Fiber bundle automatic segmentation and quantitative labeling method for white matter abnormality of Parkinson's disease

Abstract

The invention discloses a fiber bundle automatic segmentation and quantitative labeling method for white matter abnormality of Parkinson's disease, and belongs to the technical field of medical image processing. The problems that in the prior art, white matter anomaly detection depends on manual sketching or a traditional machine learning method, efficiency is low, and subjectivity is high are solved, by setting the first threshold value and the second threshold value, the abnormal fiber bundles can be recognized, the needed threshold value range can be selected according to different research purposes and clinical requirements, and the detection accuracy is improved. Therefore, reliable judgment results can be provided after definite diagnosis and in early screening; if the two thresholds are selected to be used at the same time, the method can comprehensively judge each parameter of each fiber bundle, so that the white matter abnormal fiber bundle of the Parkinson's disease patient can be more accurately identified, the accuracy of the marking result of the white matter abnormal fiber bundle of the Parkinson's disease is improved, and the accuracy of the marking result of the Parkinson's disease white matter abnormal fiber bundle is improved. The high-precision positioning function of the fiber bundle for white matter abnormality of the Parkinson's disease is achieved, and more powerful support is provided for clinical diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a fiber bundle automatic segmentation and quantitative labeling method for white matter abnormalities of Parkinson's disease. BACKGROUND

[0002] Parkinson's disease is a neurodegenerative disease, and its pathological features include not only the degeneration of dopaminergic neurons in the substantia nigra, but also the microstructural abnormalities of white matter fiber bundles, such as the decrease in FA value shown by diffusion tensor imaging (DTI).

[0003] Currently, the detection of white matter abnormalities in the prior art relies on manual delineation or traditional machine learning methods, resulting in low efficiency and strong subjectivity. SUMMARY

[0004] The present application provides a fiber bundle automatic segmentation and quantitative labeling method for white matter abnormalities of Parkinson's disease, which can not only identify abnormal fiber bundles, but also select the required threshold range according to different research purposes and clinical needs, thereby ensuring reliable judgment results in both diagnosis and early screening; if both thresholds are selected, the method can make comprehensive judgments on each parameter of each fiber bundle, forming a more comprehensive conclusion, thereby more accurately detecting abnormal white matter fiber bundles in Parkinson's disease patients, improving the accuracy of fiber bundle labeling results of Parkinson's disease white matter abnormalities, and providing stronger support for clinical diagnosis, solving the problems raised in the above background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: A fiber bundle automatic segmentation and quantitative labeling method for white matter abnormalities of Parkinson's disease, comprising the following steps: S1, collecting diffusion magnetic resonance image data with normal fiber bundles, and unifying all data to the same anatomical space to form a reference database; S2, extracting part of the data of the reference database as a healthy control group; S3, performing automatic segmentation of the fiber bundle by the TractSeg algorithm, extracting the main fiber bundle; calculating the fiber bundle parameters of the healthy control group using the fiber tracking algorithm, the fiber bundle parameters including fiber density, fiber cross section, and fiber density-cross section product; and statistically analyzing the fiber bundle parameters of each healthy control group individual to calculate the mean and standard deviation; S4, performing Z-score standardization processing on the fiber bundle parameters of the healthy control group for detecting abnormal fiber bundles; S5, collect diffusion magnetic resonance image data of the target Parkinson's disease patient, preprocess the diffusion magnetic resonance image data thereof, and calculate fiber bundle parameters of the target Parkinson's disease patient using the step S3; S6, compare the fiber bundle parameters of the target Parkinson's disease patient with the Z-score standardized database of the healthy control group, calculate the Z-score value of each fiber bundle parameter, and label the fiber bundle whose Z-score value is out of the preset threshold range as an abnormal fiber bundle; S7, according to the position and characteristics of the abnormal fiber bundle, label it as a specific fiber bundle related to Parkinson's disease, map the Z-score value of the abnormal fiber bundle to a heat map using a visualization tool, and generate detailed labeling information for each abnormal fiber bundle.

