Three-dimensional opposition movement-based parkinson's disease artificial intelligence classification method and system

Abstract

A three-dimensional finger-to-finger movement-based Parkinson's disease artificial intelligence classification method, including three-dimensional finger-to-finger segmentation, three-dimensional mode feature extraction and automatic diagnosis discrimination, based on the collected three-dimensional finger-to-finger movement, through three-dimensional finger-to-finger segmentation, the problem that the thumb and index finger do not reach the closing point and opening point at the same time in the finger-to-finger movement process is solved, and through three-dimensional mode feature extraction, the movement coordination, damaged inconsistency and modal energy entropy related features in the finger-to-finger movement process are extracted, and finally through automatic classification discrimination, whether the user has Parkinson's disease is comprehensively analyzed and discriminated by combining three-dimensional mode features and traditional features such as slowness, amplitude reduction and fatigue. And provide the system realized by the method. The present application does not need invasive detection, and can effectively assist in distinguishing Parkinson's disease patients and healthy people.

Description

Technical Field

[0001] This invention relates to the field of digital healthcare, specifically to an artificial intelligence classification method and system for Parkinson's disease based on three-dimensional finger-flicking movements. Background Technology

[0002] Parkinson's disease (Parkinson's disease) is a common neurodegenerative disorder with a prevalence of approximately 0.3% in the general population and about 1% in people over 60 years of age, making it the second most common neurodegenerative disease after Alzheimer's. Typical motor symptoms of Parkinson's disease include bradykinesia, rigidity, and tremor. Currently, Parkinson's screening primarily involves assessing motor symptoms, requiring patients to present with bradykinesia accompanied by resting tremor or rigidity. In clinical practice, the assessment of motor symptoms is mainly conducted by clinicians using Part III of the new World Movement Disorders Society Parkinson's Disease Comprehensive Assessment Scale (MDS-UPDRS), using a 5-point scale, with 0 representing normal and 4 representing the most severe condition. The assessment process is highly dependent on the physician's clinical experience, resulting in low consistency among physicians. Due to these factors, the misdiagnosis rate of Parkinson's disease is high; some literature reports a misdiagnosis rate as high as 47% in general practice hospitals at the primary care level.

[0003] In particular, finger antimagnetic movement is an important part of bradykinesia assessment, and in clinical practice, doctors mainly assess it based on traditional characteristics such as slowness, reduced amplitude, and fatigue. However, these traditional characteristics are easily affected by factors such as age and medication use. Therefore, the finger antimagnetic movement of healthy elderly people with age-related bradykinesia may be assessed as greater than 0 points, while the finger antimagnetic movement of patients with mild Parkinson's disease taking medication may be normal. This further complicates the screening of patients with mild Parkinson's disease in primary care hospitals. To address this issue, research on how to more accurately distinguish between patients with mild Parkinson's disease and healthy elderly people with age-related bradykinesia can effectively improve the accuracy of Parkinson's disease screening.

[0004] Representing finger-antipulation movements in three dimensions allows for the analysis of multi-joint motion patterns during these movements, supplementing traditional features. However, since the index finger and thumb do not always reach the closing and opening points simultaneously during finger-antipulation movements, segmentation of three-dimensional finger-antipulation movements is challenging. This makes it impossible to extract features related to the opening, closing, and fatigue processes from three-dimensional finger-antipulation movements, thus hindering the construction of an intelligent classifier for Parkinson's disease that combines traditional and three-dimensional features. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes an artificial intelligence classification method and system for Parkinson's disease based on three-dimensional antifinger movement. This method and system analyzes three-dimensional antifinger movement patterns, extracts motor coordination features, impairment inconsistency features, and modal energy entropy features, and uses three-dimensional pattern features to supplement the deficiencies of traditional features, improving robustness to factors such as bradykinesia. It also considers antifinger movement features such as bradykinesia, reduced amplitude, fatigue, motor coordination, impairment inconsistency, and modal energy entropy to construct an intelligent classifier for Parkinson's disease, which can make accurate and objective classifications of users. It has particular advantages in identifying patients with mild Parkinson's disease and can be used for large-scale population screening.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] An artificial intelligence classification method for Parkinson's disease based on three-dimensional finger-fingering motion includes the following steps:

[0008] Step 1: 3D finger segmentation, segmenting the input 3D finger motion signal to obtain the closing and opening points in the finger motion;

[0009] Step 2: 3D pattern feature extraction. Based on the segmented 3D finger motion, 3D pattern features are extracted. The 3D pattern features include motion coordination features, damage inconsistency features, and modal energy entropy features.

