Multi-scene fusion hyperactivity quantitative evaluation system and use method

Abstract

The invention provides a multi-scene fusion hyperactivity quantitative evaluation system, which is characterized in that motion data of a subject in different diagnosis scenes are acquired through a diagnosis scene data acquisition module, and motion behavior information of the subject is recorded; then, a single diagnosis scene feature extraction module is used for automatically extracting feature vectors corresponding to scene data of motion sensor time sequence data collected in the single task testing process by using a single scene representation network; using a multi-diagnosis scene efficient fusion module to take the feature vector of each scene data as the feature of a corresponding node on a graph, transmitting information among the feature representations of different scene data by adopting a graph neural network, integrating the scores of different options of a prediction scale, and optimizing a diagnosis process to reduce the task burden of a subject and improve the diagnosis accuracy of the subject. The objectivity and efficiency of symptom evaluation are remarkably improved, the accuracy and operability of ADHD clinical auxiliary diagnosis are effectively improved, and the method has wide clinical application prospects.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and specifically relates to an ADHD quantitative assessment system and a method of use for efficient integration of multiple diagnostic scenarios. Background Art

[0002] Attention deficit hyperactivity disorder (ADHD), commonly known as ADHD in my country, is one of the common neurodevelopmental disorders in childhood. Among the existing clinical diagnostic methods for ADHD, scale assessment is the most widely used. Among them, the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) is widely recognized as the authoritative standard for the diagnosis of ADHD, providing 18 symptom descriptions of ADHD. However, the scale is mainly descriptive language, lacks quantitative standards, is easily affected by human subjective factors, and has low evaluation consistency. In addition, the evaluation process is complicated, involves filling in multiple contents, and takes a long time. At present, there is still a lack of objective and accurate ADHD auxiliary diagnostic methods that are widely used in clinical practice.

[0003] For example, the Chinese patent CN111743554A discloses an attention deficit hyperactivity disorder diagnosis system based on an electroencephalogram analysis algorithm, which includes a monitoring diagnosis block, and the inner side of the monitoring diagnosis block is an arc-shaped surface suitable for the structure of the human head, a processor is arranged inside the monitoring diagnosis block, an electroencephalogram acquisition electrode is arranged on the inner side of the monitoring diagnosis block, a conductive layer is connected and installed on the top of the electroencephalogram acquisition electrode, and a retractable belt is installed on the left and right sides of the monitoring diagnosis block through a connecting mechanism, and the processor includes a data acquisition chip, a signal receiver, an algorithm module, a diagnosis classification module, a data storage module, a signal transmission module and a cloud background terminal, and the signal transmission module is also connected to a mobile device; the present invention realizes real-time electroencephalogram data monitoring and analysis processing under normal conditions or during children's play time, making the monitoring data more accurate and representative, and greatly improving the practicality and efficiency of the product. Although the patent detects the behavioral characteristics of subjects in specific aspects through the design of a single task scene, due to the different detection paradigms and interaction methods of each scene, the motor behavior patterns of subjects in each task scene often show significant differences. Although the motion sensor data in each task scene can partially reflect the abnormal behavior of the subjects, it is difficult to fully cover all the symptoms of ADHD. If the data of all task scenes are indiscriminately input into the same model, the differences and connections between the scenes will be blurred. In addition, due to the subjects' attention level, not all subjects completed all tasks during the test, resulting in missing data for some task scenes.

[0004] The main defects of the existing technology for quantitative assessment of ADHD include the following aspects:

[0005] First, it is difficult to achieve efficient and objective evaluation with the scale. The scale mainly uses descriptive language, lacks quantitative standards, is easily affected by subjective factors, and has low evaluation consistency. In addition, the evaluation process is complicated, involves filling in multiple contents, and takes a long time.

[0006] Secondly, some task scenarios have missing data. In the actual test process, due to the unstable attention of the subjects, not all of them can complete all the task scenarios, resulting in missing data for some scenarios;

[0007] Finally, a single scenario cannot achieve a comprehensive assessment. The symptoms of ADHD patients vary greatly, and a single task scenario is difficult to fully cover all relevant symptoms.

