Parkinson's subtype data clustering methods, devices, systems, and storage media
By constructing core skeleton points using a hybrid t-nearest neighbor graph and a Gaussian mixture model, and combining this with a reverse t-nearest neighbor graph to separate water basins, the problem of high-dimensional nonlinear complex structure in Parkinson's disease subtype data was solved. This approach enabled efficient and high-quality data clustering and generated verifiable subtype labels.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAIHUA UNIV
- Filing Date
- 2026-02-26
- Publication Date
- 2026-06-02
AI Technical Summary
Existing technologies struggle to effectively handle the high-dimensional, nonlinear, and complex structure of Parkinson's disease subtype data. Traditional clustering methods suffer from high computational complexity, strong parameter sensitivity, and fail to achieve efficient and high-quality subtype data clustering.
A hybrid t-nearest neighbor graph is used to construct core skeleton points and hub points. A Gaussian mixture model and a reverse t-nearest neighbor graph are combined to separate water collection basins. Water collection basins are merged through subclass connections and similarity to achieve high-dimensional data clustering.
This improved the efficiency and quality of clustering high-dimensional Parkinson's disease subtype data, generated subtype labels that could be verified with clinical indicators, and enhanced the clinical significance of the clustering.
Smart Images
Figure CN122132865A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of information processing technology, specifically relating to a method, apparatus, system, and storage medium for clustering Parkinson's subtype data. Background Technology
[0002] Parkinson's disease, a highly heterogeneous neurodegenerative disease, relies primarily on subjective assessments of motor symptoms in traditional clinical classification, making it difficult to capture the disease's inherent biological heterogeneity. With the development of digital healthcare technologies, wearable sensors, voice analysis platforms, and digital motion tasks enable high-precision, continuous quantification of patients' tremor and non-tremor motor characteristics, providing a new dimension for data-driven disease classification. Tremor characteristics typically include frequency (4-6 Hz resting tremor), amplitude, rhythmicity (acquired via accelerometers and surface electromyography), and perturbation indices in the speech signal. Non-tremor characteristics encompass multiple dimensions such as bradykinesia, postural instability, and gait disturbances, including decreasing speed and amplitude of finger tapping, coefficient of variation of gait spatiotemporal parameters, and body sway trajectories in balance posture diagrams. These multi-dimensional time-series data constitute a high-dimensional feature space, whose complex structure is difficult to effectively analyze using traditional linear classification methods.
[0003] Because Parkinson's disease is not a single disease but a syndrome composed of different subtypes, traditional clinical classifications (such as tremor-dominant / postural instability-gait disorder) are rather subjective and crude. Furthermore, an excessive number of features can easily lead to the "curse of dimensionality" challenge of high-dimensional data. Therefore, pre-specifying the number of subtypes can easily cause subjective bias because the exact number of Parkinson's disease subtypes is unknown. Thus, given that methods such as K-means and density peak clustering and their variants require pre-specifying the number of subtype categories, and that methods like DBSCAN and its variants do not fully consider the high-dimensionality and complex spatial structure of Parkinson's subtype feature data, how can we effectively identify naturally formed patient subgroups within the digital feature data of Parkinson's disease—a data space where tremor and non-tremor features exhibit high-dimensional and non-linear distributions with complex correlations? In other words, how to efficiently and effectively cluster high-dimensional, structurally complex Parkinson's disease subtype data, and how to generate subtype labels that can be correlated and validated with clinical indicators to enhance the clinical significance of clustered subtype data, has become a crucial technical problem that needs to be solved in unsupervised high-dimensional data clustering for data mining and analysis in the field of Parkinson's disease.
