Intelligent screening and evaluation system for parkinson disease based on multi-data driving

Abstract

The application provides a Parkinson's disease intelligent screening and evaluation system based on multi-data driving, belongs to the technical field of artificial intelligence, collects hand movement data of a patient based on a multi-modal sensor glove, pre-processes the collected hand movement data to obtain a feature vector corresponding to the corresponding sensor, converts a UPDRS scale evaluation result into numerical data, matches the numerical data with the feature vector to obtain a four-dimension evaluation vector of the patient, trains and tests the four-dimension evaluation vector by using a multi-source heterogeneous evaluation algorithm to obtain a clustering evaluation model, and outputs an evaluation score according to the input four-dimension evaluation vector to obtain a comprehensive disease classification of the patient. The Parkinson's disease intelligent screening and evaluation method and system based on multi-data driving can timely find the disease level of the patient, improve the objectivity and accuracy of the evaluation, and overcome the subjectivity, fuzziness and incoherence of the traditional evaluation scale.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a Parkinson's disease intelligent screening and evaluation system based on multi-data driving. BACKGROUND

[0002] The current Parkinson's disease is mainly diagnosed by artificial combined with UPDRS scale. UPDRS adopts a grading score. Doctors evaluate the clinical manifestations of patients and give corresponding scores according to the degree. The total score is 199. Each item has five levels of 0, 1, 2, 3 and 4. The higher the score, the more serious the Parkinson's disease symptoms. The scoring value of each item in 18-31 has a difference of 0.5 in four levels of 0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5 and 4. The higher the score, the more serious the PD symptoms. The traditional Parkinson's diagnosis mainly has the following problems:

[0003] (1) Large subjective bias, evaluation depends on the subjective description of patients and the naked eye observation of doctors, such as the distance, amplitude and frequency of action, lack of quantitative indicators, and different doctors have different diagnosis results.

[0004] (2) The evaluation and diagnosis of the scale takes a long time. It takes more than 30 minutes for a doctor to complete a UPDRS evaluation, which consumes a lot of time and communication costs of doctors and patients. It takes more than one year from the onset to the diagnosis.

[0005] (3) The scale response is not sufficient. The most widely used UPDRS scale cannot balance the degree of core motor symptoms of Parkinson's disease. The instructions are ambiguous, lack of quantitative indicators, and the evaluation of early Parkinson's disease patients is not ideal.

[0006] At the same time, there are many Parkinson's detection products on the market, mainly using cameras to identify visual recognition body movements, or using single-dimension sensors to collect single-dimension data, but they also have the following defects:

[0007] (1) Single evaluation dimension. Most products on the market use single dimension (such as gait or grip strength) for evaluation, which has low accuracy.

[0008] (2) Limited by site and environment. Some products also collect action nodes through cameras, which are strictly limited by site. SUMMARY

[0009] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a Parkinson's disease intelligent screening and evaluation method based on multi-data driving, which can timely find the disease level of patients, improve the objectivity and accuracy of evaluation, and overcome the subjectivity, ambiguity and incoherence of traditional evaluation scale.

[0010] The second object of the present application is to provide a Parkinson's disease intelligent screening and evaluation system based on multi-data driving, which can realize accurate detection of the hand motor function of Parkinson's disease patients through the cooperative work of multiple sensors, improve the reliability and effectiveness of the data, avoid the limitations and errors of a single sensor, and has the advantages of simplicity, comfort, low cost, easy operation, etc., so that patients can be evaluated in a family or community environment, reducing the dependence on and consumption of hospital resources.

[0011] To achieve the above-mentioned object, the present application provides the following scheme: a Parkinson's disease intelligent screening and evaluation method based on multi-data driving, comprising the following steps:

[0012] Step one, collecting hand movement data of patients based on a multi-modal sensor glove, preprocessing the collected hand movement data to obtain a feature vector corresponding to the corresponding sensor;

[0013] Step two, converting the UPDRS scale evaluation result into numerical data and matching it with the feature vector to obtain a four-dimensional evaluation vector of the patient;

[0014] Step three, training and testing the four-dimensional evaluation vector using a multi-source heterogeneous evaluation algorithm to obtain a clustering evaluation model, and outputting an evaluation score according to the input four-dimensional evaluation vector to obtain the comprehensive disease classification of the patient.