[0010] Step 3: Automatic classification and discrimination. Combining the input traditional features and 3D pattern features, the selected features are obtained through a feature selector. An intelligent classifier is used to complete the binary classification between Parkinson's disease patients and healthy people, and output the classification result data of whether the user is a Parkinson's disease patient or a healthy person.

[0011] Furthermore, in step 1, the three-dimensional finger-pairing motion is acquired by devices such as inertial sensors or cameras, and the signal is expressed in Euler angles as the three-dimensional rotational relationship between the thumb and index finger. During acquisition, the user is required to perform the finger-pairing motion in a comfortable sitting posture, placing their hand on a horizontal table, and keeping the thumb and index finger as close to the horizontal plane as possible during the motion.

[0012] Preferably, in step 1, the three-dimensional finger-like segmentation includes the following steps:

[0013] Step 1-1: Preliminary search for the closure and opening points. Using the index finger motion axis component in the three-dimensional finger motion signal, a peak search algorithm is used for the preliminary search of the closure and opening points. The calculation method for the minimum peak prominence and minimum interpeak interval of the peak search algorithm is as follows:

[0014] MPP=α×iqr(Angle F );

[0015]

[0016] Where MPP is the minimum peak prominence of the peak search algorithm, α is the ratio coefficient of minimum peak prominence to interquartile range, iqr is the interquartile range calculation function, and Angle... F Let f0 be the index finger motion axis component in the three-dimensional finger-pinch motion signal, MPD be the minimum interpeak interval of the peak search algorithm, β be the proportionality coefficient between the minimum interpeak interval and the main frequency, and f0 be the main frequency of the index finger motion axis component in the three-dimensional finger-pinch motion signal.

[0017] Step 1-2: Final selection of the closure and opening points. Search for local minima of the relative angular velocity magnitude. The local minima of the relative angular velocity magnitude near the initially searched closure and opening points are selected as the candidate point set P. The points that maximize the objective function are selected as the closure and opening points from this candidate set. The objective function for this selection process is:

[0018] argmax p∈P (-Norm(|ω(p)|)+φNorm(Angle F (p)|)

[0019] Where P is the candidate point set, ω is the magnitude of the relative angular velocity, Norm is the energy normalization function, and φ is the weight of the index finger angle in considering the selection of the closing and opening points.

[0020] Furthermore, in step 2, the three-dimensional pattern features are calculated as follows:

[0021] The motion coordination characteristics include: the number of local minima where the relative angular velocity modulus is less than the lower quartile, the number of local maxima and the maximum value of the interaxial coherence spectrum of the three-dimensional finger-like motion. The interaxial coherence spectrum of the three-dimensional finger-like motion includes the coherence spectrum between the X-axis and Y-axis components, the coherence spectrum between the X-axis and Z-axis components, and the coherence spectrum between the Y-axis and Z-axis components. The coherence spectrum between the X-axis and Z-axis components is calculated as follows:

[0022]

[0023] Among them, C XZ P represents the coherence spectrum between the X-axis and Z-axis components of the three-dimensional finger-like motion. XZ P represents the cross-power spectrum between the X-axis and Z-axis components of the three-dimensional finger-like motion. XX P is the autocorrelation spectrum of the X-axis component of the three-dimensional finger-like motion. ZZ The autocorrelation spectrum of the Z-axis component of the three-dimensional finger-like motion is shown.