[0006] Further, in S6, the fiber bundle whose Z-score value is out of the preset threshold range is labeled as an abnormal fiber bundle, comprising: Extract data samples from the Z-score standardized database of the healthy control group, and analyze the Z-score distribution of different fiber bundle parameters; According to the statistical analysis result, set an initial threshold; According to the initial threshold, set a first threshold and a second threshold respectively according to the research purpose and clinical demand, the first threshold is used to reduce false positive detection, and the second threshold is used to improve the sensitivity of detection.

[0007] Further, in S6, the fiber bundle whose Z-score value is out of the preset threshold range is labeled as an abnormal fiber bundle, further comprising: Obtain the fiber bundle parameters of the Parkinson's disease patient, and the research purpose and clinical demand, and determine to select the first threshold, the second threshold or both according to the research purpose and clinical demand; When the first threshold is selected, if the Z-score value of any one parameter of any one fiber bundle exceeds the first threshold, the parameter of the fiber bundle is labeled as high confidence abnormal; When the second threshold is selected, if the Z-score value of any one parameter of any one fiber bundle exceeds the second threshold but does not exceed the first threshold, the parameter of the fiber bundle is labeled as low confidence abnormal; When both are used, compare each parameter of each fiber bundle with the first threshold and the second threshold respectively, and obtain the comprehensive judgment result of each parameter of each fiber bundle.

[0008] Further, in S1, diffusion magnetic resonance image data with normal fiber bundles are collected, and all data are unified to the same anatomical space to form a reference database, comprising: Traverse the Z-score value of each fiber bundle parameter in the reference database, and compare it with the first threshold value and the second threshold value respectively to obtain the judgment result of each fiber bundle parameter; Obtain the data samples in the reference database whose Z-score value of the fiber bundle parameter exceeds the first threshold value range as the first data set; Obtain the data samples in the reference database whose Z-score value of the fiber bundle parameter exceeds the second threshold value range but does not exceed the first threshold value range as the second data set; Create two classification storage structures in the reference database, including: the first threshold value storage structure and the second threshold value storage structure; Match the first data set and the second data set to the corresponding data sample cache area respectively, and label the information of each fiber bundle parameter stored, including: fiber bundle name, parameter type, Z-score value, corresponding threshold category and other related information.

[0009] Further, the fiber bundle parameters of the target Parkinson's disease patient in S5 are calculated using the S3 step, including: Obtain the preprocessed diffusion magnetic resonance image data of the target Parkinson's disease patient, and match the corresponding healthy control group, and obtain the fiber tracking algorithm parameters of the healthy control group; Adjust the fiber tracking algorithm parameters of the healthy control group, fit the tensor model to its diffusion magnetic resonance image data, and obtain the diffusion tensor parameters of each voxel, including: anisotropy fraction, radial diffusion rate; In the brain atlas of the standard space, define the target area of the fiber bundle required for research; Starting from the seed point of interest in the target area, track along the main diffusion direction until the termination condition is met; During the tracking process, record the voxel positions on the fiber path in real time.

[0010] Further, the fiber bundle parameters of the target Parkinson's disease patient in S5 are calculated using the S3 step, including: Select the region of interest, count the number of fiber paths passing through the region, and divide by the volume of the region to obtain the fiber density value; Select the fiber path, count the number of voxels intersecting the fiber path, and multiply by the volume of the voxel to obtain the fiber cross-sectional value; Multiply the fiber density value by the fiber cross-sectional value to obtain the fiber density-cross-sectional product; Record the personal information of the target Parkinson's disease patient, and store the calculated fiber bundle parameters in the personal information.

[0011] Further, in S7, according to the location and characteristics of the abnormal fiber bundle, it is labeled as a specific fiber bundle related to Parkinson's disease, including: Obtain the fiber path obtained by the fiber tracking algorithm, and clearly label the name of the abnormal fiber bundle; According to the name of the abnormal fiber bundle, determine the specific location of the abnormal fiber bundle in the brain; According to the fiber density, fiber cross section and Z-score value of fiber density-cross section product of the fiber bundle, determine the abnormality degree of the fiber bundle; Create a three-dimensional brain template image, and map the Z-score value of each fiber bundle to the corresponding location; Convert the Z-score value of the abnormal fiber bundle to a specific color and generate a heat map; In the heat map, label the name, Z-score value and corresponding location information of each abnormal fiber bundle; According to the abnormality of the fiber bundle parameters, generate a comprehensive conclusion.