[0015] Preferably, in the step one, four sensors, i.e., an acceleration angular velocity sensor, a bending sensor, a pressure sensor and an electromyographic signal sensor, are installed on the multi-modal sensor glove, and the hand tremor frequency, the finger bending angle, the pinching frequency of the index finger and the thumb, and the small arm electromyographic signal data are collected in real time through the four sensors.

[0016] Preferably, the acceleration angular velocity sensor is a six-axis motion tracking device integrating a three-axis gyroscope and a three-axis accelerometer, preferably an MPU6050 sensor, which is installed above the back of the multi-modal sensor glove to measure the three-axis acceleration and three-axis angular velocity when the hand trembles.

[0017] Preferably, the bending sensor is a flexible sensor based on the principle of resistance change, preferably a Flex'4.5 bending sensor, which is installed at the index finger, middle finger and ring finger of the multi-modal sensor glove to measure the bending degree of the finger joints.

[0018] Preferably, the pressure sensor is a force-sensitive resistor based on the principle of piezoelectric effect, preferably an FSR pressure sensor, which is installed at the index finger part of the multi-modal sensor glove to measure the pressure when the index finger and the thumb pinch.

[0019] Preferably, the electromyographic signal sensor is an electrode patch based on the principle of bioelectric signals, which is worn on the upper arm of the patient to measure the electrical signals generated during muscle contraction.

[0020] Preferably, in the second step, before obtaining the four-dimensional evaluation vector of the patient, the following four aspects need to be evaluated:

[0021] In the stability evaluation aspect, the tremor frequency data of the patient's hand are collected by the acceleration angular velocity sensor, and the CNN-LSTM algorithm is used in combination with the UPDRS scale to evaluate the stability of the patient's hand.

[0022] In the flexibility evaluation aspect, the finger bending angle data are collected by the bending sensor, and after normalization processing, the TAM evaluation method is used to judge the flexibility degree of the patient's hand.

[0023] In the coordination evaluation aspect, the number of times of pinching of the index finger and the thumb is collected by the pressure sensor, and compared with the UPDRS scale to judge the coordination disorder symptoms and the coordination disorder degree of the patient.

[0024] In the muscle strength evaluation aspect, the electromyographic signal data of the patient's arm are collected by the electromyographic signal sensor, and compared with the electromyographic signal data of the normal population to judge the muscle rigidity symptoms and the muscle rigidity degree of the patient.

[0025] Preferably, in the third step, the input of the multi-source heterogeneous evaluation algorithm comes from the data of the four sensors of the multi-modal sensor glove, the obtained data is trained to obtain an evaluation sub-model, and then four sub-models are integrated by model fusion to construct a clustering evaluation model, so as to map the input multi-source heterogeneous sensor data to different evaluation levels.

[0026] The application also provides a Parkinson's disease intelligent screening and evaluation system based on multi-data driving, which comprises a perception layer, a transmission layer, a platform layer and an application layer.

[0027] The perception layer collects the hand movement data of the patient through the multi-modal sensor glove.

[0028] The transmission layer transmits the hand movement data of the patient collected by the perception layer to the next process through the wireless transmission module.

[0029] The platform layer adopts four-dimensional evaluation and multi-source heterogeneous evaluation algorithm reasoning identification based on the input hand movement data to obtain evaluation data and a clustering evaluation model.

[0030] The application layer displays the evaluation data and results of the patient by using a mobile phone, a tablet computer or an embedded device.

[0031] Preferably, it also includes a memory and a processor for executing the computer management program of the memory to realize the steps of the method for intelligent screening and evaluation of Parkinson's disease based on multi-data driving.

[0032] According to the specific technical scheme provided by the present application, compared with the prior art, the following technical effects are disclosed:

[0033] (1) Data collection dimension: In the analysis of data, the existing technical scheme has only single-dimensional data evaluation or low integration, inaccurate evaluation and inconvenient wearing, while the present application fuses four-dimensional multi-source heterogeneous data to obtain evaluation results, and the reliability of the evaluation is higher.

