A wearable multi-limb segment joint analysis method and system for hand tremor

Abstract

A wearable hand tremor multi-limb segment joint analysis method, comprising the following steps: step 1, hand multi-limb segment tremor data acquisition, synchronously collecting hand multi-limb segment tremor data through a sensing device; step 2, hand multi-limb segment data preprocessing, respectively performing posture estimation, denoising and dimension reduction processing on each limb segment; step 3, multi-limb segment joint tremor feature extraction, respectively extracting the tremor frequency, tremor amplitude and tremor variability features of each limb segment, and simultaneously extracting the coordination relationship features between limb segments; step 4, disease automatic classification and discrimination, selecting multi-limb segment joint tremor features through a feature selection algorithm, and using an intelligent classifier to complete the classification of Parkinson's disease and essential tremor. And provide a kind of described wearable hand tremor multi-limb segment joint analysis system. The present application provides a hand tremor analysis method and system for improving the classification accuracy of Parkinson's disease and essential tremor through multi-limb segment joint analysis.

Description

Technical Field

[0001] This invention pertains to the field of digital healthcare, specifically relating to a wearable method and system for multi-segment joint analysis of hand tremors. Background Technology

[0002] Hand tremor is a typical symptom of Parkinson's disease and essential tremor. Generally, hand tremors in Parkinson's patients are mainly resting tremors, while those in essential tremor patients are mainly postural tremors. However, approximately 18% of patients with essential tremor will experience resting tremor, and 90% of patients with Parkinson's disease will experience postural tremor. This makes the differential diagnosis between Parkinson's disease and essential tremor difficult in clinical practice.

[0003] Most tremor analysis systems currently focus on analyzing the frequency, amplitude, and variability of tremors in single limb segments. For example, Qingdao University. A three-dimensional posture real-time analysis system and method for hand tremors in Parkinson's disease: CN202211181166.9[P]. 2022-12-20; some visual systems, while involving the analysis of tremor frequency, amplitude, and variability in multiple limb segments, do not analyze the synergistic relationship between tremors in different limb segments. For example, Oktay AB, Kocer A. Differential diagnosis of Parkinson and essential tremor with convolutional LSTM networks[J]. Biomedical Signal Processing and Control, 2020, 56:101683. That is, using convolutional long short-term memory neural networks to differentiate between Parkinson's disease and essential tremor. Because the tremor frequency, amplitude, and variability of Parkinson's disease patients and essential tremor patients overlap, related systems have difficulty effectively distinguishing between Parkinson's disease patients and essential tremor patients with similar symptoms. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides a wearable multi-segment joint analysis method and system for hand tremors. By analyzing the tremor frequency, amplitude, and variability of each segment of the hand, and combining the synergistic relationship between tremors in different segments, it achieves automatic and accurate classification of Parkinson's disease patients and patients with essential tremor.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A method for combined multi-segment analysis of wearable hand tremor includes the following steps:

[0007] Step 1: Data acquisition of multiple limb segments of the hand. The acquisition of tremor data of multiple limb segments of the patient's hand is carried out by attaching a sensor device to each limb segment of the hand. The multiple limb segments of the hand are a subset of the candidate set of hand limb segments with more than 2 elements. The candidate set of hand limb segments includes the wrist, thumb, index finger, middle finger, ring finger, little finger and back of the hand.

[0008] Step 2: Preprocessing of hand multi-limb segment tremor data. For each limb segment, a posture fusion algorithm and a denoising filter are used to process the data to obtain the posture signal of each limb segment in Euler angle form. The principal component analysis method is used to reduce the three-dimensional Euler angle signal of each limb segment to one dimension to obtain the principal component signal of the angle of each limb segment. The limb segment refers to all elements in the set of hand multi-limb segments.

[0009] Step 3: Extract joint tremor features of multiple limb segments. The tremor frequency, tremor amplitude and tremor variability of each limb segment are obtained through power spectrum analysis and time-domain waveform analysis. The synergistic relationship between limb segment tremors is obtained through coherence spectrum analysis.