[0012] Further, in S1, collect diffusion magnetic resonance image data with normal fiber bundles, and unify all data to the same anatomical space to form a reference database, also including: According to the research needs and actual situation of data collection, set the update frequency of the reference database; Use the Euclidean distance method to calculate the similarity between the new data sample and the historical data sample; According to the similarity measurement result, calculate the repetition rate of the new data sample and the historical data sample; Identify the data sample with high repetition rate, and delete the historical data sample with high repetition rate from the reference database; Store the new data sample in the corresponding cache area to form an updated reference database.

[0013] Further, in S3, statistical analysis is performed on the fiber bundle parameters of each healthy control group individual to calculate the mean and standard deviation, including: Obtain the same fiber bundle parameter value of each individual in all healthy control groups, add all the same fiber bundle parameter values of the individuals, and then divide by the number of individuals to obtain the average value of the parameter; Calculate the square of the difference between each parameter value and the average value, sum it up, divide by the number of individuals, and then take the square root to obtain the standard deviation of the parameter.

[0014] Further, according to the abnormality of the fiber bundle parameters, a comprehensive conclusion is generated, including: In the heat map, mark the abnormality degree and distribution of the abnormal fiber bundle by different colors; Insert the generated heat map into the diagnostic report and list the labeling information of each abnormal fiber bundle for clinical use.

[0015] Further, the fiber bundle automatic segmentation and quantitative labeling method for white matter abnormalities in Parkinson's disease further comprises: Extracting fiber density, fiber cross-section, and fiber density-cross-section product from the fiber bundle parameters of the target Parkinson's disease patient; Based on the fiber density, fiber cross-section, and fiber density-cross-section product, calculating the fiber bundle abnormality degree index, and the calculation formula is as follows:

[0016] Wherein, A is the fiber bundle abnormality degree index, ranging from 0 to 1, the higher the value, the higher the degree of fiber bundle abnormality; FD is the fiber density parameter, indicating the number of fibers per unit volume in the fiber bundle of the target Parkinson's disease patient calculated by S5 step, the unit is per cubic millimeter, reflecting the density of the fiber bundle; FCS is the fiber cross-section parameter, indicating the average cross-sectional area of the fiber bundle path calculated by S5 step, the unit is square millimeter, reflecting the spatial distribution characteristics of the fiber bundle; FDP is the fiber density-cross-section product parameter, indicating the product of FD and FCS, the unit is dimensionless, reflecting the comprehensive structural characteristics of the fiber bundle; w1, w2, w3 are weight coefficients, corresponding to the weighted influence of fiber density, fiber cross-section and fiber density-cross-section product respectively, ranging from 0 to 1, and satisfying w1+w2+w3=1, determined by statistical analysis of fiber bundle parameters of healthy control group and machine learning model training; Q is the data quality factor, indicating the signal-to-noise ratio score of the diffusion magnetic resonance image data of the target Parkinson's disease patient, ranging from 0 to 10, the higher the value, the higher the data quality, calculated by weighting the image clarity and stereoscopic microscope calibration parameters evaluated in the preprocessing step; exp(-Q) is an exponential function, used to adjust the nonlinear influence of data quality factor on fiber bundle abnormality degree index, to ensure the sensitivity of abnormality degree index is reduced when the data quality is poor; According to the calculated fiber bundle abnormality degree index, the mapping relationship between the pre-set abnormality degree index and abnormality grade is determined to determine the abnormality grade of each fiber bundle; Combine the determined abnormality grade with the corresponding fiber bundle name, location information and Z-score value to generate a dynamic evaluation report of fiber bundle abnormality degree; In the dynamic evaluation report, list the abnormality degree index, abnormality grade, fiber bundle name, Z-score value and corresponding anatomical location of each fiber bundle; Map the abnormality degree index in the dynamic evaluation report to the three-dimensional brain template image; The generated dynamic evaluation report and three-dimensional brain template image are transmitted to the visualization tool for clinical diagnosis and subsequent research; The dynamic assessment report is stored in the patient database, associated with the patient's personal information and historical fiber bundle parameters, providing data support for long-term tracking of the progression of white matter abnormalities in Parkinson's disease.

