A brain network-based AI-assisted diagnostic method and system for Parkinson's disease
By processing temporal signals from the brain, a steady-state data stream and topological confidence structure are constructed, solving the problem of inaccurate brain network topology in existing technologies and achieving efficient, visualized, and quantitative assessment of Parkinson's disease auxiliary diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU SIXIANG SOFTWARE CO LTD
- Filing Date
- 2026-02-09
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies are unable to effectively construct high-confidence brain network topologies, and cannot locate and quantify functional blockage areas in brain networks, leading to misdiagnosis or missed diagnosis of Parkinson's disease.
By acquiring temporal signals from the brain, performing sliding window segmentation and signal activity filtering, a steady-state data stream is constructed. Based on the phase synchronization index and topological confidence structure, an equivalent resistance model is established to track the potential energy drop of virtual currents and generate a transmission cost subgraph and a brain network saliency heatmap.
It achieves high signal-to-noise ratio preprocessing of brain networks, removes artifact interference, identifies significant functional couplings, locates bottleneck regions in information transmission, and provides intuitive visualization and quantitative evaluation of network status.
Smart Images
Figure CN122074901A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical signal processing technology, specifically to an AI-assisted diagnostic method and system for Parkinson's disease based on brain networks. Background Technology
[0002] Parkinson's disease is a common neurodegenerative disease whose pathological mechanism involves functional abnormalities in the basal ganglia-thalamic cortex circuit. Currently, the clinical diagnosis of Parkinson's disease mainly relies on physicians' subjective observation of patients' motor symptoms such as tremor and rigidity, as well as standardized scale scores.
[0003] However, early symptoms are often subtle and vary greatly from person to person, making it easy for doctors' subjective experience to lead to misdiagnosis or missed diagnosis. Neurophysiological techniques such as electroencephalography (EEG) have been introduced to aid diagnosis, but raw brain signals are often mixed with a large amount of physiological noise from electrooculography (EOG), electromyography (EMG), and other sources. Traditional noise reduction methods struggle to remove artifacts while preserving the integrity of the signal's temporal sequence. Furthermore, existing brain network construction methods are often directly based on signal correlation, which can easily generate numerous spurious connections due to volumetric conduction effects, and there is a lack of effective physical models to characterize the information transmission efficiency of brain regions. This makes it difficult for auxiliary systems to construct high-confidence brain network topologies, locate and quantify functional blockage areas in the brain network, and ultimately fail to meet clinical diagnostic needs.
[0004] Therefore, this invention provides an AI-assisted diagnostic method and system for Parkinson's disease based on brain networks. Summary of the Invention
[0005] The purpose of this invention is to provide an AI-assisted diagnostic method and system for Parkinson's disease based on brain networks, so as to solve the above-mentioned background problems.
[0006] The objective of this invention can be achieved through the following technical solutions: A brain network-based AI-assisted diagnostic method for Parkinson's disease includes the following steps: Continuously acquired temporal brain signals are acquired and segmented into sliding windows, and a discrete sequence is established; the discrete sequence is then filtered for signal activity to establish a set of effective windows for signal activity. A tight evaluation of temporal connectivity between effective window sets is performed to obtain a boundary matching score; based on the boundary matching score, a gradient adaptive stitching process is established for the effective window set to obtain a steady-state data stream with temporal integrity. Based on steady-state data streams, signal synchronization analysis is performed between signals in different brain regions to obtain the phase synchronization index. The phase synchronization index is then used to determine whether it exhibits significant functional coupling. If it does, a topological confidence structure of the brain network is constructed. An equivalent resistance model is established based on the topological confidence structure of the brain network, and the global conductivity of the network nodes is output. Based on the global conductivity, the blockage of virtual current injection is determined. If blockage exists, the potential energy difference of virtual current between nodes is tracked, and a transmission cost subgraph is established.
[0007] As a further technical solution of the present invention, the method for performing the signal activity screening is as follows: Obtain the preset physiological noise range; Each standard deviation value in the discrete sequence is extracted and logically compared with the physiological noise range. Based on the comparison results, the corresponding sub-signal segments are graded and filtered. The discrete sequence is reorganized based on the graded filtering results to obtain an effective window set.