[0004] Currently, methods for clustering high-dimensional data mainly include dimensionality reduction-based clustering, density clustering, spectral clustering, graph clustering, tensor clustering, and subspace clustering. For example, dimensionality reduction-based clustering methods, represented by PCA+K-means and t-SNE+DBSCAN, use the core idea of "preserving key features while compressing dimensionality, reducing dimensionality before clustering" to discover patterns in high-dimensional data. They are computationally efficient and visually appealing; however, they may suffer information loss when processing nonlinear data, and the separation of the two stages may lead to suboptimal solutions. Spectral clustering, based on graph cutting, constructs a Laplacian matrix feature space based on similarity graphs to perform data clustering, capable of handling manifold data. However, it suffers from sensitivity in similarity matrix construction and high storage overhead. Density clustering discovers data clusters of arbitrary shapes based on density distribution and density connectivity, without requiring a specified number of data classes, and can handle variable density and noise. However, the overly simplistic density definition leads to parameter sensitivity, and distance metrics degenerate and are prone to failure in high-dimensional environments. Graph clustering methods, based on graph nodes and community detection algorithms, can efficiently handle sparse high-dimensional data and do not require a pre-defined number of clusters. However, the clustering quality depends on the precise edge weights and struggles to handle overlapping communities. Subspace clustering assumes that data patterns are distributed in different low-dimensional subspaces and can discover hidden subspace structures through sparse representation or matrix factorization. It can directly solve the curse of dimensionality, capture local linear structures, and resist noise interference. However, it suffers from high computational complexity, reliance on pre-specified subspace dimensions, and the potential for vectorization to destroy image spatial structure. Therefore, how to achieve high-efficiency and high-quality high-dimensional data clustering has become a crucial technical problem that urgently needs to be solved in the field of high-dimensional data clustering mining and analysis. Summary of the Invention
[0005] To address the problems existing in the prior art, this invention provides a Parkinson's subtype data clustering method, apparatus, system, and storage medium. It overcomes the shortcomings and limitations of methods such as k-means and density peak clustering with a pre-specified number of subtypes and their variants, as well as existing density clustering methods. These limitations include difficulty in handling nonlinear complex structures, high time and space complexity, and the lack of a method that can combine the strengths and weaknesses of both methods. Furthermore, it effectively solves the problems of poor efficiency and quality in high-dimensional data clustering.
[0006] To achieve the above objectives, the present invention provides the following solution: A clustering method for Parkinson's subtype data includes: Step S100: Standardize and preprocess the Parkinson's data, and calculate the mixed t-nearest neighbor graph of the data space; Step S200: Calculate the core skeleton points and pivot points based on the hybrid t-nearest neighbor graph and divergence; Step S300: Calculate the naive catch basin based on the Gaussian mixture model, the pivot point set, and the skeleton core points; Step S400: Calculate the mergeable points of the water collection basins, aggregate the water collection basins using subclass connection and subclass similarity, and obtain high-dimensional data clustering results.
[0007] Preferably, in step S100, vector normalization preprocessing is performed on the Parkinson's dataset with tremor features and without tremor features. At the same time, a mixed t-nearest neighbor graph is calculated based on the dataset and parameter t. The mixed t-nearest neighbor graph includes: the t-nearest neighbor graph and the reverse t-nearest neighbor graph of the data points, as well as the height.
[0008] Preferably, in step S300, the optimized water collection basin and the abnormal water collection basin are separated using the reverse t-nearest neighbor graph.
[0009] Preferably, in step S400, the mergeable connections between adjacent water collection basins are calculated, and the water collection basins are optimized by merging water collection basin-level similarity and subclass connections. At the same time, the remaining abnormal water collection basins are aggregated to obtain the clustering results of Parkinson's subtype data.
[0010] The present invention also provides a Parkinson's subtype data clustering device, comprising: The first processing module is used for the standardization preprocessing of Parkinson's data and to calculate the mixed t-nearest neighbor graph of the data space. The second processing module is used to calculate the core skeleton points and pivot points based on the hybrid t-nearest neighbor graph and divergence; The third processing module is used to calculate the naive catch basin based on the Gaussian mixture model, the hub point set, and the skeleton core points. The fourth processing module is used to calculate the mergeable points of the water collection basins, aggregate the water collection basins using subclass connections and subclass similarity, and obtain high-dimensional data clustering results.
[0011] Preferably, the first processing module performs vector normalization preprocessing on the Parkinson's dataset with and without tremor features, and calculates a hybrid t-nearest neighbor graph based on the dataset and parameter t; wherein, the hybrid t-nearest neighbor graph includes: the t-nearest neighbor graph and the reverse t-nearest neighbor graph of the data points, as well as the height.