[0034] (2) Evaluation method: The existing technical scheme has the problems of less hand evaluation data dimension and low sensitivity of video evaluation method for evaluating the hand clenched action of Parkinson's patients, while the present application mainly adopts the design of multi-modal sensor gloves, integrates MPU6050 sensor, bending sensor, pressure sensor and electromyographic signal sensor in the gloves, and is responsible for collecting hand tremor frequency, finger bending angle, index finger and thumb pinch frequency and small arm electromyographic signal data respectively. These data can reflect the hand movement function of Parkinson's patients, such as stability, flexibility, coordination and muscle strength. Through the cooperative work of multiple sensors, the precise detection of the hand movement function of Parkinson's patients can be realized, the reliability and effectiveness of the data can be improved, and the limitations and errors of single sensor can be avoided. The design of multi-sensor wearable gloves is simple, comfortable, low-cost and easy to operate, which can facilitate patients to evaluate in the family or community environment, reduce the dependence and consumption of hospital resources. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical scheme in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0036] Figure 1 The flow chart is provided for the method for intelligent screening and evaluation of Parkinson's disease based on multi-data driving of the present application;

[0037] Figure 2 The system panoramic view is provided for the system for intelligent screening and evaluation of Parkinson's disease based on multi-data driving of the present application;

[0038] Figure 3 The technical architecture schematic diagram is provided for the present application;

[0039] Figure 4 A stability evaluation flowchart provided for the first embodiment of the present application;

[0040] Figure 5 A data packet schematic diagram provided for the first embodiment of the present application;

[0041] Figure 6 Nine data packet schematic diagrams provided for the first embodiment of the present application;

[0042] Figure 7 A feature tensor schematic diagram provided for the first embodiment of the present application;

[0043] Figure 8 A flexibility evaluation flowchart provided for the second embodiment of the present application;

[0044] Figure 9 A feature tensor schematic diagram provided for the second embodiment of the present application;

[0045] Figure 10 A coordination evaluation flowchart provided for the third embodiment of the present application;

[0046] Figure 11 A feature tensor schematic diagram provided for the third embodiment of the present application;

[0047] Figure 12 A muscle strength evaluation flowchart provided for the fourth embodiment of the present application;

[0048] Figure 13 A feature tensor schematic diagram provided for the fourth embodiment of the present application;

[0049] Figure 14 A feature tensor fusion schematic diagram provided for the fifth embodiment of the present application; DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely below with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0051] Figure 1 A flowchart provided for a Parkinson's disease intelligent screening and evaluation method based on multiple data driving, as shown in Figure 1 includes the following steps:

[0052] Step one, based on a multi-modal sensor glove, collecting patient hand movement data, pre-processing the collected hand movement data, and obtaining the feature vector corresponding to the corresponding sensor;

[0053] Step two, convert the UPDRS scale evaluation results into numerical data and match them with the feature vector to obtain the patient's four-dimensional evaluation vector;

[0054] Step three, train and test the four-dimensional evaluation vector using a multi-source heterogeneous evaluation algorithm to obtain a clustering evaluation model, and output an evaluation score based on the input four-dimensional evaluation vector to obtain the patient's comprehensive disease classification.

[0055] In the above steps, the implementation is as follows:

[0056] Four sensors, including acceleration and angular velocity sensors, bending sensors, pressure sensors, and electromyographic signal sensors, are installed on the multi-modal sensor glove. Real-time data of hand tremor frequency, finger bending angle, index finger and thumb pinch frequency, and small arm electromyographic signal are collected through the four sensors.

[0057] The acceleration and angular velocity sensor is a six-axis motion tracking device integrating a three-axis gyroscope and a three-axis accelerometer, preferably an MPU6050 sensor, which is installed on the upper back of the multi-modal sensor glove to measure the three-axis acceleration and three-axis angular velocity when the hand trembles.

[0058] The bending sensor is a flexible sensor based on the principle of resistance change, preferably a Flex'4.5 bending sensor, which is installed on the index finger, middle finger, and ring finger of the multi-modal sensor glove to measure the bending degree of the finger joints.

[0059] The pressure sensor is a force-sensitive resistor based on the principle of piezoelectric effect, preferably an FSR pressure sensor, which is installed on the index finger part of the multi-modal sensor glove to measure the pressure when the index finger and thumb pinch.

[0060] The electromyographic signal sensor is an electrode patch based on the principle of bioelectric signal, which is worn on the upper arm of the patient to measure the electrical signal generated by muscle contraction.