[0017] Further, the fiber bundle automatic segmentation and quantitative labeling method for white matter abnormalities in Parkinson's disease further comprises: From the database of target Parkinson's disease patients, extract diffusion magnetic resonance image data collected at consecutive time points, including fiber bundle parameter data of at least three time nodes; Preprocess the fiber bundle parameter data of each time node to ensure that all data are unified to the same anatomical space, and use the TractSeg algorithm and fiber tracking algorithm in the S3 step to calculate the fiber density, fiber cross section and fiber density-cross section product of each time node, respectively; Compare the fiber bundle parameters of each time node with the Z-score standardized database of the healthy control group, and calculate the Z-score value of each fiber bundle parameter at each time node; According to the Z-score value, identify the abnormal fiber bundle at each time node that exceeds the preset threshold range, and record its name, anatomical location and abnormality degree; Perform time series analysis on the Z-score values of the same fiber bundle at different time nodes to calculate the rate of change of the Z-score values, which is calculated by the difference between the Z-score values of adjacent time nodes divided by the time interval, with the unit of Z-score change per month; According to the rate of change of the Z-score value, determine the dynamic trend of the fiber bundle abnormality; Combine the dynamic trend of each fiber bundle with the corresponding Z-score value, fiber bundle name and anatomical location to generate a dynamic tracking report of fiber bundle abnormalities; In the dynamic tracking report, list the Z-score value, rate of change, dynamic trend and corresponding anatomical location of each fiber bundle at each time node, and show the trend of Z-score value change over time through a line chart; According to the dynamic trend analysis result, generate a prediction model of abnormal fiber bundles, which is based on time series analysis and machine learning algorithm, combining historical Z-score values and rates of change to predict the abnormality degree of fiber bundles in the future time node; Combine the prediction results with the dynamic tracking report to generate a comprehensive trend analysis report, which includes the current abnormality degree, historical trend and future prediction result of each fiber bundle; The comprehensive trend analysis report is stored in the patient database and is associated with the personal information and historical fiber bundle parameters of the patient, thereby providing data support for long-term dynamic monitoring of white matter abnormalities of Parkinson's disease and clinical intervention.

[0018] Compared with the prior art, the present application has the following advantages: In the present application, by setting the first threshold value and the second threshold value, not only can the abnormal fiber bundle be identified, but also the required threshold range can be selected according to different research purposes and clinical needs, thereby ensuring that reliable judgment results can be provided in both diagnosis and early screening; if both threshold values are selected for use, the method can comprehensively judge each parameter of each fiber bundle to form a more comprehensive conclusion, thereby more accurately detecting the white matter abnormal fiber bundle of the Parkinson's disease patient, improving the accuracy of the fiber bundle labeling result of the white matter abnormality of Parkinson's disease, and realizing the function of high-precision positioning of the fiber bundle of the white matter abnormality of Parkinson's disease, thereby providing stronger support for clinical diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The flow chart of the Parkinson's disease white matter abnormality fiber bundle automatic segmentation and quantitative labeling method of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0021] To solve the problem in the prior art that the detection of white matter abnormalities relies on manual delineation or traditional machine learning methods, thereby resulting in low efficiency and strong subjectivity, please refer to Figure 1 The present embodiment provides the following technical solutions: The Parkinson's disease white matter abnormality fiber bundle automatic segmentation and quantitative labeling method comprises the following steps: S1, collect diffusion magnetic resonance image data with normal fiber bundles, and unify all the data to the same anatomical space to form a reference database; the conditions include: no history of nervous system diseases, no serious systemic diseases, age / gender matching with the target Parkinson's disease patient group, for example: if the average age of the target patient group is 65 years old, the gender ratio is 50% male and 50% female, then the healthy control group should also match this ratio as much as possible; the specific steps include: The Z-score value of each fiber bundle parameter in the reference database is traversed, and is compared with the first threshold value and the second threshold value respectively to obtain the judgment result of each fiber bundle parameter; obtain data samples with Z-score values of the fiber bundle parameters in the reference database beyond the first threshold range as a first data set; obtain data samples with Z-score values of the fiber bundle parameters in the reference database beyond the second threshold range but not beyond the first threshold range as a second data set; create two classification storage structures in the reference database, including: a first threshold storage structure and a second threshold storage structure; match the first data set and the second data set to corresponding data sample cache areas respectively, and label the information of each fiber bundle parameter stored, including: fiber bundle name, parameter type, Z-score value, corresponding threshold category, and other related information such as age, gender, etc.; set the update frequency of the reference database according to research needs and actual data collection situations; for example: update every half year or every year; use the Euclidean distance method to calculate the similarity between new data samples and historical data samples; calculate the repetition rate of new data samples and historical data samples according to the similarity measurement results; for example: if the similarity between a new sample and a historical sample exceeds a preset threshold (such as 95%), the two samples are considered to be repeated; identify data samples with high repetition rates and delete historical data samples with high repetition rates from the reference database to avoid data redundancy; store new data samples in the corresponding cache area to form an updated reference database.