[0008] As a further technical solution of the present invention: the gradient adaptive splicing process is established as follows: Set up the matching acceptance logic, input the boundary matching score into the matching acceptance logic, and determine whether to trigger the logic threshold for signal stitching; If triggered, the adjacent preceding and following windows in the effective window set will undergo sequence recombination processing based on a weighted window function to obtain a steady-state data stream.
[0009] As a further technical solution of the present invention: the method for obtaining the boundary matching score is as follows: Extract adjacent preceding and following windows from the set of valid windows, and read the data frames of a preset length from the tail of the preceding window and the head of the following window as boundary buffers; Based on the preceding window, following window, and boundary buffer in the effective window set, temporal dimension verification is performed. If the temporal verification passes, morphological dimension analysis is performed to obtain the boundary matching score.
[0010] As a further technical solution of the present invention: the method for constructing the topological confidence structure is as follows: Establish a random clutter baseline for the phase synchronization index; Based on the random clutter baseline, a significant functional coupling logic determination is performed on the original phase synchronization index to determine whether significant functional coupling exists. If significant functional coupling is observed, a functional connectivity map of significant functional coupling is constructed. The adjacency matrix of the dynamic functional connectivity graph is sparsified, weak connections with weights lower than the preset functional strength are removed, and strong connections that constitute the brain network skeleton are retained. The sparsed network structure is labeled as a topological confidence structure.
[0011] As a further technical solution of the present invention: the method for establishing the transmission cost subgraph is as follows: Global connectivity logic is determined based on global conductivity. If a blockage is detected, virtual current injection logic is triggered. If blockage exists, select the virtual source and sink points of the topological confidence structure, and perform virtual current injection and potential energy gradient analysis to establish the potential energy gradient field. In the potential energy gradient field, select the set of connecting edges with voltage potential energy drop; Extract the associated nodes of the connected edge set to form the local topology that has the greatest resistance to the virtual current, and define it as the transmission cost subgraph; As a further technical solution of the present invention, the method for establishing the potential energy gradient field is as follows: In the topological confidence structure, the node with the highest betweenness centrality is selected as the virtual source point, and the node with the lowest betweenness centrality is selected as the virtual sink point. A virtual test current of unit intensity is injected. Calculate the node voltage potential energy generated when the virtual test current flows through each node; Traverse all adjacent nodes, calculate the voltage potential energy difference between the two ends of the connecting edge, and generate the potential energy gradient field.
[0012] As a further technical solution of the present invention: gridded density analysis is performed based on the transmission cost subgraph to obtain the spatial damping distribution; a brain network saliency heatmap is generated based on the spatial damping distribution, and the spatial coverage ratio of the high-damping region is calculated, and the brain network operation status is evaluated.
[0013] As a further technical solution of the present invention: the method for evaluating the operating state of the brain network is as follows: Establish color mapping logic to convert the spatial damping distribution into a visual image; Based on the color information after rendering the visualized image, a three-dimensional visualized heatmap of brain network saliency is generated. Anatomical partitions belonging to the basal ganglia circuit in the brain network saliency heatmap were screened out and their spatial coverage percentages were extracted. Logical comparisons were performed based on the spatial coverage percentages to output distribution pattern labels as quantitative statistical indicators of network operation status.
[0014] A brain network-based AI-assisted diagnostic system for Parkinson's disease includes the following modules: Signal acquisition module: used to acquire continuously acquired brain time-series signals, perform sliding window segmentation, and establish a discrete sequence; to filter the discrete sequence for active signals and establish a set of effective windows for active signals; The temporal evaluation module is used to evaluate the temporal connectivity between effective window sets and obtain a boundary matching score. Based on the boundary matching score, gradient adaptive stitching is performed on the effective window set to obtain a steady-state data stream with temporal integrity. Confidence analysis module: Based on steady-state data stream, it performs signal synchronization analysis between signals in each brain region to obtain the phase synchronization index, and determines whether the phase synchronization index shows significant functional coupling; if so, it constructs the topological confidence structure of the brain network. Blockage Analysis Module: Based on the topological confidence structure of the brain network, an equivalent resistance model is established, and the global conductivity of the network nodes is output. Based on the global conductivity, blockage is determined by virtual current injection. If blockage exists, the potential energy difference of the virtual current between nodes is tracked, and a transmission cost subgraph is established. State assessment module: Based on the transmission cost subgraph, a gridded density analysis is performed to obtain the spatial damping distribution; based on the spatial damping distribution, a brain network saliency heatmap is generated, the spatial coverage ratio of high-damped regions is calculated, and the brain network operating state is assessed.