[0061] Before obtaining the patient's four-dimensional evaluation vector, the following four aspects need to be evaluated:

[0062] In the stability evaluation aspect, the patient's hand tremor frequency data is collected by the acceleration and angular velocity sensor, and the CNN-LSTM algorithm is used in combination with the UPDRS scale to evaluate the patient's hand stability;

[0063] In the flexibility evaluation aspect, the finger bending angle data is collected by the bending sensor and processed by normalization, and the TAM evaluation method is used to judge the flexibility of the patient's hand;

[0064] In coordination evaluation, the number of pinch of the index finger and the thumb is collected by the pressure sensor, and compared with the UPDRS scale to determine the patient's coordination disorder symptoms and coordination disorder degree;

[0065] In muscle strength evaluation, the myoelectric signal data of the patient's forearm is collected by the myoelectric signal sensor, and compared with the myoelectric signal data of the normal population to determine the patient's muscle rigidity symptoms and muscle rigidity degree.

[0066] The input of the multi-source heterogeneous evaluation algorithm comes from the data of the four sensors of the multi-modal sensor glove, the obtained data is trained to obtain an evaluation sub-model, then four sub-models are integrated by model fusion to construct a clustering evaluation model, so as to map the input multi-source heterogeneous sensor data to different evaluation levels.

[0067] As shown in Figure 2 The application further provides a multi-data-driven Parkinson's disease intelligent screening and evaluation system, which comprises a perception layer, a transmission layer, a platform layer and an application layer.

[0068] The perception layer uses Hi3861 chips of HiSilicon to carry OpenHarmony1.0 operating system, accesses acceleration angular velocity sensors, bending sensors, pressure sensors and myoelectric signal sensors, and is integrated into a multi-modal sensor glove, and collects tremor data of a patient's hand, bending angles of the index finger, the middle finger and the ring finger, pinch times of the index finger and the thumb and forearm myoelectric signal through preprocessing and filtering respectively;

[0069] The transmission layer accesses the network through WiFi and transmits data to the Huawei cloud IoT platform through the MQTT (Message Queuing Telemetry Transport) protocol;

[0070] The platform layer uses the Huawei cloud IoT DA (IoT Device Access) device access service to realize data cloud uploading, uses a rule engine to forward data to the Huawei cloud ECS (Elastic Cloud Server) server, and the ECS is deployed with a project based on Nginx (a high-performance HTTP and reverse proxy web server) + uWSGI (a web server) + Flask (a lightweight web application framework).

[0071] The evaluation algorithm comprises four-dimensional evaluation and multi-source heterogeneous evaluation algorithm, the four-dimensional evaluation comprises stability, flexibility, coordination and muscle strength evaluation, the multi-source heterogeneous evaluation algorithm can obtain a comprehensive disease classification of Parkinson's disease, and data is stored in a self-built MySQL (a relational database management system application software) database after reasoning and identification by the two algorithms.

[0072] Finally, the evaluation data and results of the testee are displayed through a mobile phone, a tablet or an embedded device, so that the disease condition can be analyzed in time and adjusted in time.

[0073] Therefore, based on the above system, screening and evaluation are carried out in the following manner, specifically as follows:

[0074] Referring to Figure 3 When detection is performed, the user first clicks the corresponding evaluation button (stability evaluation, flexibility evaluation, coordination evaluation, muscle strength evaluation) in the APP, sends an HTTP request to the ECS server, the server calls the Huawei cloud IoT command issuing API interface, and the starting of the data acquisition device can be controlled; the corresponding sensor starts collecting data immediately, and the data collection is completed after 5S-10S, and is transmitted to the IoT platform through the MQTT protocol, the data is forwarded to the Huawei cloud ECS server through the data forwarding service in the IoT platform rule engine, the quantitative data is stored in the self-built MySQL database, and the corresponding algorithm is executed to evaluate the disease condition, and finally displayed to the APP.

[0075] The application will be further described through specific embodiments.

[0076] Example 1

[0077] Acceleration angle velocity acquisition and stability evaluation, the specific scheme includes:

[0078] Referring to Figure 4 The patient sits on a chair, the elbow joint is close to the trunk, the upper arm and the lower arm are kept at 90 degrees, the hands are placed on the chair armrest, and the feet are comfortably placed on the floor, and kept for 10 seconds.

[0079] Command issuing: the user clicks "stability evaluation" in the APP, the ECS server receives the request to call the command issuing API interface for message communication, and controls the starting of the data acquisition device, and the accelerometer and the angular velocity meter in the MPU6050 start collecting data immediately.