[0022] The beneficial effects achieved by the above content are: by classifying and storing fiber bundle parameters according to comparison results, the fiber bundle parameters in the reference database can be classified in detail, so that the required fiber bundle parameter information can be quickly indexed in actual application; not only helps to improve the accuracy of diagnosis, but also provides clear, systematic and comprehensive data support for subsequent analysis and research, greatly improving the usability and research efficiency of fiber bundle parameters in the reference database.

[0023] S2, extract part of the data of the reference database as a healthy control group; wherein the sampling data is not less than 30 samples, ensuring that the extracted samples are representative in key variables such as age and gender, and are matched with the target Parkinson's disease patient population.

[0024] S3, automatically segmenting the fiber bundle by the TractSeg algorithm to extract the main fiber bundle; calculating the fiber bundle parameters of the healthy control group using the fiber tracking algorithm, including fiber density, fiber cross-section, and fiber density-cross-section product; and statistically analyzing the fiber bundle parameters of each individual in the healthy control group to calculate the mean and standard deviation, including: obtaining the same fiber bundle parameter value of each individual in all healthy control groups, adding the same fiber bundle parameter values of all individuals, and then dividing by the number of individuals to obtain the mean value of the parameter; calculating the square of the difference between each parameter value and the mean value, summing and dividing by the number of individuals, and then taking the square root to obtain the standard deviation of the parameter.

[0025] S4, Z-score standardization of the fiber bundle parameters of the healthy control group for detecting abnormal fiber bundles; S5, collecting the diffusion magnetic resonance image data of the target Parkinson's disease patient, pre-processing the diffusion magnetic resonance image data, including head motion correction, noise reduction, distortion correction, and bias field correction to improve image quality, and calculating the fiber bundle parameters of the target Parkinson's disease patient using the S3 step, including: Obtaining the pre-processed diffusion magnetic resonance image data of the target Parkinson's disease patient and matching the corresponding healthy control group, and obtaining the fiber tracking algorithm parameters of the healthy control group, such as the starting threshold of fiber tracking and the step size; Adjusting the fiber tracking algorithm parameters of the healthy control group, fitting the tensor model to the diffusion magnetic resonance image data to obtain the diffusion tensor parameters of each voxel, including anisotropy fraction and radial diffusion rate; In the brain atlas of the standard space (such as MNI space), define the target area of the fiber bundle to be studied; for example, to extract the left corticospinal tract, define the target area at the posterior limb of the internal capsule and the cerebral peduncle where it passes through; Starting from the seed point of interest in the target area, tracking along the main diffusion direction until the termination condition is met, such as the FA value being lower than a certain threshold; during the tracking process, the voxel positions on the fiber path are recorded in real time; select the region of interest, such as the corpus callosum and the internal capsule, and count the number of fiber paths passing through the region, and divide by the volume of the region to obtain the fiber density value; Select the fiber path, count the number of voxels intersecting the fiber path, and multiply by the volume of the voxel to obtain the fiber cross-section value; multiply the fiber density value by the fiber cross-section value to obtain the fiber density-cross-section product; record the personal information of the target Parkinson's disease patient, and store the calculated fiber bundle parameters in the personal information, including patient number, fiber bundle name, and parameter value, for subsequent analysis.