[0024] The characteristics of the damage inconsistency include: the ratio of the root mean square of the relative angular velocity magnitude during the opening process to the root mean square of the relative angular velocity magnitude during the closing process; the opening process refers to the time period from the closing point to the next opening point; the closing process refers to the time period from the opening point to the next closing point.

[0025] The modal energy entropy characteristics include: modal energy entropy along each axis of relative angular velocity. The modal energy entropy along each axis of relative angular velocity includes the modal energy entropy of the X-axis component, the Y-axis component, and the Z-axis component. The modal energy entropy of the X-axis component is calculated as follows:

[0026]

[0027] Among them EN X V is the modal energy entropy of the X-axis component of the relative angular velocity. j X Let n be the i-th intrinsic mode function of the relative angular velocity X-axis component. The intrinsic mode function can be obtained using mode decomposition algorithms such as discrete mode decomposition and empirical mode decomposition, where n is the number of layers in the mode decomposition algorithm.

[0028] Furthermore, in step 3, the automatic classification and discrimination process includes the following modules:

[0029] The feature selector selects features based on the input feature set obtained during training.

[0030] The intelligent classifier categorizes users into Parkinson's disease patients or healthy individuals based on the classification results obtained from the trained classifier.

[0031] In step 3, the feature set for the feature selector is obtained through the following steps:

[0032] Step 3-1: 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.

[0033] Step 3-2: Determine the selected feature set. Use the sequential floating feature selection algorithm to obtain the selected feature set from the candidate feature set.

[0034] Preferably, the intelligent classifier is constructed using a selected feature set, employing a support vector machine, random forest, or artificial neural network.

[0035] Preferably, α and β are set to 0.5.

[0036] Preferably, φ is set to 0.6.

[0037] Preferably, the mode decomposition algorithm is selected as the diversification mode decomposition algorithm, and the number of layers n of the mode decomposition algorithm is set to 5.

[0038] Preferably, the intelligent classifier is constructed using the support vector machine algorithm to reduce the sample size required for training.

[0039] An artificial intelligence classification system for Parkinson's disease based on three-dimensional finger-flicking motion includes a sensing device, a data synchronization and forwarding device, and a host computer. The sensing device is used to collect acceleration, angular velocity, and magnetic force data of the thumb and index finger by attaching it to corresponding positions of the thumb and index finger, and sends the data to the data synchronization and forwarding device. The data synchronization and forwarding device controls the sensing device to perform synchronous data acquisition and forwards the collected data to the host computer. The host computer includes a data receiving module, a data processing module, and a user interface. 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 is used to send the analysis results of the data processing to the user interface for display.

[0040] Furthermore, the sensing device integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, and the sensing device is lightweight enough not to affect the user's finger movements.

[0041] Furthermore, the data processing module calculates the posture of the thumb and index finger respectively using a posture fusion algorithm, and calculates the three-dimensional finger-to-finger motion and relative angular velocity based on the posture of the thumb and index finger. Then, it extracts the traditional features and three-dimensional pattern features of the finger-to-finger motion, and uses an intelligent classifier to determine the classification result data of whether the user is a Parkinson's disease patient or a healthy person.

[0042] Furthermore, the user interface includes functions for adding, deleting, and modifying patient demographic information, and can display the waveform of three-dimensional finger-pointing movements in real time, as well as the analysis results of traditional features and three-dimensional pattern features in the patient's finger-pointing movements, and intelligent classification results.

[0043] Preferably, the sensing devices are fixed at the distal joints of the thumb and index finger, respectively.

[0044] Preferably, the sensing device uses an MPU9250 nine-axis sensing chip design to ensure that the sensing device is lightweight.

[0045] Preferably, the host computer includes a database for storing users' population information and previous finger-matching analysis results, which facilitates doctors' query and comparison.

[0046] The beneficial effects of this invention are mainly reflected in:

[0047] 1. This invention provides an artificial intelligence classification method and system for Parkinson's disease based on three-dimensional finger-flicking motion. It can supplement traditional features with three-dimensional pattern features, reduce the impact of age and medication on classification results, and improve the accuracy of identifying Parkinson's disease patients.