[0008] Therefore, how to make full use of the differences and complementarities of data from different task scenarios in the case of irregular missing data in multiple scenarios, and comprehensively analyze the motor behavior performance of subjects in various scenarios to achieve a more accurate quantitative assessment of ADHD symptoms is a key issue that needs to be solved urgently. Summary of the invention

[0009] Purpose of the invention: In order to overcome the above shortcomings, the present invention provides a multi-scenario fusion ADHD quantitative assessment system and a method of use. The system combines the DSM-5 scale and expert clinical diagnosis experience to design 10 scenario interactive tasks that can comprehensively cover DSM-5 symptoms. Each task also includes different difficulty levels and is evaluated in the form of games. During the evaluation process, wearable devices are used to collect the tester's motion data and task completion status in real time. Finally, an objective quantitative assessment of ADHD symptoms is achieved through an intelligent analysis algorithm, which effectively promotes a comprehensive and high-precision quantitative assessment of ADHD symptoms, thereby improving the efficiency and accuracy of clinical auxiliary diagnosis of ADHD.

[0010] Technical solution: In order to achieve the above-mentioned purpose, the present invention provides a multi-scenario fusion ADHD quantitative assessment system, which includes: a diagnostic scenario data acquisition module, a single diagnostic scenario feature extraction module and a multi-diagnostic scenario efficient fusion module. The diagnostic scenario data acquisition module collects the motion data of the subject in different diagnostic scenarios and records the subject's motion behavior information; then the single diagnostic scenario feature extraction module uses the single diagnostic scenario feature extraction module to automatically extract the feature vector corresponding to the scene data from the motion sensor time series data collected during a single task test using a single scenario characterization network; then the multi-diagnostic scenario efficient fusion module uses the feature vector of each scene data as the feature of the corresponding node on the graph, and uses a graph neural network to transfer information between the feature representations of different scene data to comprehensively predict the scores of different options in the scale.

[0011] As a further preferred embodiment of the present invention, the diagnostic scenario data acquisition module includes a large touch screen computer, a set of 3D printed interactive devices and six wearable motion sensors.

[0012] As a further preferred embodiment of the present invention, a scene test program is run on the large touch screen computer, and six wearable sensors are respectively fixed on the tester's head, waist, left and right wrists, and left and right ankles to fully perceive the tester's whole body posture changes and exercise intensity. During the test, the tester completes the designated scene task according to the prompts of the large touch screen computer, and the touch screen computer records the relevant data of the task completion.

[0013] As a further preferred embodiment of the present invention, the wearable sensor collects motion data including acceleration and angular velocity.

[0014] As a further preferred embodiment of the present invention, the data used by the single diagnostic scene feature extraction module for quantitative assessment of ADHD is the time series data collected by the subject through the motion sensor during the completion of the scene task. For the single scene data, a single scene representation network with learnable parameters is used to extract features from the single scene data. The single scene representation network uses a convolutional neural network with shared parameters between scenes to extract information. The motion sensor data of different data channels input at the same time are subjected to a step-by-step convolution operation along the time dimension, so that the multi-source time series data is abstracted through multiple convolution layers to obtain the final feature representation;

[0015] A feature extractor is constructed based on a deep residual network structure, and parameters are shared between scenes. The motion data of a single scene is temporally characterized by one-dimensional convolution. Since each subject completes a different number of scenes, it is impossible to perform unified batch processing, and each sample can only be input in sequence. Therefore, the BatchNormalization layer in the deep residual network is replaced by the Layer Normalization layer. Batch normalization is to normalize the same channel features of different samples, while layer normalization is to normalize between different channels of a single sample, which is suitable for multi-dimensional temporal data feature extraction where the channel features within the sample are closely related. Adding the LayerNormalization layer and the ReLU activation layer to the network can not only speed up the training process, but also stabilize the parameters, reduce the problems of gradient vanishing and overfitting, and thus improve the feature expression ability. The pooling layer uses the maximum pooling operation, and downsampling is performed by selecting the maximum value in the local area. The weights of the network are randomly initialized, and the convolution kernel sizes are 1×7 and 1×3. The channels are first upsampled and then downsampled. Finally, channel dimension reduction is achieved through a 1×1 convolution kernel to promote cross-channel information interaction and integration. After the feature map is flattened, it is input into two fully connected layers for processing.

[0016] Compared with the traditional deep residual network, the single scene representation network has been adjusted in three aspects. First, the two-dimensional convolution is changed to one-dimensional temporal convolution; second, the batch normalization is changed to layer normalization; finally, a 1×1 convolution kernel is added for channel dimensionality reduction. By sharing the parameters of the single scene representation network, the model structure can be simplified, allowing it to share parameters between multiple task scenarios, ensuring that the model can still achieve the same or even better generalization performance at a smaller capacity.