[0010] Step 4: Automatic disease classification and identification. The feature selector obtains the selected multi-limb combined tremor features, and the intelligent classifier completes the classification of Parkinson's disease and essential tremor to determine whether the user is a Parkinson's disease patient or an essential tremor patient.

[0011] Furthermore, the process of step 2 is as follows:

[0012] Step 2-1: Limb segment attitude estimation. For the sensing signals of each limb segment, the quaternion attitude fusion algorithm is used to fuse the accelerometer, gyroscope and magnetometer signals to obtain the drift-free attitude estimation signal of each limb segment, and the attitude signal is converted into a three-dimensional Euler angle signal.

[0013] Step 2-2: Denoising and dimensionality reduction of posture signals. For the Euler angle signals of each limb segment, the signal frequency is limited to the common frequency band of hand tremor by a denoising filter. Then, the three-dimensional Euler angle signal is reduced to a one-dimensional angle principal component signal by principal component analysis algorithm.

[0014] Furthermore, in step 3, the characteristics of multi-segmental tremor include:

[0015] Tremor frequency characteristics: the principal frequency, median frequency, frequency dispersion, and mean instantaneous tremor frequency of the principal component signals of each limb segment angle. The method for calculating the mean instantaneous tremor frequency is as follows:

[0016]

[0017] in Let n be the average instantaneous tremor frequency of any segment within the set of multiple hand segments, where the set of multiple hand segments is a selected collection of hand segments. S This represents the number of zero-crossing points in the principal component signal of the limb segment where the derivative is greater than 0. The time interval between the zero-crossing point of the i-th derivative greater than 0 and the (i+1)-th derivative greater than 0 in the principal component signal of the limb segment angle;

[0018] Tremor amplitude characteristics: root mean square, peak frequency, maximum peak value, and average peak value of principal component signals at different limb segments;

[0019] Tremor variability characteristics: peak value coefficient of variation, instantaneous frequency coefficient of variation, and amplitude stability coefficient. The amplitude stability coefficient is calculated as follows:

[0020]

[0021] MSI S The wrist amplitude stability coefficient is given by the formula for any segment of the hand within the set of multiple limb segments. The maximum value between the zero-crossing point of the i-th derivative of the principal component signal of the limb segment angle and the zero-crossing point of the (i+1)-th derivative, iqr is the interquartile range function;

[0022] The characteristics of the coordination relationship between limb segment tremors: the phase difference of the principal component signals of limb segment angles, wherein the phase difference of the principal component signals of limb segment angles is the phase difference between any two distinct principal component signals of limb segment angles in the set of multiple limb segments of the hand, and the calculation method is as follows:

[0023]

[0024] Where S1 and S2 are any two distinct limb segments in the set of multiple hand limb segments. The coherence spectrum is the principal component signal of angle S1 of the hand limb segment and the principal component signal of angle S2 of the hand limb segment. The autocorrelation spectrum is the principal component signal of the S1 angle of the hand limb segment. The autocorrelation spectrum is the principal component signal of the S2 angle of the hand limb segment. Let be the phase difference of the principal component signal of the angle between the hand limb segment S1 and the hand limb segment S2, where angle is a function taking the imaginary angle.

[0025] In step 4, the feature set for the feature selector is obtained through the following steps:

[0026] Step S1: Candidate feature set selection. Regularized regression algorithms such as lasso regression, ridge regression, or elastic net regression are used to initially screen the extracted 3D pattern features and traditional features to obtain the candidate feature set.

[0027] Step S2: The selected feature set is determined by using the sequential floating feature selection algorithm to obtain the selected feature set from the candidate feature set.

[0028] The intelligent classifier described herein is constructed using a selected feature set, employing support vector machines, random forests, or artificial neural networks.

[0029] The collected tremor data included resting tremor and postural tremor data in the Wing-beating posture.

[0030] The noise reduction filter uses an FIR filter with a cutoff frequency of 3-12Hz to avoid signal distortion.