[0026] S6, compare the fiber bundle parameters of the target Parkinson's disease patient with the Z-score standardized database of the healthy control group, calculate the Z-score value of each fiber bundle parameter, and label the fiber bundle whose Z-score value exceeds the preset threshold range as an abnormal fiber bundle, including: Extract data samples from the Z-score standardized database of the healthy control group, and analyze the Z-score distribution of different fiber bundle parameters; According to the statistical analysis result, set an initial threshold value; for example, according to the 95% confidence interval of the normal distribution, the initial threshold value is preliminarily set as |Z|>2.0; According to the initial threshold value, combined with the research purpose and clinical demand, a first threshold value and a second threshold value are respectively set, the first threshold value is used to reduce false positive detection, for example: |Z|>2.0, the second threshold value is used to improve the sensitivity of detection, for example: |Z|>1.96; if the research purpose is to reduce false positive results, and is a confirmed state, the first threshold value is selected; if the research purpose is to improve the sensitivity of diagnosis, and is early screening, the second threshold value is selected; if the clinical demand needs to comprehensively judge the multiple fiber bundle parameters of the patient, both the first threshold value and the second threshold value are selected.

[0027] Obtain the fiber bundle parameters of the Parkinson's disease patient, and the research purpose and clinical demand, and determine to select the first threshold value, the second threshold value or both according to the research purpose and clinical demand; When the first threshold value is selected, if the Z-score value of any one parameter of any one fiber bundle exceeds the first threshold value, the parameter of the fiber bundle is labeled as high confidence abnormal; for example, if the research purpose is to diagnose Parkinson's disease, and the Z-score value of the FA parameter of a certain fiber bundle of the patient is-2.5, which is less than the first threshold value of-2.0, it is labeled as high confidence abnormal; When the second threshold value is selected, if the Z-score value of any one parameter of any one fiber bundle exceeds the second threshold value but does not exceed the first threshold value, the parameter of the fiber bundle is labeled as low confidence abnormal; for example, if the research purpose is to screen Parkinson's disease early, and the Z-score value of the RD parameter of a certain fiber bundle of the patient is +1.98, but to further determine the accuracy of parameter comparison, due to random variation or measurement error, the parameter of the patient is further compared with the first threshold value, and the result is greater than the second threshold value of +1.96 but less than the first threshold value of +2.0, which is labeled as low confidence abnormal; When both are selected, the parameters of each fiber bundle are compared with the first threshold value and the second threshold value respectively to obtain the comprehensive judgment result of each parameter of each fiber bundle; for example: if the Z-score value of the FA parameter (anisotropy fraction) of a certain fiber bundle is -2.5 (high confidence abnormality), and the Z-score value of the RD parameter (radial diffusion rate) is +2.8 (low confidence abnormality), it can be concluded that the FA of the fiber bundle is significantly reduced and the RD is significantly increased, suggesting the presence of typical demyelination; for another example: if the Z-score value of the FA parameter of a certain fiber bundle is -1.98 (low confidence abnormality), and the Z-score value of the RD parameter is +2.05 (high confidence abnormality), it can be concluded that the FA of the fiber bundle may be reduced and the RD is significantly increased, suggesting that there may be demyelination; thus more accurate pathological diagnosis information can be provided to help clinicians better understand the patient's condition.