[0080] Data acquisition and filtering: the sampling rate is set to 100Hz, a Butterworth filter is used as a low-pass filter to filter out high-frequency noise, the frequency of hand tremor of a Parkinson's patient is between 3-8Hz, the cutoff frequency is selected as 10Hz, and the acceleration and angular velocity data of each axis are filtered and processed, so that the frequency of hand tremor of a Parkinson's patient is retained, and the noise or irrelevant signal components with higher frequency are filtered out.

[0081] Data assembly and transmission: The filtered data is divided into samples using a sliding window method. A 200*6 (200 rows and 6 columns) window is applied to the filtered data, and the window is moved from one end to the other end to generate continuous subsequences. The window slides down by 100 rows, and the data in the next window is extracted as the next subsequence. This process continues until the window slides to the end of the filtered data. That is, each data packet contains continuous data within 2 seconds. By dividing the data using a 50% overlapping sliding window, the continuity and completeness of the data are ensured. Therefore, as shown in Figure 5 , the data of a single axis (each axis of three-axis acceleration and three-axis angular velocity) within 2 seconds is 200 rows and 1 column, and the data of six-axis data within 2 seconds is 200 rows and 6 columns. The data within 2 seconds is assembled into one data packet. The wireless module uploads the data to the Huawei IoT platform after the assembly of a single data packet is completed. A total of 9 data packets of 10 seconds of collected data are uploaded, as shown in Figure 6 , that is, 1800 rows and 6 columns of data are obtained. The IoT platform forwards the data packets to the self-built application server for subsequent processing.

[0082] Data processing and feature extraction: Referring to Figure 7 , feature extraction is performed on the 9 data packets. For one data packet, six-axis data (three-axis acceleration and three-axis angular velocity) is extracted, including mean (representing the central tendency of the data), standard deviation (representing the dispersion of the data set), skewness (representing the asymmetry of the data distribution), kurtosis (representing the degree of peak or flatness of the data distribution), spectral peak (the frequency value of the highest peak in the spectrum), and frequency energy (representing the energy distribution of the signal at different frequencies). A total of 6 features are extracted, and the features are standardized by Z-Score to obtain a 6x6 standardized feature matrix. A total of 9 6x6 standardized feature matrices are obtained from the 9 data packets.

[0083] AI inference evaluation: For 1 feature matrix, a 6x6 feature matrix is input into a model combined with convolutional neural network (CNN) and long short-term memory network (LSTM). CNN is used for feature fusion and extraction, which receives the feature matrix and outputs a feature map. LSTM receives the feature map as a sequence vector input to further extract time series features. The model structure includes 3 layers of convolution and 2 layers of LSTM layer, each layer of convolution uses 64 feature maps and a kernel size of 3 time steps. The LSTM layer models and processes the feature map in sequence. A Dropout layer is applied on the LSTM layer to reduce the overfitting of the neural network. Finally, the extracted features are classified through a fully connected layer to obtain the hand stability evaluation grade of the data packet. The stability evaluation grade is calculated for 9 feature matrices respectively, and the average value is obtained, and finally the hand stability evaluation grade is obtained, which is divided into 0, 1, 2, 3, 4 five levels. The higher the level, the more severe the symptoms of Parkinson's patients.

[0084] Example two

[0085] Bending data collection and flexibility evaluation, the specific scheme includes:

[0086] Referring to Figure 8 As shown in the figure, the patient sits on a chair, the elbow joint is close to the trunk, the upper arm and the lower arm are kept at 90 degrees, the index finger and the thumb are separated, the other three fingers are naturally flexed, the patient's palm is opened as much as possible, and the five-finger gripping-opening action is alternately performed at the fastest speed for 10 seconds.

[0087] Command issuing: the user clicks "flexibility evaluation" in the APP, the ECS server receives the request to call the command issuing API interface for message communication, and controls the start of the data collection device. The bending sensor starts collecting data immediately.

[0088] Data collection and filtering: the sampling rate is set to 100Hz, for each bending sensor, a median filter is used to filter the collected bending signal, and 5 is selected as the window size (the window is a fixed length of data point set, and the data in a certain time period is selected from the signal sequence. The window size is set to 5, which means that 5 consecutive data points are selected from the signal each time), which can respond to signal changes faster and smooth signals better. The data point set in the window is sorted, and the data point at the middle position is selected as the median. The median replaces the data point at the center position of the window, that is, the median is used as the filtered output value. The window is slid forward by a fixed step, and the data of the next window is filtered. The above filtering process is repeated until the filtering is completed.