[0017] Since each subject completes a different number of scenario tasks and there are varying degrees of data missing, the multi-diagnostic scenario efficient fusion module uses a graph neural network to achieve the fusion of multi-scenario data and sets the scenario fusion graph to an undirected graph.

[0018] As a further preferred embodiment of the present invention, the number of nodes on the undirected graph is equal to the number of scenario tasks completed by the subject, the differences and connections between the scenario data are described by nodes and connecting edges, and the features of the data of each scene are extracted through a single scene representation network as node features on the scene fusion graph.

[0019] The scene fusion graph is defined as G = (V, E), where V represents the node set, represents a scene s node, |V| is the number of nodes, i.e. the number of task scenes, E represents the edge set, represents the relationship between scene nodes v1 and v2, and e 1,2 =e 2,1 , |E| is the relationship number. If |E|=1 is set, it indicates whether the scenes are similar. At the same time, the scene fusion graph can also be defined as G=(H,A), where H={h1,h2,…,h S} represents the features of all nodes on the scene fusion graph G, A represents the adjacency matrix of G, and A 1，2 A value of 1 indicates that there is a connecting edge between the v1 and v2 nodes at the corresponding position;

[0020] After obtaining the node feature set H = {h1, h2, ..., h S} and the connection relationship E, a graph attention network can be constructed to realize the fusion of feature information of nodes on the graph, that is, each scene. The graph attention network can adaptively pay attention to more important nodes in the process of message transmission between nodes by applying the attention mechanism in the graph neural network. Its basic node message transmission process is to calculate the attention coefficient and then perform weighted summation. The formula for calculating the attention coefficient is as follows:

[0021]

[0022] Among them, LeakyReLU is the selected activation function, is the learnable vector of the model layer l, W(l) is the affine transformation matrix of the model layer l, ·||· is the vector concatenation operator, is the feature of point i at layer l, where is the neighbor set of node f under the connection relationship E. The calculated attention coefficient is used as the weight to weight the neighbor nodes. The formula is as follows:

[0023]

[0024] Consider a multi-head attention mechanism to strengthen attention to different aspects of the model, that is, use multiple groups of attention at the same time when updating node information, and concatenate the generated features. The specific formula is as follows:

[0025]

[0026] Among them, ·||· is the vector concatenation operator, N head is the number of different attentions in the multi-head attention mechanism. The symbol with the subscript n represents the corresponding symbol under the nth attention mechanism. In summary, through the multi-layer graph attention network, each node obtains a representation vector that integrates multi-scene features. This representation vector can be further used for subsequent quantitative evaluation tasks;

[0027] The mapping network learns the mapping function between the vector representation of the network output and the labels of the scale score results. Given the output vector g at the graph level, the final classification problem can be defined as y = F(g), where y represents the label of a scale score and F represents the mapping function. Specifically, the output vector of each node in the graph attention network Global mean pooling is performed to obtain the graph-level output vector g. After two layers of fully connected network mapping, the specific scoring categories of each item on the scale are predicted, and finally the scale is made objective.

[0028] A multi-scenario fusion ADHD quantitative assessment method based on graph neural network aims to achieve quantitative assessment of ADHD symptoms with efficient fusion of multiple diagnostic scenarios. Based on artificial intelligence technology, graphs are used to model a variety of different ADHD diagnostic scenarios, a reasonable scenario connection strategy is designed, a scenario fusion graph is constructed based on scenario similarity, and a graph attention network is used to realize information interaction between scenario data, adaptively give different attention to each scene, fuse motion sensor data from multiple scenes, predict DSM-5 scale score values, and finally achieve a comprehensive and efficient quantitative assessment of ADHD symptoms.

[0029] The method for using the multi-scenario fusion ADHD quantitative assessment system of the present invention comprises the following steps:

[0030] Step 1: Wear the six wearable sensors equipped with the system on the subject's head, waist, left and right wrists, and left and right ankles, complete the connection debugging, and start running the scene test program. The system guides the subject to complete the preset scenario tasks, collects the acceleration, angular velocity and other data of the motion sensor in real time, and records the tester's performance in the scene;

[0031] Step 2: After data collection is completed, the system uses the improved deep residual network to extract fine-grained features from the data of each scene, and then integrates and analyzes these single-scene features through the graph attention network, comprehensively utilizing the complementary information of each task scene to generate a comprehensive evaluation result for the subject;

[0032] Step 3: The system completes the DSM-5 scale score prediction based on the analysis results of the algorithm. The system provides a quantitative evaluation report for each subject's performance to help doctors refer to their behavioral characteristics in different scenarios and provide strong support for clinical diagnosis;

[0033] Step 4: The final quantitative evaluation report will be automatically saved to the system, and a graphical interface will be provided to show the specific performance of the subject in each task scenario. These data can be used as a reference for doctors and parents for further clinical diagnosis or treatment plan formulation. The system also has a long-term monitoring function and can evaluate progress and adjust goals based on historical data.