[0015] The beneficial effects of this invention are as follows: 1. By performing sliding window segmentation on continuously acquired temporal brain signals and establishing a discrete sequence, the fluctuation characteristics of the signals can be quantified. Signal activity is filtered by combining a preset physiological noise range, and hierarchical filtering logic is used to remove high-amplitude artifacts and low-amplitude silent dead zones, improving the signal-to-noise ratio of the input data during the data preprocessing stage. Boundary matching scoring with dual temporal and morphological verification is used to assess the physical connection tightness between adjacent windows, which helps reduce logical breaks caused by direct splicing of fragmented data. Gradient-based adaptive splicing processing is employed, enabling non-linear transitions in the signal waveform at the splicing points, which helps reduce the step effect or abrupt artifacts that may occur with traditional hard splicing.
[0016] 2. By constructing a random clutter baseline using a phase synchronization index combined with a substitute data method, the significance of signal coupling strength between brain regions is tested, identifying substantial functional couplings working in concert amidst background noise. Constructing a topological confidence structure facilitates the removal of redundant weak connections from the whole-brain network, enhancing the network model's ability to represent physiological mechanisms. The brain network topological confidence structure is transformed into an equivalent resistance model, and global conductivity is calculated, quantifying the information transmission efficiency of the brain network through physical circuit concepts. Virtual current injection and potential energy drop tracking help locate nodes generating high impedance effects in the network topology. Potential energy gradient-based analysis methods are beneficial for reflecting bottleneck regions in the information interaction process, constructing a transmission cost subgraph reflecting the network transmission stagnation state.
[0017] 3. Map the transmission cost subgraph to a three-dimensional spatial damping distribution through grid density analysis, and further transform it into a brain network significance heat map, realizing the intuitive conversion of data features from the graph theory space to the anatomical space and enhancing the visualization of data results. Calculate the spatial coverage ratio of high-damping regions and output distribution pattern labels, simplifying the network state into statistical indicators, which is conducive to reflecting the physical range and distribution pattern of abnormal brain network function regions and providing intuitive and quantitative references for professionals to understand the operating state of the brain network. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The present invention will be further described below with reference to the accompanying drawings.
[0019] Figure 1 is a flowchart of an AI-assisted Parkinson's disease diagnosis method based on brain networks according to the present invention; Figure 2 is a functional module diagram of an AI-assisted Parkinson's disease diagnosis system based on brain networks in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0020] In order to make the technical means, creative features, achieved purposes, and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments. Embodiment 1
[0021] As Figure 1 shown, an AI-assisted Parkinson's disease diagnosis method based on brain networks includes the following steps: S10. Obtain continuously collected brain time series signals, perform sliding window segmentation, and establish a dispersion sequence; perform signal activity screening on the dispersion sequence to establish a set of effective windows with signal activity; Among them, the method of obtaining continuously collected brain time series signals, performing sliding window segmentation, and establishing a dispersion sequence is as follows: In some embodiments, obtain the brain region time series signals of multi-channel scalp brain signals covering the whole brain through an electroencephalogram acquisition device; Set a fixed time window length and a sliding step size, and construct a sliding window that moves unidirectionally along the time axis; Exemplarily, the method of performing sliding window segmentation is: set the time window length to T and the sliding step size to S (where S < T), control the sliding window to slide依次 along the brain time series signal from the starting moment with a step size of S, and truncate the continuous signal into several sub-signal segments with overlapping time series; Perform statistical calculations on the amplitudes of all sampling points within each sub-signal segment, and extract the statistic representing the severity of signal fluctuation as the dispersion degree of the sliding window; Preferably, the method for calculating the degree of dispersion is as follows: the standard deviation algorithm is used to calculate the degree of deviation of the voltage value of each sampling point in the sub-signal segment from the mean, and the corresponding standard deviation value is obtained; Arrange the standard deviation values corresponding to all sub-signal segments in chronological order to generate a discrete sequence that maps one-to-one with the time window; The method for filtering the discrete sequence to establish an effective window set of signal activity is as follows: Obtain the preset physiological noise range; Among them, the physiological noise range is composed of an upper threshold and a lower threshold that characterize the intensity of normal human brain electrical activity; Each standard deviation value in the discrete sequence is traversed and logically compared with the physiological noise range. Based on the comparison results, the corresponding sub-signal segments are graded and filtered. Based on the graded filtering results, the discrete sequence is reorganized to obtain the effective window set. The preferred method for hierarchical screening is as follows: If the standard deviation of the current sliding window is higher than the upper limit threshold, it indicates that the signal in the sliding window contains high-amplitude artifact interference (such as electrooculography, electromyography, or body movement noise). The sliding window is determined to be a noise frequency band, marked as an invalid window, and removed. If the standard deviation of the current sliding window is lower than the lower threshold, it indicates that the signal in the sliding window is in a flat baseline state or a dead zone state caused by poor equipment contact, lacking effective physiological information. The current sliding window is determined to be a silent frequency band, and the sliding window is marked as an invalid window and removed. If the standard deviation of the current sliding window is within the closed interval between the upper and lower thresholds, it indicates that the signal fluctuation of the sliding window conforms to the physiological characteristics of neuronal firing in the cerebral cortex, and the current sliding window is determined to be an active frequency band. Extract the time indices of all sub-signal segments identified as active frequency bands and reassemble them into a set of effective windows for active signals.