[0012] As a preferred option, the third processing module uses a reverse t-nearest neighbor graph to separate the optimized water collection basin and the abnormal water collection basin.
[0013] As a preferred option, the fourth processing module calculates the mergeable connections between adjacent water collection basins, optimizes the water collection basins by merging water collection basin-level similarity and subclass connections, and aggregates the remaining abnormal water collection basins to obtain the clustering results of Parkinson's subtype data.
[0014] The present invention also provides a Parkinson's subtype data clustering system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a Parkinson's subtype data clustering method when executed by the processor.
[0015] The present invention also provides a storage medium storing a computer program that executes a Parkinson's subtype data clustering method when running.
[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention addresses the problem of Parkinson's disease subtype clustering. First, it fully leverages the core skeleton point method based on the mixture t-nearest neighbor graph, constructing naive catch basins based on pivot points and Gaussian mixture models. Then, it uses the reverse t-nearest neighbor graph to separate optimized and anomalous catch basins. Finally, it obtains the subtype data clustering results using basin-level merging points and basin-level similarity. This invention addresses the shortcomings and limitations of methods such as k-means and density peak clustering, their variants, and existing clustering methods of different densities, including difficulty in handling nonlinear complex structures and high time and space complexity. Furthermore, no method has yet been able to combine the strengths of both methods, and there are currently no methods or systems that utilize unsupervised clustering to mine Parkinson's disease subtype data. This invention effectively improves the efficiency and quality of clustering methods on high-dimensional subtype data. Attached Figure Description
[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of the Parkinson's subtype data clustering method according to an embodiment of the present invention; Figure 2 To achieve hybridization according to embodiments of the present invention t Flowchart of the nearest neighbor graph; Figure 3 This is a flowchart illustrating the calculation of the core skeleton points and pivot point set according to an embodiment of the present invention; Figure 4 This is a flowchart illustrating the calculation of the water collection basin according to an embodiment of the present invention; Figure 5 This is a flowchart illustrating the merging of water collection basins according to an embodiment of the present invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0021] Example 1 like Figure 1 As shown, this invention provides a Parkinson's subtype data clustering method to address the problems of unpredicted number of clusters, complex data spatial structure, and the curse of dimensionality in Parkinson's subtype clustering, including: Step S100: Standardize and preprocess the Parkinson's dataset based on pre-specified parameters. t Calculate the mixed t-nearest neighbor graph of dataset D, mainly including data points. t's nearest neighbor Reverse t nearest neighbor The first attribute of a node is its height.
[0022] In this embodiment, the purpose of step S100 is to calculate the t-nearest neighbors, height, core skeleton points, and hub points of each data point in the Parkinson's dataset D. Specifically, based on the input dataset to be clustered, the calculation is performed for each data point. The t-nearest neighbors, reverse t-nearest neighbors, and height are used to obtain a mixed t-nearest neighbor graph.
[0023] In one exemplary embodiment, please refer to Figure 2 To obtain a mixture t The flowchart of the nearest neighbor graph, step S100 specifically includes steps S101 to S103, which are described in detail below: Step S101: Perform standardization preprocessing on the input dataset, and then calculate the standardization for each sample point. of t Nearby, Reverse t Nearest neighbors are used to obtain the mixed edges between nodes in the mixed nearest neighbor graph.
[0024] Step S102: Calculate the in-degree coefficient of each node based on the mixed edge and the shared t-nearest neighbor, wherein the in-degree coefficient... The calculation formula is as follows: ; In the formula, Let represent any single data point, sample, or node in the mixed nearest neighbor graph of the Parkinson dataset D. Indicates exponentiation. Representing data points In-degree in the t-nearest neighbor graph, Representing data points t's nearest neighbor.
[0025] Step S103: Calculate data points based on the mixed t-nearest neighbor and average shared distance. The height is calculated using the following formula: ; In the formula, Let represent any single data point, sample, or node in the Parkinson's dataset D, or a mixed nearest neighbor graph, where t represents the number of nearest neighbors. Representing data points t's nearest neighbor, Representing data points height, Representing data points Its neighbors distance, Representing data points The in-degree coefficient of the nearest neighbor of t. Representing data points In-degree in the reverse t-nearest neighbor graph.