[0034] Beneficial effects: Compared with the prior art, the multi-scenario fusion ADHD quantitative assessment system and use method of the present invention have the following advantages:

[0035] (1): Using the motion sensors worn by the test subjects as data sources, relying on the task scenarios integrated in the existing ADHD clinical diagnosis methods, the real-time collection and analysis of the test subjects' multi-dimensional scenario data is realized, and the final scale score is predicted through intelligent algorithms. This overcomes the limitations of traditional clinical diagnosis that relies on the subjective judgment of doctors, the poor consistency of evaluation results, and the time-consuming evaluation process, so as to achieve efficient and objective scale evaluation;

[0036] (2): The system can reduce human interference through automated analysis, provide more accurate evaluation results, and significantly improve diagnostic efficiency;

[0037] (3): The improved deep residual network is used to extract fine-grained features from the subject's motion data in each scene. The one-dimensional temporal convolution is used to capture the subtle changes in motion behavior. At the same time, the parameters of the shared single-scene representation network are used to achieve efficient sharing of the model between multiple task scenes to achieve fine-grained motion feature extraction. This simplifies the model structure, reduces the use of computing resources, and maintains efficient generalization performance at a smaller model capacity, enabling the model to accurately identify motion features in each scene.

[0038] (4): Through the multi-scene data fusion method, the graph attention network is used to learn the association and complementary information between different task scenes, and the data characteristics between each scene are comprehensively considered. The system can extract key features from multi-scene data and accurately reflect the overall movement behavior pattern of the subject;

[0039] (5): The multi-scenario fusion approach not only improves the efficiency of data utilization, but also greatly improves the accuracy of the final evaluation results, ensuring accurate prediction of the scale score:

[0040] (6): By optimizing the diagnostic process, the task burden of the subjects is reduced, the objectivity and efficiency of symptom assessment are significantly improved, and the accuracy and operability of clinical auxiliary diagnosis of ADHD are effectively improved, which has broad clinical application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The interaction mode and corresponding DSM-5 related symptoms for each task. DETAILED DESCRIPTION

[0042] The present invention is a multi-scenario fusion ADHD quantitative assessment system based on graph neural network, which combines wearable devices and artificial intelligence technology to provide a more objective and accurate quantitative assessment for auxiliary diagnosis of ADHD. The present invention intelligently integrates motion data under multiple task scenarios, comprehensively analyzes the motion behavior characteristics of the subjects, and finally generates quantitative assessment results. The following is a specific implementation method of the implementation process of ADHD quantitative assessment:

[0043] Example

[0044] Background: A 9-year-old child is suspected of having ADHD, with symptoms such as difficulty concentrating and hyperactivity. The parents hope to obtain a more objective diagnosis through scientific evaluation. This system collects motion data through multiple task scenarios and performs intelligent analysis on the data to help doctors complete quantitative evaluation.

[0045] 1. Equipment configuration and data collection

[0046] The six wearable sensors equipped with the system are worn on the subject's head, waist, left and right wrists, and left and right ankles, and the connection debugging is completed, and the scene test program is started. The system guides the subject to complete the preset scenario tasks, collects the acceleration, angular velocity and other data of the motion sensor in real time, and records the tester's performance in the scene;

[0047] 2. Intelligent analysis and quantitative evaluation

[0048] After data collection is completed, the system uses an improved deep residual network to extract fine-grained features from the data of each scene, and then integrates and analyzes these single-scene features through a graph attention network, comprehensively utilizing the complementary information of each task scene to generate a comprehensive evaluation result for the subject;

[0049] 3. Scale scoring and dynamic adjustment

[0050] Based on the analysis results of the algorithm, the system completes the DSM-5 scale score prediction. For each subject's performance, the system provides a quantitative evaluation report to help doctors refer to their behavioral characteristics in different scenarios and provide strong support for clinical diagnosis.