[0028] The above-mentioned beneficial effects: by setting the first threshold value and the second threshold value, not only can abnormal fiber bundles be identified, but also the required threshold range can be selected according to different research purposes and clinical needs, thereby ensuring that reliable judgment results can be provided in both confirmed diagnosis and early screening; if both thresholds are selected for simultaneous use, the method can make comprehensive judgments on each parameter of each fiber bundle to form a more comprehensive conclusion, thereby more accurately detecting white matter abnormal fiber bundles in Parkinson's disease patients, improving the accuracy of Parkinson's disease white matter abnormal fiber bundle labeling results, and achieving high-precision positioning of Parkinson's disease white matter abnormal fiber bundles to provide stronger support for clinical diagnosis.

[0029] S7、According to the location and characteristics of the abnormal fiber bundle, it is labeled as a specific fiber bundle related to Parkinson's disease, such as corpus callosum, internal capsule, external capsule, etc. The Z-score value of the abnormal fiber bundle is mapped to the heat map using a visualization tool, and detailed labeling information is generated for each abnormal fiber bundle, including fiber bundle name, abnormality degree (Z-score value), location, etc. The specific steps include: Obtain the fiber path obtained by the fiber tracking algorithm, and clearly label the name of the abnormal fiber bundle, such as "corpus callosum", "internal capsule", etc.; according to the name of the abnormal fiber bundle, determine the specific location of the abnormal fiber bundle in the brain; for example: the corpus callosum connects the main fiber bundle of the two hemispheres of the brain; the internal capsule is located between the basal ganglia and the thalamus, and contains important ascending and descending fibers; the external capsule is located outside the internal capsule and contains fibers connecting the cortex and basal ganglia; the corticospinal tract is a motor fiber bundle from the cerebral cortex to the spinal cord; the corticopontine tract is a fiber bundle from the cerebral cortex to the pons; According to the fiber density, fiber cross-section, and Z-score value of the fiber density-cross-section product of the fiber bundle, the abnormality degree of the fiber bundle is determined; for example: the FA parameter Z-score value of the left corticospinal tract is detected to be -2.5 (high confidence abnormality), and the RD parameter Z-score value is +2.8 (high confidence abnormality); A three-dimensional brain template image is created, and the Z-score value of each fiber bundle is mapped to the corresponding position; the Z-score value of the abnormal fiber bundle is converted into a specific color, and a heat map is generated; in the heat map, the name, Z-score value and corresponding position information of each abnormal fiber bundle are labeled; the labeling information includes: fiber bundle name: left corticospinal tract; abnormality degree: FA significantly decreased (Z=-2.5), RD significantly increased (Z=+2.8); position: left cerebral hemisphere, from the precentral gyrus to the medulla oblongata; According to the abnormality of the fiber bundle parameters, a comprehensive conclusion is generated; and in the heat map, the abnormality degree and distribution of the abnormal fiber bundle are marked by different colors; the generated heat map is inserted into the diagnosis report, and the labeling information of each abnormal fiber bundle is listed for clinical use.

[0030] Working principle: By collecting diffusion magnetic resonance image data of healthy controls, reliable benchmarks are provided for subsequent anomaly detection; secondly, the TractSeg algorithm is used for automatic segmentation and parameter calculation of fiber bundles, ensuring the accuracy and efficiency of fiber bundle extraction; through Z-score standardization processing and double threshold comparison mechanism, the sensitivity and specificity of anomaly detection can be flexibly adjusted according to research purposes and clinical needs, effectively distinguishing high and low confidence abnormal fiber bundles; in addition, the Z-score value of the abnormal fiber bundle is mapped to the heat map, and detailed labeling information is generated, providing intuitive and comprehensive visualization results for clinicians and researchers, which helps to quickly understand and apply detection results, significantly improves the accuracy and efficiency of Parkinson's disease diagnosis, and realizes the function of high-precision positioning of Parkinson's disease white matter abnormalities, providing strong support for early screening and diagnosis.