[0022] S20. Perform a tight evaluation of the temporal connectivity between the effective window set to obtain a boundary matching score; based on the boundary matching score, perform gradient adaptive stitching processing on the effective window set to obtain a steady-state data stream with temporal integrity. The method for performing both temporal and morphological consistency checks on the effective window set to obtain the boundary matching score is as follows: In some embodiments, adjacent preceding and following windows are extracted from the set of valid windows, and a data frame of a preset length is read from the tail of the preceding window and the head of the following window as a boundary buffer. Based on the preceding window, following window and boundary buffer in the effective window set, temporal dimension verification is performed. If the temporal verification is passed, morphological dimension analysis is performed to obtain the boundary matching score. The preferred method for performing time-series and morphological dimension verification is as follows: Calculate the absolute time difference between the end time of the preceding window and the start time of the following window. If the time difference is less than the preset receptive field tolerance of TCN (Temporal Convolutional Network), then the temporal verification is passed. Perform a first-order difference operation on the signal data within the boundary buffer to extract the instantaneous rate of change vector of the signal at the breakpoint; Calculate the cosine similarity between the rate of change vector at the tail of the preceding window and the rate of change vector at the head of the following window; The absolute time difference and cosine similarity are input into a pre-set logistic regression algorithm, which outputs a probability value between 0 and 1 to obtain a boundary matching score, which is used to characterize whether two windows belong to the same structured data stream. Among them, the method for determining the physically continuous steady-state data stream by performing weighted gradient-based adaptive stitching based on boundary matching scoring is as follows: S201. Set the matching acceptance logic, input the boundary matching score into the matching acceptance logic, and determine whether to trigger the logic threshold of signal stitching. If the boundary matching score is higher than or equal to the preset matching acceptance threshold, it indicates that the two windows are close in time and have a smooth transition in waveform trend, and there is no abrupt artifact. Then the logic threshold for signal stitching is executed. Conversely, if the boundary matching score is lower than the preset matching acceptance threshold, the change in the boundary matching score will be continuously monitored. S202. If triggered, perform sequence recombination processing based on weighted window function on adjacent preceding and following windows in the effective window set to obtain a steady-state data stream. Preferably, the method for performing sequence recombination processing based on weighted window functions is as follows: determine the overlapping area between the preceding window and the following window in terms of time sequence; Construct a temporal window function with edge attenuation characteristics (such as the Hanning window), and use the temporal window function to perform weighted fusion of corresponding data points of two windows in the overlapping area to generate a fused frame with a smooth transition. It should be noted that, within the overlapping region, the time-domain window function performs complementary energy attenuation on the tail data of the preceding window and the head data of the following window to suppress high-frequency transition artifacts at the signal truncation point and preserve the phase envelope characteristics of the signal. By using fused frames to logically connect the effective portions of the preceding window and the effective portions of the following window, a quasi-steady-state data stream with suppressed edge effects is formed. The quasi-steady-state data stream is encapsulated into standardized time-series samples, stored in a standardized dataset of a distributed storage system, and labeled as a steady-state data stream. Example 2