[0051] 4. Feedback and application of evaluation results

[0052] The final quantitative evaluation report will be automatically saved in the system, and a graphical interface will be provided to show the specific performance of the subjects in each task scenario. Figure 1 The interactive mode of each task and the corresponding DSM-5 related symptoms are shown. Since each detection task focuses on different interactive modes when designed, the body parts of the subjects interact are different. The similarity between scenes can be calculated based on this. Specifically, it is believed that the subtasks of different difficulties in the same scene are similar, and the connection edges are directly established. It is believed that the tasks with the same body movement mode are similar, and the connection edges can be established. The movement mode is set to only hand movement or whole body movement. Four different body movement modes can be encoded according to the presence or absence of these two movement modes. Then, the connection edges between the scene nodes are established according to the similarity of the body movement modes in each task scene, and the scene fusion graph is constructed. For each subject, that is, each sample, due to the difference in the number of completed test scenes, the scene fusion graph corresponding to each sample is irregular, which is also one of the starting points for the present invention to choose to use a graph to represent multi-scene data. The number of nodes in each graph is different, and each node in the graph has a different number of neighbor nodes, but the graph neural network can also be used for representation learning to achieve information fusion.

[0053] These data can be used as a reference for doctors and parents for further clinical diagnosis or treatment planning. The system also has long-term monitoring capabilities and can evaluate progress and adjust goals based on historical data.

[0054] The present invention introduces a scale-driven wearable device-assisted diagnosis and evaluation system, and combines intelligent algorithms to analyze scene-related data to achieve rapid, objective, and quantitative evaluation of ADHD symptoms, effectively reducing human interference and improving evaluation efficiency. The data representation method based on the graph structure organizes the data of each scene in the form of a graph, avoiding the impact of irregular data missing on the overall evaluation results. This method effectively improves data integrity and the reliability of the evaluation. By introducing a graph attention network, effective information interaction between multi-scene data is carried out, the characteristics of each scene are integrated, and the scoring results of the DSM-5 scale are accurately predicted to provide a comprehensive symptom assessment. Therefore, the present invention has strongly promoted a comprehensive and high-precision quantitative evaluation of ADHD symptoms, thereby improving the efficiency and accuracy of clinical auxiliary diagnosis of ADHD.

[0055] The above embodiments are exemplary, and their purpose is to illustrate the technical concept and features of the present invention so that people familiar with this field can understand the content of the present invention and implement it accordingly, and they cannot limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-scenario fusion ADHD quantitative assessment system, characterized by: It includes: The diagnostic scenario data acquisition module, the single diagnostic scenario feature extraction module and the multi-diagnostic scenario efficient fusion module collect the motion data of the subjects in different diagnostic scenarios and record the subjects' motion behavior information through the diagnostic scenario data acquisition module; then the single diagnostic scenario feature extraction module uses the single scenario representation network to automatically extract the feature vector corresponding to the scene data from the motion sensor time series data collected during a single task test; then the multi-diagnostic scenario efficient fusion module uses the feature vector of each scene data as the feature of the corresponding node on the graph, and uses the graph neural network to transfer information between the feature representations of different scene data to comprehensively predict the scores of different options in the scale.

2. According to claim 1, a multi-scenario fusion ADHD quantitative assessment system is characterized by: The diagnostic scenario data acquisition module includes a large touch screen computer, a set of 3D printed interactive devices and six wearable motion sensors.

3. The multi-scenario fusion ADHD quantitative assessment system according to claim 1 is characterized by: The scenario test program is run on the large touch screen computer, and six wearable sensors are respectively fixed on the tester's head, waist, left and right wrists, and left and right ankles to fully perceive the tester's whole body posture changes and exercise intensity.

4. The multi-scenario fusion ADHD quantitative assessment system according to claim 2, characterized in that: The wearable sensor collects motion data including acceleration and angular velocity.

5. The multi-scenario fusion ADHD quantitative assessment system according to claim 1, characterized in that: The single diagnostic scene feature extraction module uses a single scene characterization network with learnable parameters to extract features from the data of a single scene. The single scene characterization network uses a convolutional neural network with shared parameters between scenes to extract information. It performs step-by-step convolution operations along the time dimension on motion sensor data from different data channels input at the same time, so that the multi-source time series data is abstracted through multiple convolutional layers to obtain the final feature representation.

6. A multi-scenario fusion ADHD quantitative assessment system according to claim 5, characterized in that; In the single diagnosis scene feature extraction module, the two-dimensional convolution is changed to one-dimensional temporal convolution, the batch normalization is changed to layer normalization, and a 1×1 convolution kernel is added for channel dimensionality reduction.