[0026] Step S200: Calculate the core skeleton points and pivot points based on the hybrid t-nearest neighbor graph and divergence.
[0027] In this embodiment, the core skeleton point factor and pivot factor are calculated based on the mixed t-nearest neighbor and divergence to obtain the core skeleton point set and pivot point set.
[0028] In one exemplary embodiment, please refer to Figure 3 The flowchart for calculating the core skeleton point set and the hub point set is as follows. Step S200 includes steps S201 to S204, which are described in detail below.
[0029] Step S201: Based on the mixed t-nearest neighbor graph, calculate the in-degree and divergence of each data point. The formula for calculating divergence is as follows: ; In the formula, Let represent any single data point, sample, or node in the Parkinson's dataset D, or a mixed nearest neighbor graph, where t represents the number of nearest neighbors. Representing data points t's nearest neighbor, Representing data points divergence, Represents data point pairs The Euclidean distance between them.
[0030] Step S202: Based on data points The median of the height of the nearest neighbor of t and variance Calculate the single-nature core skeleton point factor values, which include the forward core skeleton point factor values: and backward core skeleton point factor values: The calculation formulas for both are as follows: ; In the formula, and Let t represent any two distinct data points or samples in the Parkinson's dataset D, and let t represent the number of nearest neighbors. Representing data points t's nearest neighbor, This represents median operations. Represents the calculated data points The height of each of the nearest neighbors of t. This represents variance calculation. This represents absolute value operations, with True and False representing logical true and logical false values, respectively.
[0031] Then, based on the reverse t-nearest neighbor and single-core skeleton point factor values, the data points are calculated. Core skeleton point factor values: The formula is as follows: ; In the formula, and Let t represent any two distinct data points or samples in the Parkinson's dataset D, and let t represent the number of nearest neighbors. Representing data points t's nearest neighbor, Representing data points t's nearest neighbor The single-sex core skeleton point factor value, It is an indicator function. The range of values for is [0,1]. Represents data point pairs Whether a reverse t-nearest neighbor relationship is satisfied is an index function; 1 indicates that the relationship is satisfied, and 0 indicates otherwise. Represents data point pairs At least one forward core skeleton point factor value is satisfied: and backward core skeleton point factor values: True and False represent logical true and logical false values, respectively.
[0032] Step S203: Based on data points Its local neighborhood t-nearest neighbor height, and mixed t-nearest neighbor, are calculated. The in-degree and divergence, where the divergence is... The calculation formula is as follows: ; In the formula, Let represent any single data point, sample, or node in the Parkinson's dataset D, or a mixed nearest neighbor graph, where t represents the number of nearest neighbors. Representing data points t's nearest neighbor, Represents data point pairs The Euclidean distance between them.
[0033] Next, using the divergence and the reverse t-nearest neighbor, the pivot factor is calculated using the following formula: ; In the formula, Let represent any single data point, sample, or node in the Parkinson's dataset D, or a mixed nearest neighbor graph, where t represents the number of nearest neighbors. Representing data points t's nearest neighbor, Representing data points Appearing at other points t The number of times in the nearest neighbor list, Representing data points The reverse t nearest neighbor, Represents the modulo operation of a set.
[0034] Based on the predefined domain value If data points The pivot factor is greater than Furthermore, within the local core skeleton point region, data points The pivot factor satisfies the following conditions: but It is a hub point. Represents the calculated data points The hub factor of each member of the t nearest neighbors.
[0035] Step 204: Based on the core skeleton point factor values and pivotal factors and threshold This allows us to obtain the core skeleton point set and the hub point set.
[0036] Step S300: Based on the Gaussian mixture model and the set of pivot points, as well as the core points of the skeleton, calculate the naive water collection basin, and then use the reverse t-nearest neighbor graph to separate the optimized water collection basin and the abnormal water collection basin.
[0037] In this embodiment, the t-nearest neighbor height and Gaussian mixture model, as well as the naive water collection basin of the hub point set, are used. Then, the optimized water collection basin and the abnormal water collection basin are separated from the naive water collection basin by using the reverse t-nearest neighbor.
[0038] In one exemplary embodiment, please refer to Figure 4 The flowchart for calculating the optimized water collection basin and the abnormal water collection basin is provided. Step S300 includes steps S301 to S302, which are described in detail below.