[0023] Please see Figure 1 As shown, an AI-assisted diagnostic method for Parkinson's disease based on brain networks includes the following steps: S30. Based on steady-state data stream, perform signal synchronization analysis between signals in each brain region to obtain the phase synchronization index, and determine whether the phase synchronization index shows significant functional coupling; if so, construct the topological confidence structure of the brain network. The method for obtaining the phase synchronization index by performing signal synchronization analysis between brain regions based on steady-state data streams is as follows: In some embodiments, steady-state data streams are parsed and multidimensional temporal signal matrices corresponding to different brain region channels are extracted; Hilbert transform is performed on the time-series signal in each dimension of the multidimensional time-series signal matrix to convert the EEG signal in the real domain into an analytic signal in the complex domain, thereby extracting the instantaneous phase angle corresponding to each sampling point; Select any two brain region channels and calculate the phase difference of the instantaneous phase angle of the two brain region channels at the same moment; The time length of the steady-state data stream is used as the integration interval. The complex exponent of the phase difference is averaged and its magnitude is taken as the phase synchronization exponent. It should be noted that the phase synchronization index is a scalar between 0 and 1. The closer the value is to 1, the more strictly the two brain regions maintain a consistent pace in terms of neural oscillation frequency, that is, there is potential information exchange. Among them, the method for determining whether the phase synchronization index exhibits significant functional coupling, and if so, constructing the topological confidence structure of the brain network, is as follows: S301. Establish a random clutter baseline with phase synchronization index; Preferably, the method for establishing a random clutter baseline is as follows: using the substitution data method, the time-series signals in the steady-state data stream are phase-scrambled or randomly rearranged to destroy the original time correlation and generate a set of substitution data containing only background noise. Calculate the phase synchronization index of the substitute data and use the 95th percentile of the statistical distribution as the random coupling threshold; S302. Based on the random clutter baseline, perform a significant functional coupling logic determination on the original phase synchronization index to determine whether significant functional coupling exists. Preferably, the logical determination method is to compare the original phase synchronization index calculated in S301 with the random coupling threshold. If the original phase synchronization index is lower than or equal to the random coupling threshold, the connection is determined to be random clutter and is an invalid connection, and its weight is reset to zero. If the original phase synchronization index is higher than the random coupling threshold, it indicates that the statistical significance of the connection strength is significantly higher than the background noise, and it is determined that it presents significant functional coupling, that is, there is substantial collaborative work between brain regions. S303. If significant functional coupling is observed, a functional connectivity map of significant functional coupling shall be established. Preferably, the method for establishing the functional connectivity graph is as follows: Extract all brain region node pairs identified as having significant functional coupling and their corresponding phase synchronization indices; Using brain region nodes as vertices in graph theory and the selected phase synchronization index as edge weights, a weighted undirected graph is constructed as the dynamic functional connectivity graph corresponding to the steady-state data stream. The method for constructing the topological confidence structure of brain networks is as follows: The adjacency matrix of the dynamic functional connectivity graph is sparsified, weak connections with weights lower than the preset functional strength are removed, and strong connections that constitute the brain network skeleton are retained. Preferably, the sparsification process is as follows: set a network density retention threshold (preferably retaining the top 20% of the strongest connections), and extract only the connection edges with the highest weights in the adjacency matrix; The sparsed adjacency matrix is mapped back to the brain anatomical template to identify key paths in the basal ganglia-thalamic cortex loop and the sparsed network structure is marked as a topological confidence structure. A weighted adjacency matrix is established based on the phase synchronization index in the topological confidence structure; Understandably, the purpose of determining the topological confidence structure of brain networks is to: Function 1: Reduce spurious connections: Through dual filtering of random clutter baseline and sparsification, spurious synchronization caused by volume conduction effect or measurement noise is removed, ensuring that the constructed network structure reflects the real neural conduction path; Function 2: Targeting core pathological features: The pathological changes in Parkinson's disease are often subtle and hidden in the complex signals of the whole brain. By locking onto high-confidence topological confidence structures, the functional abnormalities of the basal ganglia circuit can be highlighted.