7. The multi-scenario fusion ADHD quantitative assessment system according to claim 1, characterized in that; The multi-diagnostic scenario efficient fusion module uses a graph neural network to achieve the fusion of multi-scenario data and sets the scene fusion graph as an undirected graph.

8. The multi-scenario fusion ADHD quantitative assessment system according to claim 1, characterized in that; The number of nodes on the undirected graph is equal to the number of scene tasks completed by the subject. The differences and connections between the scene data are described by nodes and connecting edges. The features of the data of each scene are extracted through a single scene representation network as node features on the scene fusion graph.

9. The multi-scenario fusion ADHD quantitative assessment system according to claim 8, characterized in that; The scene fusion graph is defined as G = (V, E), where V represents a node set, represents a scene s node, |V| is the number of nodes, i.e. the number of task scenes, E represents the edge set, represents the relationship between scene nodes v1 and v2, and e 1，2 =e 2，1 , |E| is the relationship number. If |E|=1 is set, it indicates whether the scenes are similar. At the same time, the scene fusion graph can also be defined as G=(H, A), where H={h1, h2, ..., h S } represents the features of all nodes on the scene fusion graph G, A represents the adjacency matrix of G, and A 1，2 A value of 1 indicates that there is a connecting edge between the v1 and v2 nodes at the corresponding position; After obtaining the node feature set H = {h1, h2, ..., h S } and the connection relationship E, a graph attention network can be constructed to realize the fusion of feature information of nodes on the graph, that is, each scene. The graph attention network can adaptively pay attention to more important nodes in the process of message transmission between nodes by applying the attention mechanism in the graph neural network. Its basic node message transmission process is to calculate the attention coefficient and then perform weighted summation. The formula for calculating the attention coefficient is as follows: Among them, LeakyReLU is the selected activation function, is the learnable vector of the model layer l, W (l) is the affine transformation matrix of the model layer l, ·||· is the vector concatenation operator, is the feature of point i at layer l, where is the neighbor set of node i under the connection relationship E. The calculated attention coefficient is used as the weight to sum the neighbor nodes. The formula is as follows: The multi-head attention mechanism is used to strengthen the attention on different aspects of the model, that is, multiple groups of attention are used simultaneously when updating node information, and the generated features are spliced. The specific formula is as follows: Among them, ·||· is the vector concatenation operator, N head is the number of different attentions in the multi-head attention mechanism. The symbol with the subscript n represents the corresponding symbol under the nth attention mechanism. In summary, through the multi-layer graph attention network, each node obtains a representation vector that integrates multi-scene features. This representation vector can be further used for subsequent quantitative evaluation tasks; The mapping network learns the mapping function between the vector representation of the network output and the labels of the scale score results. Given the output vector g at the graph level, the final classification problem can be defined as y = F(g), where y represents the label of a scale score and F represents the mapping function. Specifically, the output vector of each node in the graph attention network Perform global mean pooling to obtain the graph-level output vector g ,After two layers of fully connected network mapping, the specific scoring categories of each item of the scale are predicted, and the scale is finally made objectivized.

10. A method for using the multi-scenario fusion ADHD quantitative assessment system according to any one of claims 1 to 6, characterized in that: The steps include: Step 1: Wear the six wearable sensors equipped with the system on the subject's head, waist, left and right wrists, and left and right ankles, complete the connection debugging, and start running the scene test program. The system guides the subject to complete the preset scenario tasks, collects the acceleration, angular velocity and other data of the motion sensor in real time, and records the tester's performance in the scene; Step 2: After data collection is completed, the system uses the improved deep residual network to extract fine-grained features from the data of each scene, and then integrates and analyzes these single-scene features through the graph attention network, comprehensively utilizing the complementary information of each task scene to generate a comprehensive evaluation result for the subject; Step 3: The system completes the DSM-5 scale score prediction based on the analysis results of the algorithm. The system provides a quantitative evaluation report for each subject's performance, helping doctors refer to their behavioral characteristics in different scenarios and providing strong support for clinical diagnosis; Step 4: The final quantitative evaluation report will be automatically saved to the system, and a graphical interface will be provided to show the specific performance of the subject in each task scenario. These data can be used as a reference for doctors and parents for further clinical diagnosis or treatment plan formulation. The system also has a long-term monitoring function and can evaluate progress and adjust goals based on historical data.

Images