[0048] 2. This invention proposes a three-dimensional finger-pointing motion segmentation method to achieve three-dimensional finger-pointing motion segmentation.

[0049] 3. This invention extracts three three-dimensional pattern features for three-dimensional antiphagonic movement: motor coordination features, impairment inconsistency features, and modal energy entropy features. Among these, motor coordination features and modal energy entropy features are more sensitive to motor function impairment in patients with mild Parkinson's disease, while impairment inconsistency can reflect the differences in the degree of motor function impairment in patients with mild Parkinson's disease. Combining the three-dimensional pattern features with traditional features can yield a more comprehensive feature set reflecting antiphagonic movement function.

[0050] 4. This invention combines three-dimensional pattern features and traditional features for automatic classification and discrimination of Parkinson's disease. An intelligent classifier is trained using multi-directional antidigital motor function features, enabling binary classification between Parkinson's disease patients and healthy individuals, with particular advantages in identifying patients with mild Parkinson's disease. Attached Figure Description

[0051] Figure 1 This is a flowchart of the artificial intelligence classification method for Parkinson's disease based on three-dimensional finger-fingering motion, as described in this invention.

[0052] Figure 2 This is a schematic diagram of the structure of the artificial intelligence classification system for Parkinson's disease based on three-dimensional finger movements, as proposed in this invention. Detailed Implementation

[0053] The present invention will now be further described with reference to the accompanying drawings.

[0054] Example 1

[0055] Reference Figure 1 An artificial intelligence classification method for Parkinson's disease based on three-dimensional finger-sign movement includes three-dimensional finger-sign segmentation, three-dimensional pattern feature extraction, and automatic classification and discrimination, comprising the following steps:

[0056] Step 1: 3D finger-linked segmentation. 3D finger-linked segmentation includes the following steps:

[0057] Step 1-1: Preliminary search for the closure and opening points. Using the index finger motion axis component in the three-dimensional finger motion signal, a peak search algorithm is used for the preliminary search of the closure and opening points. The calculation method for the minimum peak prominence and minimum interpeak interval of the peak search algorithm is as follows:

[0058] MPP=α×iqr(Angle F );

[0059]

[0060] Where MPP is the minimum peak prominence of the peak search algorithm, α is the ratio coefficient of minimum peak prominence to interquartile range, iqr is the interquartile range calculation function, and Angle... F Let α be the index finger motion axis component in the three-dimensional finger-pinch motion signal, MPD be the minimum peak-to-peak interval of the peak search algorithm, β be the proportionality coefficient between the minimum peak-to-peak interval and the dominant frequency, and f0 be the dominant frequency of the index finger motion axis component in the three-dimensional finger-pinch motion signal. Preferably, α and β are set to 0.5.

[0061] Step 1-2: Final selection of the closure and opening points. Search for local minima of the relative angular velocity magnitude. The local minima of the relative angular velocity magnitude near the initially searched closure and opening points are used as a candidate point set. The points that maximize the objective function are selected from this candidate set as the chosen closure and opening points. The objective function for this selection process is:

[0062] argmax p∈P (-Norm(|ω(p)|)+φNorm(Angle F (p)|)

[0063] Where P is the candidate point set, ω is the magnitude of the relative angular velocity, Norm is the energy normalization function, and φ is the weight of the index finger angle in considering the selection of the closure and opening points. Preferably, φ is set to 0.6.

[0064] Step 2: 3D pattern feature extraction. Based on the segmented 3D finger motion and relative angular velocity, extract motion coordination, damage inconsistency and modal energy entropy features;

[0065] Motion coordination characteristics include: the number of local minima where the relative angular velocity modulus is less than the lower quartile; the number and maximum of local maxima in the interaxial coherence spectrum of the three-dimensional finger-like motion. The interaxial coherence spectrum of the three-dimensional finger-like motion includes the coherence spectrum between the X-axis and Y-axis components, the coherence spectrum between the X-axis and Z-axis components, and the coherence spectrum between the Y-axis and Z-axis components. Taking the coherence spectrum between the X-axis and Z-axis components as an example, the calculation method is as follows:

[0066]

[0067] Among them, C XZ P represents the coherence spectrum between the X-axis and Z-axis components of the three-dimensional finger-like motion. XZP represents the cross-power spectrum between the X-axis and Z-axis components of the three-dimensional finger-like motion. XX P is the autocorrelation spectrum of the X-axis component of the three-dimensional finger-like motion. ZZ The autocorrelation spectrum of the Z-axis component of the three-dimensional finger-like motion is shown.