[0024] S40. Based on the topological belief structure of the brain network, establish an equivalent resistance model and output the global conductivity of the network nodes; based on the global conductivity, determine the blockage of virtual current injection; if there is a blockage, track the potential energy difference of the virtual current between nodes and establish a transmission cost subgraph. Among them, the equivalent resistance model based on the topological confidence structure of brain networks is established, and the global conductivity of network nodes is output in the following way: In some embodiments, the weighted adjacency matrix of the topological confidence structure is extracted and converted into a Laplace matrix; Extract the weighted adjacency matrix of the topological confidence structure and construct the graph Laplacian matrix; The effective resistance distance between any two nodes is calculated based on the graph Laplacian matrix, which serves as the equivalent resistance model. Preferably, the method for calculating the effective resistance distance is as follows: Calculate the Moore-Penrose generalized inverse of the Laplace matrix. Based on the formula Calculate the effective resistance distance between nodes i and j; The equivalent resistance matrix of the entire network is decomposed into eigenvalues, the smallest non-zero eigenvalue is extracted, and defined as the global conductivity characterizing the network synchronization capability. The method for determining blockage based on virtual current injection using global conductivity is as follows: S401. Global connectivity logic is determined based on global conductivity. If a blockage is determined, virtual current injection logic is triggered. Preferably, the method for performing global connectivity logic determination is to compare the calculated global conductivity with the preset standard manifold reference conductivity; If the global conductivity is higher than or equal to the reference conductivity, the network transmission is determined to be in a superconducting state, the entire network is marked as a low-cost region, and no further tracking is performed. If the global conductivity is lower than the reference conductivity, the network is determined to be in a high-damping state, i.e., there is a blockage, and the virtual current injection logic is automatically triggered. S402. If blockage exists, select the virtual source and sink points of the topological confidence structure, and perform virtual current injection and potential energy gradient analysis to establish the potential energy gradient field. Preferably, the method for selecting virtual source and sink points of the topological confidence structure and performing virtual current injection and potential energy gradient analysis to establish the potential energy gradient field is as follows: In the topological confidence structure, the node with the highest betweenness centrality is selected as the virtual source point, and the node with the lowest betweenness centrality is selected as the virtual sink point. A virtual test current of unit intensity is injected. Based on Kirchhoff's circuit laws, the node voltage potential energy generated when the virtual test current flows through each node is calculated. Traverse all adjacent nodes, calculate the voltage potential energy difference between the two ends of the node connection edge, and generate the potential energy gradient field; The method for establishing the transmission cost subgraph is as follows: In the potential energy gradient field, select the set of connection edges whose voltage potential energy drop exceeds the impedance threshold (preferably twice the average drop of the entire network); Extract the associated nodes of the connected edge set to form the local topology that has the greatest resistance to the virtual current, and define it as the transmission cost subgraph.
[0025] S50. Perform gridded density analysis based on the transmission cost subgraph to obtain the spatial damping distribution; generate a brain network saliency heatmap based on the spatial damping distribution, calculate the spatial coverage ratio of high-damped regions, and evaluate the brain network operation status. Among them, the spatial damping distribution is obtained by performing gridded density analysis based on the transmission cost subgraph as follows: The nodes contained in the transmission cost subgraph established in S40 are defined as high-damped nodes; By introducing a pre-defined brain anatomical atlas (such as the AAL template), the brain space is divided into several independent anatomical partition units. Establish the positional index relationship between network nodes and anatomical partition units; For example, the method for performing gridded density statistics is as follows: Traverse each anatomical partition unit, count the number of high-damping nodes falling into the unit, calculate the ratio of the number of high-damping nodes to the total number of network nodes in the unit, and define it as the local damping density of the unit. The local damping density of all anatomical units is combined according to spatial location to form a spatial damping distribution data matrix covering the entire brain. The method for generating a saliency heatmap of the brain network based on spatial damping distribution, calculating the spatial coverage ratio of high-damping regions, and evaluating the operational status of the brain network is as follows: S501. Establish color mapping logic to convert the spatial damping distribution into a visual image; Preferably, the method for establishing color mapping logic is to set a color gradient spectrum; The local damping density values of the three-dimensional grid cells are normalized and mapped. Based on the mapping results and combined with a color gradient spectrum, the spatial damping distribution is transformed into a visualization. For example, the way to convert the spatial damping distribution into a visualization is to set a color gradient spectrum, transitioning from cool blue to warm red, while normalizing the local damping density values of the three-dimensional grid cells to the range of 0 to 1. The closer the value is to 0, the more blue the corresponding 3D grid cell is rendered, indicating that the regional information