[0031] The regularized regression algorithm chooses elastic net regression to address the problem that the total number of features is greater than the number of samples.

[0032] The classifier is trained using the support vector machine algorithm.

[0033] A multi-segment joint analysis system for hand tremor includes a sensing device and a computing unit. The sensing device is used to synchronously collect acceleration, angular velocity, and magnetic force data of each segment of the hand at corresponding positions and wirelessly transmit the data. The computing unit includes a data receiving module and a data processing module, which can be deployed in the embedded processing module of the sensing device or on a host computer. The data receiving module receives the data collected by the sensing device, decodes it, and sends it to the data processing module. The data processing module analyzes and processes the data and generates analysis results.

[0034] Furthermore, the sensing device includes N sensing units and one embedded processing module. The N sensing units collect tremor data from each segment of the hand. Each sensing unit integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The embedded processing module includes an embedded MCU and a wireless communication module, responsible for synchronizing the data from the N sensing units and wirelessly transmitting it to the data forwarding module. N is the number of elements in the set of multiple hand segments, which can be any integer from 2 to 6. Each hand segment refers to all elements in the set of multiple hand segments.

[0035] Furthermore, when the sensing device is used to collect vibration data, the user should sit comfortably in a chair with armrests, wear the sensing device, and conduct the vibration test.

[0036] Furthermore, the data processing module calculates the posture information of each limb segment through a posture fusion algorithm. After preprocessing the posture information, it extracts features such as tremor frequency, tremor amplitude, tremor variability, and synergistic relationship between limb segment tremors. Then, through an intelligent classifier, it determines whether the user is a Parkinson's disease patient or an essential tremor patient.

[0037] Preferably, the sensing unit uses an MPU9250 nine-axis sensing chip to improve the integration of the sensing unit and reduce its size.

[0038] Preferably, the wrist, thumb, index finger, middle finger, ring finger, and little finger are selected from the multiple limb segments of the hand, and N is set to 6.

[0039] Preferably, the five sensing units are respectively strapped to the distal interphalangeal joints of the thumb, index finger, middle finger, ring finger and little finger. The sensing units and the embedded processing module are integrated together and strapped to the wrist, thereby improving the integration of the sensing device and making the sensing device lighter.

[0040] Preferably, the embedded processing module uses 6 sets of SPI buses and adopts register bit operation mode to ensure the synchronization of the sensing units.

[0041] Preferably, when the computing unit is deployed in an embedded processing module, the sensing device only sends the combined features of multiple limb segments, and the classification results of Parkinson's disease and essential tremor.

[0042] Preferably, when the computing unit is deployed on a host computer, it should output the analysis results to the user interface. Simultaneously, it should interact with a database to store the user's basic demographic information, medical records, and previous tremor analysis results, thus facilitating access by doctors.

[0043] The beneficial effects of this invention are mainly reflected in:

[0044] 1. This invention can objectively analyze the tremor amplitude, tremor frequency, tremor variability, and synergistic relationship between tremors in each limb segment.

[0045] 2. This invention improves the classification accuracy of Parkinson's disease and essential tremor through multi-limb segment combined analysis, which is beneficial for clinical use.

[0046] 3. This invention can simultaneously collect tremor motion information from various limb segments and record tremor motion data of various limb segments in real time, objectively and accurately. Attached Figure Description

[0047] Figure 1 This is a flowchart of a wearable hand tremor multi-segment joint analysis method provided by the present invention.

[0048] Figure 2 This is a schematic diagram of the structure of a wearable hand tremor multi-limb segment joint analysis system provided by the present invention. Detailed Implementation

[0049] The present invention will now be further described with reference to the accompanying drawings.

[0050] Reference Figure 1 and Figure 2 A wearable method for combined analysis of multiple limb segments in hand tremors, the method comprising the following steps:

[0051] Step 1: Data acquisition of multiple limb segments of the hand. The acquisition of tremor data of multiple limb segments of the patient's hand is carried out by a sensor device attached to the fingers and wrist.