[0089] Data assembly and transmission: similar to the acceleration angular velocity assembly and transmission method, the filtered angle is divided into samples using a sliding window with 50% overlap. Each data packet contains continuous data within 2 seconds. The data of one finger is 200 rows x 1 column, and the data of five fingers is 200 rows x 5 columns. The data is assembled into one data packet, and transmitted to the Internet of Things platform every 2 seconds through the wireless module. A total of 10 seconds of data is uploaded, with a total of 9 data packets, resulting in 1800 rows x 6 columns of data.

[0090] Data processing and feature extraction: referring to Figure 9 the figure, feature extraction is performed on the 9 data packets. For one data packet, feature extraction is performed on the five finger data, including average, maximum, amplitude, average speed, average acceleration, and frequency, a total of 6 features. The features are standardized by Z-Score, and a 6x5 standardized feature matrix is obtained. A total of 9 6x5 standardized feature matrices are obtained from 9 data packets.

[0091] TAM evaluation method for flexibility: for one feature matrix, each feature is multiplied by the corresponding weight, and the scores of the 6 features are added to obtain the feature scores of the 5 fingers. The average of the feature scores of the 5 fingers is the hand flexibility score of the data packet. The flexibility scores of the 9 feature matrices are calculated and averaged to obtain the final hand flexibility score. According to the standard grade division rule of healthy people, the hand flexibility score is mapped to a grade of 0 to 4. Grades 0 to 0.2 are grade 4, grades 0.2 to 0.4 are grade 3, grades 0.4 to 0.6 are grade 2, grades 0.6 to 0.8 are grade 1, and grades 0.8 to 1.0 are grade 0. The higher the grade, the more severe the symptoms of Parkinson's patients.

[0092] Example three

[0093] Finger pinch pressure data acquisition and coordination evaluation, the specific scheme includes:

[0094] Referring to Figure 10 the figure, the patient sits on a chair with the elbow joint close to the torso, keeping the upper arm and forearm at 90 degrees. The index finger and thumb are separated, and the other three fingers are naturally flexed. The patient extends the thumb and index finger, opens them as much as possible, and alternately taps them at the fastest speed for 10 seconds.

[0095] Command issuance: the user clicks "coordination evaluation" in the APP, and the ECS server receives the request to call the command issuance API interface for message communication to control the start of the data acquisition device, and the pressure sensor starts collecting data immediately.

[0096] Data acquisition and filtering: The sampling rate is set to 100 Hz. For the pressure sensor, similar to the bending signal filtering, a median filter is used to filter the collected pressure signal, with a window size of 5, so that the window slides in the signal.

[0097] Data assembly and transmission: Similar to the acceleration and angular velocity assembly and transmission method, the filtered pressure values are divided into samples using a sliding window with 50% overlap. Each data packet contains continuous data for 2 seconds, and 200 rows of 1 column of data are assembled into 1 data packet. The wireless module transmits a data packet to the Internet of Things platform every 2 seconds, a total of 10 seconds of data, a total of 9 data packets, resulting in 1800 rows of 1 column of data.

[0098] Data processing and feature extraction: Referring to Figure 11 , feature extraction is performed on the 9 data packets. For a data packet, feature extraction is performed on the pressure signal, including 6 features: mean, maximum, amplitude, average pinch speed, maximum pinch speed, and frequency. The features are standardized by Z-Score, resulting in a 6x1 standardized feature matrix. Nine data packets result in nine 6x1 standardized feature matrices.

[0099] TAM evaluation method for flexibility: For a feature matrix, each feature is multiplied by the corresponding weight, and the scores of the 6 features are added to obtain a feature score, which reflects the hand coordination score of the data packet. The flexibility scores of the 9 feature matrices are calculated and averaged to obtain the hand flexibility score. According to the standard grade division rules for healthy people, the hand coordination score is mapped to a grade of 0 to 4, 0 to 0.2 for grade 4, 0.2 to 0.4 for grade 3, 0.4 to 0.6 for grade 2, 0.6 to 0.8 for grade 1, and 0.8 to 1.0 for grade 0. The higher the grade, the more severe the symptoms of Parkinson's patients.