[0039] Step S301: Using the core skeleton point set and pivot point set, along with a Gaussian mixture model, perform preliminary clustering of the high-dimensional dataset to obtain the naive water basin. The calculation formula for the Gaussian mixture model is as follows: ; In the formula, Let H represent any single data point, sample, or node in the Parkinson's dataset D, and let H represent the modulus of the intersection of the core skeleton point and the pivot point, i.e., the number of elements. This represents the k-th Gaussian distribution. This represents the weight of the k-th Gaussian distribution, where the sum of all weights is 1. and Let represent the mean and variance of the k-th Gaussian model in the Gaussian mixture model.
[0040] Step S302: Based on the t-nearest neighbor graph and the reverse t-nearest neighbor graph, separate the optimized water collection basin and the abnormal water collection basin contained in the naive water collection basin to avoid the aggregation of members of different classes and subclasses of different classes.
[0041] Step S400: Calculate the mergeable points of the water collection basins, aggregate the water collection basins using subclass connections and subclass similarity, and obtain high-dimensional data clustering results.
[0042] In one exemplary embodiment, please refer to Figure 5 The flowchart for obtaining high-dimensional data clustering by aggregating water collection basins is as follows. Step S400 includes steps S401 to S403, which are described in detail below.
[0043] Step S401: Based on the mixed t-nearest neighbor graph, optimize the collection basins and abnormal collection basins, calculate the connection points between collection basins, and then obtain the points where collection basins can be merged. The calculation formula is as follows: ; In the formula, Let represent any single data point, sample, or node in the Parkinson's dataset D, or a mixed nearest neighbor graph, where t represents the number of nearest neighbors. Representing data points t's nearest neighbor, Representing data points All core skeletal points of the water collection basin, express The basin point of the water collection basin, Indicates calculation The t-nearest neighbors of the reverse t-nearest neighbor graph of the corresponding catchment basin. Indicates height calculation, express If the height satisfies the relaxed 2σ Laida's rule (without using absolute values) of the height sets of all its reverse-nearest neighbor members in the catch basins, return a logical true value: True or a logical false value: False. This represents median arithmetic. If the core framework point of the water collection basin... satisfy If a state can be merged, it is true; otherwise, it is false.
[0044] Step 402: Calculate the similarity at the catchment level The calculation formula is as follows: ; In the formula, and Let represent any two distinct data points or samples in the Parkinson's dataset D, or any two distinct nodes in the mixed nearest neighbor graph, where t represents the number of nearest neighbors. Representing data points t's nearest neighbor, Representing data points and Similarity between the respective catch basins; and This represents the 2σ Laida's rule for relaxation without using absolute values. This represents the mean operation. This represents variance calculation. This indicates height value calculation. Representing data points In the reverse t-nearest neighbor graph G, True represents logical truth and False represents logical falsehood.
[0045] Then, the merging of adjacent optimized water collection basins is performed based on water collection basin connectivity (node pairs) and water collection basin level similarity.
[0046] Step 403: Prioritize the reverse t-nearest neighbor rule, and then use the t-nearest neighbor rule to aggregate abnormal water collection basins into adjacent optimized water collection basins to obtain high-dimensional data clustering.
[0047] Example 2 The present invention also provides a Parkinson's subtype data clustering device, comprising: The first processing module is used for the standardization preprocessing of Parkinson's data and to calculate the mixed t-nearest neighbor graph of the data space. The second processing module is used to calculate the core skeleton points and pivot points based on the hybrid t-nearest neighbor graph and divergence; The third processing module is used to calculate the naive catch basin based on the Gaussian mixture model, the hub point set, and the skeleton core points. The fourth processing module is used to calculate the mergeable points of the water collection basins, aggregate the water collection basins using subclass connections and subclass similarity, and obtain high-dimensional data clustering results.
[0048] As one embodiment of the present invention, the first processing module performs vector normalization preprocessing based on the Parkinson's dataset with tremor features and non-tremor features, and calculates a hybrid t-nearest neighbor graph based on the dataset and parameter t; wherein, the hybrid t-nearest neighbor graph includes: the t-nearest neighbor graph of the data points and the reverse t-nearest neighbor graph, as well as the height.