[0052] Step 2: Preprocessing of hand multi-limb segment tremor data. For each limb segment, a posture fusion algorithm and a denoising filter are used to process the data to obtain the posture signal of each limb segment in Euler angle form. The principal component analysis method is used to reduce the three-dimensional Euler angle signal of each limb segment to one dimension to obtain the principal component signal of the angle of each limb segment.

[0053] Step 3: Extract joint tremor features from multiple limb segments. The tremor frequency, tremor amplitude, and tremor variability of each limb segment are obtained through power spectrum analysis and time-domain waveform analysis. The synergistic relationship between limb segment tremors is obtained through coherence spectrum analysis.

[0054] Step 4: Automatic disease classification and identification. The feature selector obtains the selected multi-limb combined tremor features, and the intelligent classifier completes the classification of Parkinson's disease and essential tremor to determine whether the user is a Parkinson's disease patient or an essential tremor patient.

[0055] Furthermore, the specific steps for preprocessing the multi-segment tremor data of the hand are as follows:

[0056] Step 2-1: Limb segment attitude estimation. For the sensing signals of each limb segment, the quaternion attitude fusion algorithm is used to fuse the accelerometer, gyroscope and magnetometer signals to obtain the drift-free attitude estimation signal of each limb segment, and the attitude signal is converted into a three-dimensional Euler angle signal.

[0057] Step 2-2: Denoising and dimensionality reduction of posture signals. For the Euler angle signals of each limb segment, the signal frequency is limited to the common frequency band of hand tremor by a denoising filter. Then, the three-dimensional Euler angle signal is reduced to a one-dimensional angle principal component signal by principal component analysis algorithm.

[0058] Furthermore, the aforementioned multi-segmental combined tremor features include:

[0059] Tremor frequency characteristics: the principal frequency, median frequency, frequency dispersion, and mean instantaneous tremor frequency of the principal component signals of each limb segment angle. The method for calculating the mean instantaneous tremor frequency is as follows:

[0060]

[0061] in Let n be the average instantaneous tremor frequency of any segment within the set of multiple hand segments, where the set of multiple hand segments is a selected collection of hand segments. S This represents the number of zero-crossing points in the principal component signal of the limb segment where the derivative is greater than 0. The time interval between the zero-crossing point of the i-th derivative greater than 0 and the zero-crossing point of the (i+1)-th derivative greater than 0 in the principal component signal of the limb segment angle is denoted as .

[0062] Tremor amplitude characteristics: root mean square, peak frequency, maximum peak value, and average peak value of principal component signals for each limb segment.

[0063] Tremor variability characteristics: peak value coefficient of variation, instantaneous frequency coefficient of variation, and amplitude stability coefficient. The amplitude stability coefficient is calculated as follows:

[0064]

[0065] MSI S The wrist amplitude stability coefficient is given by the formula for any segment of the hand within the set of multiple limb segments. iqr is the maximum value between the zero-crossing point of the i-th derivative of the principal component signal of the limb segment angle that is greater than 0 and the zero-crossing point of the (i+1)-th derivative that is greater than 0, and iqr is the interquartile range function.

[0066] The characteristics of the coordination relationship between limb segment tremors: the phase difference of the principal component signals of limb segment angles, wherein the phase difference of the principal component signals of limb segment angles is the phase difference between any two distinct principal component signals of limb segment angles in a set of multiple limb segments of the hand, and the specific calculation method is as follows:

[0067]

[0068] Where S1 and S2 are any two distinct limb segments in the set of multiple hand limb segments. P is the coherence spectrum of the principal component signals of hand limb segment S1 and S2 angles. S1S1 The autocorrelation spectrum of the principal component signal at angle S1 of the hand limb segment, P S2S2 The autocorrelation spectrum of the principal component signal at the S2 angle of the hand limb segment, Phase S1S2 Let be the phase difference of the principal component signal of the angle between the hand limb segment S1 and the hand limb segment S2, where angle is a function taking the imaginary angle.