[0100] Example Four

[0101] Electromyographic signal acquisition and muscle strength evaluation, the specific scheme includes:

[0102] Referring to Figure 12 , the patient sits on a chair, the patient's palm is downward, the arm is straight in front of the body, and the palm is continuously turned completely upward and downward at the fastest speed for 10 seconds.

[0103] Command issuance: The user clicks "muscle strength evaluation" in the APP, the ECS server receives the request and calls the command issuance API interface for message communication to control the start of the data acquisition device, and the electromyographic sensor starts collecting data immediately.

[0104] Data acquisition and filtering: the sampling rate is set to 100 Hz, an adaptive filter based on the least mean square (LMS) algorithm is used to remove noise and interference, a random number generator is used to generate an initial weight coefficient vector close to 0 to set the weight coefficients of the adaptive filter, and the weight coefficients of the filter are adjusted according to the iterative update rule of the LMS algorithm. Finally, the effective signal of the electromyogram is extracted.

[0105] Data assembly and transmission: similar to the above assembly and transmission method, the filtered electromyogram signal is divided into samples using a sliding window overlap of 50%, each data packet contains continuous data within 2 seconds, 200 rows and 1 column of data are assembled into 1 data packet, and 1 data packet is transmitted to the Internet of Things platform every 2 seconds through the wireless module. A total of 10 seconds of data is uploaded, a total of 9 data packets, that is, 1800 rows and 1 column of data are obtained.

[0106] Data processing and feature extraction: referring to Figure 13 , feature extraction is performed on 9 data packets. For 1 data packet, feature extraction is performed on the electromyogram signal, including 6 features such as mean value, peak-to-peak value, root mean square, power spectral density, frequency band energy and frequency. The normalized feature matrix of 6x1 is obtained by Z-Score standardization of the features, and 9 normalized feature matrices of 6x1 are obtained for 9 data packets.

[0107] TAM evaluation method to evaluate muscle strength: for 1 feature matrix, each feature is multiplied by the corresponding weight to obtain the normalized feature, and the scores of the 6 features are added to obtain the feature score, which reflects the muscle strength evaluation score of the data packet. The muscle strength evaluation score is calculated for 9 feature matrices, and the average value is calculated to obtain the final muscle strength evaluation score. According to the standard grade division rule of healthy people, the hand coordination score is mapped to the grade of 0 to 4, 0 to 0.2 is grade 4, 0.2 to 0.4 is grade 3, 0.4 to 0.6 is grade 2, 0.6 to 0.8 is grade 1, and 0.8 to 1.0 is grade 0. The higher the grade, the more severe the symptoms of Parkinson's patients.

[0108] Example five

[0109] Referring to Figure 14 , a multi-source heterogeneous comprehensive evaluation algorithm is used for training and testing to obtain a clustering evaluation model, specifically:

[0110] The standardized feature tensors of hand stability, hand flexibility, hand coordination and muscle strength evaluation are obtained through the four-dimensional evaluation of example one, example two, example three and example four, and the shapes are 9x6x6, 9x6x5, 9x6x1 and 9x6x1 respectively. The second dimension is spliced to obtain a standardized fusion feature tensor of 9x6x13, and each data is in the range of 0 to 1.

[0111] The 9x6x13 normalized fusion feature tensor is taken as input, the number of LSTM units is set to 256, and the input data is first processed through the LSTM layer, the data is converted into a fixed length feature representation through the global maximum pooling layer, the feature conversion and fusion are performed using the fully connected layer, and the ReLU activation function is used as the activation function of the fully connected layer. The features are processed again using the LSTM layer and the fully connected layer. Finally, classification is performed through the output layer to obtain the comprehensive disease classification of Parkinson's disease.