[0068] The characteristics of the damage inconsistency include: the ratio of the root mean square of the relative angular velocity magnitude during the opening process to the root mean square of the relative angular velocity magnitude during the closing process; the opening process refers to the time period from the closing point to the next opening point; the closing process refers to the time period from the opening point to the next closing point.

[0069] The modal energy entropy characteristics include: modal energy entropy along each axis of relative angular velocity. The modal energy entropy along each axis of relative angular velocity includes the modal energy entropy of the X-axis component, the Y-axis component, and the Z-axis component. Taking the modal energy entropy of the X-axis component as an example, the calculation method is as follows:

[0070]

[0071] Among them EN X V is the modal energy entropy of the X-axis component of the relative angular velocity. j X Let be the i-th intrinsic mode function of the relative angular velocity X-axis component. The intrinsic mode function can be obtained using mode decomposition algorithms such as discrete mode decomposition and empirical mode decomposition. n is the number of layers in the mode decomposition algorithm. Preferably, the discrete mode decomposition algorithm is selected, and the number of layers n in the mode decomposition algorithm is set to 5.

[0072] Step 3: Automatic classification and discrimination. The automatic classification and discrimination process includes the following modules:

[0073] The feature selector selects features based on the input feature set obtained during training.

[0074] The intelligent classifier categorizes users into Parkinson's disease patients or healthy individuals based on the classification results obtained from the trained classifier.

[0075] The feature set to be selected by the feature selector is obtained by the following steps:

[0076] Step 3-1: 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.

[0077] Step 3-2: Determine the selected feature set. Use the sequential floating feature selection algorithm to obtain the selected feature set from the candidate feature set.

[0078] The intelligent classifier is constructed using a selected feature set, employing a support vector machine, random forest, or artificial neural network. Preferably, the intelligent classifier is constructed using a support vector machine algorithm to reduce the sample size required for training.

[0079] Example 2

[0080] Reference Figure 1 and Figure 2 An artificial intelligence classification system for Parkinson's disease based on three-dimensional finger-flicking motion is disclosed, comprising a sensing device, a data synchronization and forwarding device, and a host computer. The sensing device is used to collect acceleration, angular velocity, and magnetic force data of the thumb and index finger by attaching it to corresponding positions on the thumb and index finger, and then sends the data to the data synchronization and forwarding device. The data synchronization and forwarding device controls the sensing device to perform synchronous data acquisition and forwards the collected data to the host computer. The host computer includes a data receiving module, a data processing module, and a user interface. 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 sends the analysis results of the data processing to the user interface for display.

[0081] The aforementioned sensing device integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, and is sufficiently lightweight to not interfere with the user's finger movements. Preferably, the sensor uses an MPU9250 chip to improve integration and make the sensing device sufficiently lightweight.

[0082] The data processing module calculates the posture of the thumb and index finger respectively through a posture fusion algorithm, and calculates the three-dimensional finger-to-finger motion and relative angular velocity based on the posture of the thumb and index finger. Then, it extracts the traditional features and three-dimensional pattern features of the finger-to-finger motion, and uses an intelligent classifier to determine whether the user is a Parkinson's disease patient or a healthy person. The data processing process is described in Example 1.

[0083] The user interface includes functions for adding, deleting, and modifying patient information, and can display the waveform of three-dimensional finger-pointing movements in real time, as well as the analysis results of traditional features and three-dimensional pattern features in the patient's finger-pointing movements, and intelligent classification results.