[0069] Furthermore, the selection feature set for the feature selector is obtained through the following steps:

[0070] Step S1: Candidate feature set selection. Regularized regression algorithms such as lasso regression, ridge regression, or elastic net regression are used to initially screen the extracted 3D pattern features and traditional features to obtain the candidate feature set.

[0071] Step S2: The selected feature set is determined by using the sequential floating feature selection algorithm to obtain the selected feature set from the candidate feature set.

[0072] Furthermore, the intelligent classifier is constructed using the selected feature set using support vector machines, random forests, or artificial neural networks.

[0073] Preferably, the wrist, thumb, index finger, middle finger, ring finger, and little finger are selected from the multiple limb segments of the hand.

[0074] Preferably, the collected tremor data includes tremor data of resting tremor and postural tremor.

[0075] Preferably, the noise reduction filter uses an FIR filter with a cutoff frequency of 3-12Hz to avoid signal distortion.

[0076] Preferably, the regularized regression algorithm uses the elastic net regularization algorithm to address the problem that the total number of features is greater than the number of samples.

[0077] Preferably, the classifier is trained using the support vector machine algorithm.

[0078] Reference Figure 2 A wearable multi-segment joint analysis system for hand tremor includes a sensing device and a computing unit. The sensing device is used to synchronously collect acceleration, angular velocity, and magnetic force data of each limb segment by attaching it to corresponding positions on the back of the fingers and wrist, and wirelessly transmit the data. The computing unit includes a data receiving module and a data processing module, which can be deployed in the embedded processing module of the sensing device or on a host computer. The data receiving module receives the data collected by the sensing device, decodes it, and sends it to the data processing module. The data processing module analyzes and processes the data and generates analysis results. The multiple limb segments include the thumb, index finger, middle finger, ring finger, little finger, and wrist.

[0079] Furthermore, the sensing device includes N sensing units and one embedded processing module. The N sensing units collect tremor data from each segment of the hand. Each sensing unit integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The embedded processing module includes an embedded MCU and a wireless communication module, which are responsible for synchronizing the data from the N sensing units and wirelessly transmitting it to the data forwarding module.

[0080] The data processing module fuses the triaxial acceleration, triaxial angular velocity, and triaxial magnetic field strength data of each limb segment using complementary filters to obtain the posture information of each limb segment, and extracts the tremor frequency, tremor amplitude, variability characteristics, and coordination relationship between limb segments.

[0081] Preferably, the wrist, thumb, index finger, middle finger, ring finger, and little finger are selected from the multiple limb segments of the hand, and N is set to 6.

[0082] Preferably, the five sensing units are respectively strapped to the distal interphalangeal joints of the thumb, index finger, middle finger, ring finger, and little finger. The sensing unit 6 and the embedded processing module are integrated together and strapped to the wrist, thereby improving the integration of the sensing device and making the sensing device lighter.

[0083] Preferably, the embedded processing module uses 6 sets of SPI buses and adopts register bit operation mode to ensure the synchronization of the sensing units.

[0084] Preferably, the computing unit is deployed on a host computer and includes a database and a user interface. The database is used to store the patient's demographic and medical records, and can also record the patient's historical test data results, facilitating doctors' understanding of the patient's disease progression and thus adjusting treatment measures. The user interface should include functions for adding, deleting, and modifying the patient's demographic and medical records, and should be able to display the tremor frequency, tremor amplitude, tremor variability, and coordination relationships between tremors in different limb segments.

[0085] Based on the above methods and systems, this invention collected tremor data of the left and right hands from 19 age- and sex-matched Parkinson's disease patients and 12 patients with essential tremor, and constructed a dataset. The dataset contains 38 hand tremor data points from Parkinson's disease patients and 24 hand tremor data points from patients with essential tremor. There were no significant differences in resting tremor and postural tremor scores between the Parkinson's disease and essential tremor patients. The accuracy results of leave-one-out validation are shown in Table 1, which compares the accuracy of this invention with other methods.