[0112] The various embodiments are described in a progressive manner in the specification, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0113] The principles and implementation manners of the present application are described by using specific examples in the present application. The above embodiment is only used to help understand the method of the present application and its core idea; meanwhile, for the general skilled in the art, the specific implementation manner and application range can be changed according to the idea of the present application. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A system for intelligent screening and assessment of Parkinson's disease based on multi-data-driven methods, characterized in that, The system comprises a perception layer, a transmission layer, a platform layer, and an application layer; The sensing layer collects patient hand movement data through a multimodal sensor glove; The transmission layer transmits the patient's hand movement data collected by the sensing layer to the next process via a wireless transmission module; The platform layer, based on the input hand motion data, uses a four-dimensional evaluation and multi-source heterogeneous evaluation algorithm for reasoning and recognition to obtain evaluation data and a clustering evaluation model. The application layer displays the patient's assessment data and results using mobile phones, tablets, or embedded devices; It also includes a memory and a processor, wherein the processor is used to execute computer management programs of the memory to implement the steps of a method for intelligent screening and assessment of Parkinson's disease based on multiple data-driven approaches; Methods for intelligent screening and assessment of Parkinson's disease based on multiple data-driven approaches include: Step 1: Collect patient hand motion data based on multimodal sensor gloves, preprocess the collected hand motion data to obtain the feature vectors corresponding to the respective sensors; Step 2: Convert the UPDRS scale assessment results into numerical data and match them with the feature vector to obtain the patient's four-dimensional assessment vector. Step 3: Use the multi-source heterogeneous assessment algorithm to train and test the four-dimensional assessment vector to obtain the clustering assessment model, and output the assessment score according to the input four-dimensional assessment vector to obtain the patient's comprehensive condition classification. In step two, before obtaining the patient's four-dimensional assessment vector, the following four aspects need to be evaluated: In terms of stability assessment, the patient's hand tremor frequency data were collected by an accelerometer and angular velocity sensor, and the hand stability of the patient was assessed by using a CNN-LSTM algorithm combined with the UPDRS scale. In terms of dexterity assessment, finger bending angle data are collected by bending sensors, and after normalization, the TAM assessment method is used to determine the degree of dexterity of the patient's hand. In terms of coordination assessment, the number of times the index finger and thumb pinch together is collected by a pressure sensor and compared with the UPDRS scale to determine the patient's coordination disorder symptoms and degree of coordination disorder. In terms of muscle strength assessment, electromyography (EMG) signal data of the patient's forearm is collected by an EMG signal sensor and compared with EMG signal data of normal people to determine the patient's muscle rigidity symptoms and degree of muscle rigidity. In step three, the input of the multi-source heterogeneous evaluation algorithm comes from the data of the four sensors of the multimodal sensor glove. The obtained data is used to train an evaluation sub-model. Then, through model fusion, the four sub-models are integrated to construct a cluster evaluation model, thereby mapping the input multi-source heterogeneous sensor data to different evaluation levels.

2. The system for intelligent screening and assessment of Parkinson's disease based on multi-data-driven methods according to claim 1, characterized in that, In step one, four types of sensors, namely an acceleration and angular velocity sensor, a bending sensor, a pressure sensor, and an electromyography (EMG) signal sensor, are installed on the multimodal sensor glove. The four sensors collect data in real time, including hand tremor frequency, finger bending angle, number of pinches between the index finger and thumb, and forearm EMG signal data.

3. The system for intelligent screening and assessment of Parkinson's disease based on multi-data-driven methods according to claim 2, characterized in that, The acceleration and angular velocity sensor is a six-axis motion tracking device integrating a three-axis gyroscope and a three-axis accelerometer. The acceleration and angular velocity sensor is an MPU6050 sensor, which is installed on the upper back of the multimodal sensor glove to measure the three-axis acceleration and three-axis angular velocity when the hand trembles.

4. The system for intelligent screening and assessment of Parkinson's disease based on multi-data-driven methods according to claim 2, characterized in that, The bending sensor is a flexible sensor based on the principle of resistance change. The bending sensor is a Flex'4.5 bending sensor, which is installed on the index, middle, and ring fingers of the multimodal sensor glove to measure the degree of bending of the finger joints.

5. The system for intelligent screening and assessment of Parkinson's disease based on multi-data-driven methods according to claim 2, characterized in that, The pressure sensor is a force-sensitive resistor based on the piezoelectric effect principle. The pressure sensor is an FSR pressure sensor, which is installed on the index finger of the multimodal sensor glove to measure the pressure when the index finger and thumb make a pinching motion.

6. The system for intelligent screening and assessment of Parkinson's disease based on multi-data-driven methods according to claim 2, characterized in that, The electromyography (EMG) sensor is an electrode pad based on the principle of bioelectric signals. It is worn on the patient's forearm to measure the electrical signals generated during muscle contraction.

Images