transmission is smooth. The closer the value is to 1, the red the corresponding 3D mesh cell is, which means that there is high impedance accumulation in the regional information transmission, that is, a high damping area. S502. Based on the color information after rendering the visualized image, generate a three-dimensional visualized brain network saliency heatmap. S503. Screen out the anatomical partition units of the basal ganglia circuit in the brain network saliency heatmap and extract the spatial coverage ratio. Based on the spatial coverage ratio, perform logical comparison and output the distribution pattern label as a quantitative statistical indicator of the network operation status. Preferably, the method for extracting the spatial coverage ratio is as follows: count the number of partition units with local damping density greater than a preset threshold, calculate the ratio of the number of partition units in all partition units of the basal ganglia circuit, and use it as the spatial coverage ratio. The spatial coverage ratio is compared with a preset interval threshold, and the distribution pattern label (such as discrete distribution, clustered distribution, diffuse distribution) is output as a quantitative statistical indicator of network operation status. It is understandable that the purpose of outputting quantitative statistical indicators is: Function 1: Transforms transmission costs into anatomical images, and through the highlighting of heat maps, assists professionals in quickly identifying the spatial location of data anomalies, providing objective informatics references; Secondly, by calculating the spatial coverage ratio, it reflects the change in the physical range of the high impedance region, providing a reusable mathematical benchmark for subsequent comparison of data differences at different time points, rather than directly giving a disease diagnosis conclusion. Example 3
[0026] Please see Figure 2 As shown, an AI-assisted diagnostic system for Parkinson's disease based on brain networks includes the following modules: Signal acquisition module: used to acquire continuously acquired brain time-series signals, perform sliding window segmentation, and establish a discrete sequence; to filter the discrete sequence for active signals and establish a set of effective windows for active signals; The temporal evaluation module is used to evaluate the temporal connectivity between effective window sets and obtain a boundary matching score. Based on the boundary matching score, gradient adaptive stitching is performed on the effective window set to obtain a steady-state data stream with temporal integrity. Confidence analysis module: Based on steady-state data stream; performs signal synchronization analysis between signals in each brain region to obtain the phase synchronization index, and determines whether the phase synchronization index shows significant functional coupling; if so, it constructs the topological confidence structure of the brain network; Blockage Analysis Module: Based on the topological confidence structure of the brain network, an equivalent resistance model is established, and the global conductivity of the network nodes is output. Based on the global conductivity, blockage is determined by virtual current injection. If blockage exists, the potential energy difference of the virtual current between nodes is tracked, and a transmission cost subgraph is established. State assessment module: Based on the transmission cost subgraph, a gridded density analysis is performed to obtain the spatial damping distribution; based on the spatial damping distribution, a brain network saliency heatmap is generated, the spatial coverage ratio of high-damped regions is calculated, and the brain network operating state is assessed.
[0027] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.
Claims
1. A brain network-based AI-assisted diagnostic method for Parkinson's disease, characterized in that, Includes the following steps: Continuously acquired temporal brain signals are acquired and segmented into sliding windows, and a discrete sequence is established; the discrete sequence is then filtered for active signals to establish a set of effective windows for active signals. A tight evaluation of temporal connectivity between effective window sets is performed to obtain a boundary matching score; based on the boundary matching score, gradient adaptive stitching processing is performed on the effective window sets to obtain a steady-state data stream with temporal integrity. Based on steady-state data streams, signal synchronization analysis is performed between signals from different brain regions to obtain the phase synchronization index. It is then determined whether the phase synchronization index shows significant functional coupling. If it does, a topological confidence structure of the brain network is constructed. An equivalent resistance model is established based on the topological belief structure of brain networks, and the global conductivity of network nodes is output. Blockage determination is performed based on global conductivity for virtual current injection. If blockage exists, the potential energy difference of virtual current between nodes is tracked, and a transmission cost subgraph is established.
2. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 1, characterized in that: The method for performing the signal activity filtering is as follows: Obtain the preset physiological noise range; Each standard deviation value in the discrete sequence is extracted and logically compared with the physiological noise range. Based on the comparison results, the corresponding sub-signal segments are graded and filtered. The discrete sequence is reorganized based on the graded filtering results to obtain an effective window set.
3. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 1, characterized in that: The gradient adaptive stitching process is established as follows: Set up the matching acceptance logic, input the boundary matching score into the matching acceptance logic, and determine whether to trigger the logic threshold for signal stitching; If triggered, a sequence recombination process based on a weighted window function is performed on adjacent preceding and succeeding windows in the effective window set to obtain a steady-state data stream.
4. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 3, characterized in that: The boundary matching score is obtained as follows: Extract adjacent preceding and following windows from the set of valid windows, and read the data frames of a preset length from the tail of the preceding window and the head of the following window as boundary buffers; Based on the preceding window, following window, and boundary buffer in the effective window set, temporal dimension verification is performed. If the temporal verification passes, morphological dimension analysis is performed to obtain the boundary matching score.
5. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 1, characterized in that: The topological confidence structure is constructed as follows: Establish a random clutter baseline for the phase synchronization index; Based on the random clutter baseline, a significant functional coupling logic determination is performed on the original phase synchronization index to determine whether significant functional coupling exists. If significant functional coupling is observed, a functional connectivity map of significant functional coupling is constructed. The adjacency matrix of the dynamic functional connectivity graph is sparsified, weak connectivity edges with weights lower than the preset functional strength are removed, and strong connectivity edges that constitute the brain network skeleton are retained. The sparsed network structure is labeled as a topological confidence structure.
6. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 1, characterized in that: The transmission cost subgraph is constructed as follows: Global connectivity logic is determined based on global conductivity. If a blockage is detected, virtual current injection logic is triggered. If blockage exists, select the virtual source and sink points of the topological confidence structure, and perform virtual current injection and potential energy gradient analysis to establish the potential energy gradient field. In the potential energy gradient field, select the set of connecting edges with voltage potential energy drop; Extract the associated nodes of the connected edge set to form the local topology that has the greatest resistance to the virtual current, and define it as the transmission cost subgraph.
7. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 6, characterized in that: The method for establishing the potential energy gradient field is as follows: In the topological confidence structure, the node with the highest betweenness centrality is selected as the virtual source point, and the node with the lowest betweenness centrality is selected as the virtual sink point. A virtual test current of unit intensity is injected. Calculate the node voltage potential energy generated when the virtual test current flows through each node; Traverse all adjacent nodes, calculate the voltage potential energy difference between the two ends of the connecting edge, and generate the potential energy gradient field.
8. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 1, characterized in that: Gridded density analysis is performed based on the transmission cost subgraph to obtain the spatial damping distribution; a brain network saliency heatmap is generated based on the spatial damping distribution, the spatial coverage ratio of high-damped regions is calculated, and the brain network operation status is evaluated.
9. The AI-assisted diagnostic method for Parkinson's disease based on brain networks according to claim 1, characterized in that: The method for assessing the operational status of the brain network is as follows: Establish color mapping logic to convert the spatial damping distribution into a visual image; Based on the color information after rendering the visualized image, a three-dimensional visualized brain network saliency heatmap is generated. Anatomical partitions belonging to the basal ganglia circuit in the brain network saliency heatmap were screened out and their spatial coverage percentages were extracted. Logical comparisons were performed based on the spatial coverage percentages to output distribution pattern labels as quantitative statistical indicators of network operation status.
10. A brain network-based AI-assisted diagnostic system for Parkinson's disease, used to implement any one of the brain network-based AI-assisted diagnostic methods for Parkinson's disease as described in claims 1-9, characterized in that, Includes the following modules: Signal acquisition module: used to acquire continuously acquired brain time-series signals, perform sliding window segmentation, and establish a discrete sequence; to filter the discrete sequence for active signals and establish a set of effective windows for active signals; The temporal evaluation module is used to evaluate the temporal connectivity between effective window sets and obtain a boundary matching score. Based on the boundary matching score, gradient adaptive stitching is performed on the effective window set to obtain a steady-state data stream with temporal integrity. Confidence analysis module: based on steady-state data stream; Signal synchronization analysis was performed between brain regions to obtain the phase synchronization index, and it was determined whether the phase synchronization index showed significant functional coupling; if so, the topological confidence structure of the brain network was constructed. Blockage Analysis Module: Based on the topological confidence structure of brain networks, an equivalent resistance model is established, and the global conductivity of network nodes is output. Blockage determination is performed based on global conductivity for virtual current injection. If blockage exists, the potential energy difference of virtual current between nodes is tracked, and a transmission cost subgraph is established. State assessment module: Based on the transmission cost subgraph, a gridded density analysis is performed to obtain the spatial damping distribution; based on the spatial damping distribution, a brain network saliency heatmap is generated, the spatial coverage ratio of high-damped regions is calculated, and the brain network operating state is assessed.