[0084] Preferably, the data synchronization and forwarding device uses two I2C buses to ensure synchronization of the sensors at the thumb and index finger through register bit operation.

[0085] Based on the above method and system, this invention was validated on a self-collected dataset including 49 Parkinson's disease patients and 29 healthy controls. In the self-collected dataset, the age and sex of the Parkinson's disease patients and healthy controls were matched. The validation results for right-hand finger opposition are shown in Table 1.

[0086] Features used Five-fold cross-means accuracy (%) Traditional features only 86.1 Traditional features + publicly available 3D model features 87.5 Feature set of the present invention 95.8

[0087] Table 1

[0088] Specifically, this invention further validated its accuracy in distinguishing between patients with mild Parkinson's disease and healthy individuals with age-related slowing in a subset of data containing only patients with mild Parkinson's disease. This subset of data included 11 age- and sex-matched patients with mild Parkinson's disease and 28 healthy controls, with both patients and healthy controls having an antiphagia score of 0 or 1. The validation results for right-hand antiphagia are shown in Table 2.

[0089] Features used Five-fold cross-means accuracy (%) Traditional features only 73.3 Traditional features + publicly available 3D model features 84.6 Feature set of the present invention 97.4

[0090] Table 2

[0091] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[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. An artificial intelligence classification system for Parkinson's disease based on three-dimensional finger-signaling, characterized in that, The system includes a sensing device, a data synchronization and forwarding device, and a host computer. The sensing device is used to collect acceleration, angular velocity, and magnetic force data of the thumb and index finger by attaching it to corresponding positions on the thumb and index finger, and sends it to the data synchronization and forwarding device. The data synchronization and forwarding device controls the sensing device to collect data synchronously and forwards the collected data to the host computer. The host computer includes a data receiving module, a data processing module, and a user interface. 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 is used to send the analysis results of the data processing to the user interface for display. The Parkinson's disease artificial intelligence classification method based on three-dimensional finger-counting motion implemented by the three-dimensional finger-counting motion artificial intelligence classification system includes the following steps: Step 1: 3D finger segmentation, segmenting the input 3D finger motion signal to obtain the closing and opening points in the finger motion; Step 2: 3D pattern feature extraction. Based on the segmented 3D finger motion, 3D pattern features are extracted. The 3D pattern features include motion coordination features, damage inconsistency features, and modal energy entropy features. Step 3: Automatic classification and discrimination. Combining the input traditional features and 3D pattern features, the selected features are obtained through a feature selector. An intelligent classifier is used to complete the binary classification between Parkinson's disease patients and healthy people, and output the classification result data of whether the user is a Parkinson's disease patient or a healthy person.

2. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 1, characterized in that, In step 1, the three-dimensional finger-pairing motion is acquired by an inertial sensor or camera, and the signal is expressed in Euler angles as the three-dimensional rotational relationship between the thumb and index finger. During acquisition, the user is required to perform the finger-pairing motion in a comfortable sitting position. During the finger-pairing motion, the user needs to place their hand on a horizontal table and make the thumb and index finger open and close in the horizontal plane during the motion.

3. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 1 or 2, characterized in that, In step 1, the three-dimensional finger-like segmentation includes the following steps: Step 1-1: Preliminary search for the closure and opening points. Using the index finger motion axis component in the three-dimensional finger motion signal, a peak search algorithm is used for the preliminary search of the closure and opening points. The calculation method for the minimum peak prominence and minimum interpeak interval of the peak search algorithm is as follows: ； ； in The minimum peak prominence of the peak search algorithm is given. It is the ratio of minimum peak prominence to interquartile range. This is the interquartile range calculation function. The index finger motion axis component in the aforementioned three-dimensional finger-pointing motion signal. The minimum interpeak interval of the peak search algorithm described above. It is the ratio of the minimum peak interval to the dominant frequency. The dominant frequency of the index finger motion axis component in the aforementioned three-dimensional finger-to-finger motion signal; Steps 1-2: Final selection of the closure and opening points; searching for local minima of the relative angular velocity magnitude; and using the local minima of the relative angular velocity magnitude near the initially searched closure and opening points as the candidate point set. The points that maximize the objective function are selected from the candidate point set as the selected closure and opening points. The objective function for the selection process is: ； in For the candidate point set, The magnitude of the relative angular velocity is given. Let be the energy normalization function. The weight of the index finger angle in the selection of the closure and opening points.

4. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 1 or 2, characterized in that, In step 2, the three-dimensional pattern features are calculated as follows: The motion coordination characteristics include: the number of local minima where the relative angular velocity modulus is less than the lower quartile, the number of local maxima and the maximum value of the interaxial coherence spectrum of the three-dimensional finger-like motion. The interaxial coherence spectrum of the three-dimensional finger-like motion includes the coherence spectrum between the X-axis and Y-axis components, the coherence spectrum between the X-axis and Z-axis components, and the coherence spectrum between the Y-axis and Z-axis components. The coherence spectrum between the X-axis and Z-axis components is calculated as follows: ； in, This refers to the coherence spectrum between the X-axis and Z-axis components of the three-dimensional finger-like motion. This refers to the cross-power spectrum between the X-axis and Z-axis components of the three-dimensional finger-like motion. The autocorrelation spectrum of the X-axis component of the three-dimensional finger-like motion is given. The autocorrelation spectrum of the Z-axis component of the three-dimensional finger-like motion is shown. The characteristics of the damage inconsistency include: the ratio of the root mean square of the relative angular velocity magnitude during the opening process to the root mean square of the relative angular velocity magnitude during the closing process; the opening process refers to the time period from the closing point to the next opening point; the closing process refers to the time period from the opening point to the next closing point. The modal energy entropy characteristics include: modal energy entropy along each axis of relative angular velocity. The modal energy entropy along each axis of relative angular velocity includes the modal energy entropy of the X-axis component, the Y-axis component, and the Z-axis component. The modal energy entropy of the X-axis component is calculated as follows: ； in The modal energy entropy of the X-axis component of the relative angular velocity is given. The first of the relative angular velocity X-axis components One eigenmode function The first of the relative angular velocity X-axis components The intrinsic mode functions are obtained using diversifying mode decomposition and empirical mode decomposition algorithms. This represents the number of layers in the mode decomposition algorithm.

5. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 1 or 2, characterized in that, In step 3, the automatic classification and discrimination process includes the following modules: The feature selector selects features based on the input feature set obtained through training. The intelligent classifier categorizes users into Parkinson's disease patients or healthy individuals based on the classification results obtained from the trained classifier.

6. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 5, characterized in that, In step 3, the feature set for the feature selector is obtained through the following steps: Step 3-1: Candidate feature set selection. Lasso regression, ridge regression or elastic net regression regularized regression algorithm is used to preliminarily screen the extracted 3D pattern features and traditional features to obtain the candidate feature set; Step 3-2: Determine the selected feature set. Use the sequential floating feature selection algorithm to obtain the selected feature set from the candidate feature set.

7. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 5, characterized in that, The intelligent classifier described herein is constructed using a selected feature set, employing support vector machines, random forests, or artificial neural networks.

8. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 1, characterized in that, The aforementioned sensing device integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, and the sensing device does not interfere with the user's finger movements.

9. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 1 or 8, characterized in that, The data processing module calculates the posture of the thumb and index finger separately using a posture fusion algorithm, and calculates the three-dimensional finger-to-finger motion and relative angular velocity based on the posture of the thumb and index finger. Then, it extracts the traditional features and three-dimensional pattern features of the finger-to-finger motion, and uses an intelligent classifier to determine whether the user is a Parkinson's disease patient or a healthy person.

10. The Parkinson's disease artificial intelligence classification system based on three-dimensional finger-finger movements as described in claim 1 or 8, characterized in that, The user interface includes functions for adding, deleting, and modifying patient information, and can display the waveform of three-dimensional finger-pointing movements in real time, as well as the analysis results of traditional features and three-dimensional pattern features in the patient's finger-pointing movements, and intelligent classification results.

Images