[0086]

[0087] Table 1

[0088] Specifically, the accuracy of the method of the present invention within the dataset was verified when N=2, in order to facilitate the selection of a suitable set of hand segments in cost-sensitive scenarios. The verification results are shown in Table 2, which presents the accuracy of the method of the present invention when N=2.

[0089]

[0090]

[0091] Table 2

[0092] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A wearable multi-segment joint analysis system for hand tremors, characterized in that, The system includes a sensing device and a computing unit. The sensing device is used to synchronously collect acceleration, angular velocity, and magnetic force data of each limb segment by attaching it to corresponding positions on each limb segment and wirelessly transmit the data. The computing unit includes a data receiving module and a data processing module, which can be deployed in the embedded processing module of the sensing device or on a host computer. The data receiving module receives the data collected by the sensing device, decodes it, and sends it to the data processing module. The data processing module analyzes and processes the data and generates analysis results. The multi-limb joint analysis method implemented by the system includes the following steps: Step 1: Data acquisition of multiple limb segments of the hand. The tremor data of multiple limb segments of the hand is collected by attaching a sensor device to each limb segment of the hand. The multiple limb segments of the hand are a subset of the candidate set of limb segments with more than 2 elements. The candidate set of limb segments of the hand includes the wrist, thumb, index finger, middle finger, ring finger, little finger and back of the hand. Step 2: Preprocessing of hand multi-limb segment tremor data. For each limb segment, a posture fusion algorithm and a denoising filter are used to process the data to obtain the posture signal of each limb segment in Euler angle form. The principal component analysis method is used to reduce the three-dimensional Euler angle signal of each limb segment to one dimension to obtain the principal component signal of the angle of each limb segment. The limb segment refers to all elements in the set of hand multi-limb segments. Step 3: Extract joint tremor features of multiple limb segments. The tremor frequency, tremor amplitude and tremor variability of each limb segment are obtained through power spectrum analysis and time-domain waveform analysis. The synergistic relationship between limb segment tremors is obtained through coherence spectrum analysis. Step 4: Automatic disease classification and identification. The feature selector obtains the selected multi-limb combined tremor features, and the intelligent classifier completes the classification of Parkinson's disease and essential tremor to determine whether the user is a Parkinson's disease patient or an essential tremor patient. In step 3, the characteristics of multi-segmental tremor include: Tremor frequency characteristics: the principal frequency, median frequency, frequency dispersion, and mean instantaneous tremor frequency of the principal component signals of each limb segment angle. The method for calculating the mean instantaneous tremor frequency is as follows: in Let n be the average instantaneous tremor frequency of any segment within the set of multiple hand segments, where the set of multiple hand segments is a selected collection of hand segments. S T represents the number of zero-crossing points in the principal component signal of this limb segment whose derivative is greater than 0. i S The time interval between the zero-crossing point of the i-th derivative greater than 0 and the (i+1)-th derivative greater than 0 in the principal component signal of the limb segment angle; Tremor amplitude characteristics: root mean square, peak frequency, maximum peak value, and average peak value of principal component signals at different limb segments; Tremor variability characteristics: peak value coefficient of variation, instantaneous frequency coefficient of variation, and amplitude stability coefficient. The amplitude stability coefficient is calculated as follows: Where MSIS is the wrist amplitude stability coefficient of any segment in the set of multiple hand segments, M i S The maximum value between the zero-crossing point of the i-th derivative of the principal component signal of the limb segment angle and the zero-crossing point of the (i+1)-th derivative, iqr is the interquartile range function; The characteristics of the coordination relationship between limb segment tremors: the phase difference of the principal component signals of limb segment angles, wherein the phase difference of the principal component signals of limb segment angles is the phase difference between any two distinct principal component signals of limb segment angles in the set of multiple limb segments of the hand, and the calculation method is as follows: Where S1 and S2 are any two distinct limb segments in the set of multiple hand limb segments. The coherence spectrum is the principal component signal of angle S1 of the hand limb segment and the principal component signal of angle S2 of the hand limb segment. The autocorrelation spectrum is the principal component signal of the S1 angle of the hand limb segment. The autocorrelation spectrum is the principal component signal of the S2 angle of the hand limb segment. Let be the phase difference of the principal component signal of the angle between the hand limb segment S1 and the hand limb segment S2, where angle is a function taking the imaginary angle.

