively, wherein the eye movement data is used to calculate the gaze point drift variance, and the nine-axis inertial measurement data is used to extract the limb micro-motion energy integral value; The dynamic coupling analysis unit (3) receives the scene parameters, the gaze point drift variance, and the frequency energy spectrum of the limb movement, constructs a three-dimensional attention deviation index, and then correlates the attention dispersibility, movement inhibition defect, and decision delay of ADHD symptoms through a nonlinear mapping relationship; The adaptive modulation unit (4) adjusts the scene parameters according to the three-dimensional attention deviation index, and simultaneously generates a personalized training strategy based on the covariance matrix of the gaze point drift variance and the frequency energy spectrum of the limb movement, and finally integrates the three-dimensional attention deviation index curve, the scene parameter adjustment record, and the user behavior improvement rate to output a three-dimensional evaluation atlas; The multi-modal data acquisition unit (2) acquires data as follows: Calibrate the eye movement tracking module, acquire the gaze point coordinate sequence of the eye movement trajectory at a preset frequency during task execution, and perform denoising processing on the acquired data; Initialize and calibrate the nine-axis inertial measurement module, continuously acquire limb movement data when the user executes the VR task, extract the frequency energy spectrum of the limb movement through frequency domain analysis method, and calculate the gaze point drift variance and the limb micro-motion energy integral value. 2.The VR task-based ADHD evaluation and training system of claim 1, wherein, The VR task generation unit (1) generates a programmable virtual task scene as follows: Initialize the basic parameters of the attention load dynamic adjustment algorithm, set the initial range of visual interference density, target object appearance timing entropy value, and response time window dynamic compression rate; According to the preset task type, arrange the basic scene elements in the virtual scene, construct the initial task framework, and according to the attention load dynamic adjustment algorithm, monitor the user feedback data in real time during task execution, dynamically adjust the visual interference density, target object appearance timing entropy value, and response time window dynamic compression rate according to the preset rule, and generate a dynamically changing virtual task scene. 3.The VR task-based ADHD evaluation and training system of claim 1, wherein, The dynamic coupling analysis unit (3) constructs a three-dimensional attention deviation index as follows: Normalize the received scene parameters, gaze point drift variance, and frequency energy spectrum of the limb movement to unify the data scale, and based on the normalized data, construct sub-indices from three dimensions of visual interference response, eye movement stability, and limb movement control, combine the sub-indices by weighting, and form a three-dimensional attention deviation index. 4.The VR task-based ADHD evaluation and training system of claim 3, wherein, The dynamic coupling analysis unit (3) correlates ADHD symptoms through a nonlinear mapping relationship as follows: A database of ADHD symptoms is established, including typical data samples of different degrees of attention dispersion, motor inhibition defects, and decision-making delays. A nonlinear mapping model is trained using machine learning methods, with the three-dimensional attention deviation index as input and the ADHD symptom data as output. The real-time calculated three-dimensional attention deviation index is input into the trained nonlinear mapping model to obtain the correlation results of the attention dispersion, motor inhibition defects, and decision-making delays of ADHD symptoms. 5.The VR task-based ADHD evaluation and training system of claim 4, wherein, The dynamic coupling analysis unit (3) also performs the following steps when training the nonlinear mapping model: The data in the ADHD symptom database is divided into training set, validation set and test set according to a certain proportion. During the training process, the validation set is used to periodically evaluate the model and adjust the model hyperparameters. The test set is used to verify the final performance of the trained model.

6. The VR task-based ADHD assessment and training system of claim 1, wherein, The adaptive modulation unit (4) adjusts the scene parameters as follows: Set different threshold intervals for the three-dimensional attention deviation index, each interval corresponds to a different scene parameter adjustment strategy, monitor the three-dimensional attention deviation index in real time, and determine its threshold interval; According to the threshold interval, adjust the visual interference density, target object appearance timing entropy value, and response time window dynamic compression rate in the VR task generation unit (1) according to the preset adjustment strategy.

7. The VR task-based ADHD assessment and training system of claim 1, wherein, The adaptive modulation unit (4) generates individualized training strategies as follows: Calculate the covariance matrix of the gaze point drift variance and the frequency energy spectrum of the limb movement, analyze the correlation between the two, and according to the correlation characteristics reflected by the covariance matrix, combined with the individual characteristics and historical training data of the user, formulate targeted training objectives; Based on the training objectives and the current task scenario, design an individualized training strategy that includes task difficulty adjustment, training duration planning, and feedback methods. 8.The VR task-based ADHD evaluation and training system of claim 7, wherein, The adaptive modulation unit (4) also performs the following steps when designing individualized training strategies: Introduce a reinforcement learning mechanism, give appropriate rewards or punishments according to the user's real-time performance during training, dynamically adjust the task difficulty, training duration, and feedback methods in the individualized training strategy according to the reward or punishment feedback, and continuously optimize the training strategy until the maximum optimization times are reached. 9.The VR task-based ADHD assessment and training system of claim 1, wherein, The adaptive modulation unit (4) outputs the three-dimensional evaluation atlas as follows: Collect the three-dimensional attention deviation index curve, record the change trend during the training process, organize the scene parameter adjustment records, specify the time, content and adjustment basis of each adjustment, calculate the behavior improvement rate of the user during the training process, evaluate the training effect, visualize the three-dimensional attention deviation index curve, scene parameter adjustment records and user behavior improvement rate, and generate a three-dimensional evaluation atlas.

Images