[0049] As one embodiment of the present invention, the third processing module uses a reverse t-nearest neighbor graph to separate the optimized water collection basin and the abnormal water collection basin.
[0050] As one embodiment of the present invention, the fourth processing module calculates the mergeable connections between adjacent water collection basins, optimizes the water collection basins by merging water collection basin-level similarity and subclass connections, and aggregates the remaining abnormal water collection basins to obtain the Parkinson's subtype data clustering results.
[0051] Example 3 The present invention also provides a Parkinson's subtype data clustering system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a Parkinson's subtype data clustering method when executed by the processor.
[0052] Example 4 The present invention also provides a storage medium storing a computer program that executes a Parkinson's subtype data clustering method when running.
[0053] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims
1. A clustering method for Parkinson's subtype data, characterized in that, include: Step S100: Standardize and preprocess the Parkinson's data, and calculate the mixed t-nearest neighbor graph of the data space; Step S200: Calculate the core skeleton points and pivot points based on the hybrid t-nearest neighbor graph and divergence; Step S300: Calculate the naive catch basin based on the Gaussian mixture model, the pivot point set, and the skeleton core points; Step S400: Calculate the mergeable points of the water collection basins, aggregate the water collection basins using subclass connection and subclass similarity, and obtain high-dimensional data clustering results.
2. The Parkinson's subtype data clustering method as described in claim 1, characterized in that, In step S100, vector normalization preprocessing is performed on the Parkinson's dataset with tremor features and without tremor features. At the same time, a mixed t-nearest neighbor graph is calculated based on the dataset and parameter t. The mixed t-nearest neighbor graph includes: the t-nearest neighbor graph and the reverse t-nearest neighbor graph of the data points, as well as the height.
3. The Parkinson's subtype data clustering method as described in claim 3, characterized in that, In step S300, the optimized water collection basin and the abnormal water collection basin are separated using the reverse t-nearest neighbor graph.
4. The Parkinson's subtype data clustering method as described in claim 4, characterized in that, In step S400, the mergeable connections between adjacent water collection basins are calculated, and the water collection basins are optimized by merging water collection basin-level similarity and subclass connections. At the same time, the remaining abnormal water collection basins are aggregated to obtain the clustering results of Parkinson's subtype data.
5. A Parkinson's subtype data clustering device, characterized in that, include: The first processing module is used for the standardized preprocessing of Parkinson's data and to calculate the mixed t-nearest neighbor graph of the data space. The second processing module is used to calculate the core skeleton points and pivot points based on the hybrid t-nearest neighbor graph and divergence; The third processing module is used to calculate the naive water basin based on the Gaussian mixture model, the hub point set, and the skeleton core points. The fourth processing module is used to calculate the mergeable points of the water collection basins, aggregate the water collection basins using subclass connections and subclass similarity, and obtain high-dimensional data clustering results.
6. The Parkinson's subtype data clustering device as described in claim 5, characterized in that, The first processing module performs vector normalization preprocessing on the Parkinson's dataset with and without tremor features. At the same time, it calculates a mixed t-nearest neighbor graph based on the dataset and parameter t. The mixed t-nearest neighbor graph includes: the t-nearest neighbor graph and the reverse t-nearest neighbor graph of the data points, as well as the height.
7. The Parkinson's subtype data clustering device as described in claim 6, characterized in that, The third processing module uses a reverse t-nearest neighbor graph to separate the optimized water collection basin and the abnormal water collection basin.
8. The Parkinson's subtype data clustering device as described in claim 7, characterized in that, The fourth processing module calculates the mergeable connections between adjacent water collection basins, optimizes water collection basins by merging water collection basin-level similarity and subclass connections, and aggregates the remaining abnormal water collection basins to obtain the clustering results of Parkinson's subtype data.
9. A Parkinson's subtype data clustering system, characterized in that, include: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program performing the Parkinson's subtype data clustering method as described in any one of claims 1-4 when executed by the processor.
10. A storage medium, characterized in that, The storage medium stores a computer program that, when executed, performs the Parkinson's subtype data clustering method as described in any one of claims 1-4.