2. The wearable multi-segment joint analysis system for hand tremor as described in claim 1, characterized in that, The process of step 2 is as follows: Step 2-1: Limb segment attitude estimation. For the sensing signals of each limb segment, the quaternion attitude fusion algorithm is used to fuse the accelerometer, gyroscope and magnetometer signals to obtain the drift-free attitude estimation signal of each limb segment, and the attitude signal is converted into a three-dimensional Euler angle signal. Step 2-2: Denoising and dimensionality reduction of posture signals. For the Euler angle signals of each limb segment, the signal frequency is limited to the common frequency band of hand tremor by a denoising filter. Then, the three-dimensional Euler angle signal is reduced to a one-dimensional angle principal component signal by principal component analysis algorithm.

3. A wearable multi-segment combined analysis system for hand tremor as described in claim 1 or 2, characterized in that, In step 4, the feature set for the feature selector is obtained through the following steps: Step S1: Candidate feature set selection. Regularized regression algorithms such as lasso regression, ridge regression, or elastic net regression are used to initially screen the extracted 3D pattern features and traditional features to obtain the candidate feature set. Step S2: The selected feature set is determined by using the sequential floating feature selection algorithm to obtain the selected feature set from the candidate feature set.

4. A wearable multi-segment combined analysis system for hand tremor as described in claim 1 or 2, characterized in that, The intelligent classifier described above is constructed using a selected feature set, employing support vector machines, random forests, or artificial neural networks. The collected tremor data included resting tremor and postural tremor data in the Wing-beating posture; The regularized regression algorithm chooses elastic net regression to address the problem that the total number of features is greater than the number of samples.

5. A wearable multi-segment combined analysis system for hand tremor as described in claim 1, characterized in that, The sensing device includes N sensing units and one embedded processing module. The N sensing units collect tremor data from each segment of the hand. Each sensing unit integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The embedded processing module includes an embedded MCU and a wireless communication module, which is responsible for synchronizing the data from the N sensing units and wirelessly transmitting it to the data forwarding module. N is the number of elements in the set of multiple hand segments, which can be any integer from 2 to 6. Each hand segment is all the elements in the set of multiple hand segments.

6. A wearable multi-segment combined analysis system for hand tremor as described in claim 1 or 5, characterized in that, When the sensor is used to collect vibration data, the user should sit comfortably in a chair with armrests, wear the sensor, and perform the vibration test.

7. A wearable multi-segment combined analysis system for hand tremor as described in claim 1 or 5, characterized in that, The data processing module calculates the posture information of each limb segment through a posture fusion algorithm. After preprocessing the posture information, it extracts the tremor frequency, tremor amplitude, tremor variability, and synergistic relationship features between limb segment tremors. Then, it uses an intelligent classifier to determine whether the user is a Parkinson's disease patient or an essential tremor patient.

8. A wearable multi-segment combined analysis system for hand tremor as described in claim 1 or 5, characterized in that, The sensing unit uses a highly integrated nine-axis sensing chip. The candidate fixed positions of the sensing unit include the distal interphalangeal joints of the thumb, index finger, middle finger, ring finger and little finger, as well as the wrist. The embedded processing module is strapped to the wrist. The embedded processing module uses 6 SPI buses and register bit operation mode to ensure the synchronization of the sensing units; when the computing unit is deployed in the embedded processing module, the sensing device only sends the multi-limb joint features, Parkinson's disease and essential tremor classification results. When the computing unit is deployed on the host computer, it should output the analysis results to the user interface and interact with the database to store the user's basic population information, medical records, and previous tremor analysis results.

Images