[0031] The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease also includes: Extracting fiber density, fiber cross-section and fiber density-cross-section product from the fiber bundle parameters of the target Parkinson's disease patient; Based on the fiber density, fiber cross-section and fiber density-cross-section product, the fiber bundle abnormality index is calculated, and the calculation formula is as follows:

[0032] wherein A is the fiber bundle abnormality degree index, ranging from 0 to 1, the higher the value, the higher the degree of fiber bundle abnormality; FD is the fiber density parameter, representing the number of fibers per unit volume in the fiber bundle of the target Parkinson's disease patient calculated by the S5 step, with units of per cubic millimeter, reflecting the density of the fiber bundle; FCS is the fiber cross-section parameter, representing the average cross-sectional area of the fiber bundle path calculated by the S5 step, with units of square millimeters, reflecting the spatial distribution characteristics of the fiber bundle; FDP is the fiber density-cross-section product parameter, representing the product of FD and FCS, with units of dimensionless, reflecting the comprehensive structural characteristics of the fiber bundle; w1, w2, w3 are weight coefficients, corresponding to the weighted influence of fiber density, fiber cross-section, and fiber density-cross-section product, respectively, ranging from 0 to 1, and satisfying w1+w2+w3=1, determined by statistical analysis of fiber bundle parameters of healthy control group and machine learning model training; Q is the data quality factor, representing the signal-to-noise ratio score of the diffusion magnetic resonance image data of the target Parkinson's disease patient, ranging from 0 to 10, the higher the value, the higher the data quality, calculated by weighting the image clarity and stereoscopic microscope calibration parameters evaluated in the preprocessing step; exp(-Q) is an exponential function, used to adjust the nonlinear influence of the data quality factor on the fiber bundle abnormality degree index, ensuring that the sensitivity of the abnormality degree index is reduced when the data quality is poor; According to the calculated fiber bundle abnormality degree index, the mapping relationship between the pre-set abnormality degree index and abnormality level is checked to determine the abnormality level of each fiber bundle; The determined abnormality level is combined with the corresponding fiber bundle name, location information, and Z-score value to generate a dynamic assessment report of fiber bundle abnormality degree; in the dynamic assessment report, the abnormality degree index, abnormality level, fiber bundle name, Z-score value, and corresponding anatomical location of each fiber bundle are listed; The abnormality degree index in the dynamic assessment report is mapped to a three-dimensional brain template image; The generated dynamic assessment report and three-dimensional brain template image are delivered to a visualization tool for clinical diagnosis and subsequent research; The dynamic assessment report is stored in the patient database, associated with the patient's personal information and historical fiber bundle parameters, providing data support for long-term tracking of the progression of Parkinson's disease white matter abnormalities.

[0033] Working principle: By extracting fiber density (FD), fiber cross-section (FCS) and fiber density-cross-section product (FDP) from the fiber bundle parameters of the target Parkinson's disease patient, and combining the data quality factor (Q) to calculate the fiber bundle abnormality degree index (A), the quantitative evaluation of the fiber bundle abnormality degree is realized. Fiber density (FD) is defined as the number of fibers per unit volume (unit: per cubic millimeter), which is calculated by the fiber tracking algorithm in step S5. Specifically, the number of fiber paths passing through the region is counted and divided by the volume of the region along the main diffusion direction from the seed point of the region of interest. Fiber cross-section (FCS) is defined as the average cross-sectional area of the fiber bundle path (unit: square millimeter), which is obtained by counting the number of voxels intersecting the fiber path and multiplying the voxel volume. Fiber density-cross-section product (FDP) is the product of FD and FCS (dimensionless), which reflects the comprehensive structural characteristics of the fiber bundle. The weight coefficients w1, w2, w3 (range 0 to 1, satisfying w1+w2+w3=1) are determined by statistical analysis of the fiber bundle parameters of the healthy control group and machine learning model (such as support vector machine or random forest) training. The specific steps are as follows: Extract the statistical distribution of FD, FCS, FDP from the healthy control group database, calculate the contribution rate of each parameter to abnormal classification (such as through feature importance analysis), and optimize the weight allocation through cross-validation. The data quality factor (Q, range 0 to 10) is calculated by the signal-to-noise ratio score in the preprocessing step and the stereomicroscope calibration parameter, specifically: the signal-to-noise ratio is obtained by image gray gradient variance analysis, the stereomicroscope calibration parameter is calculated by comparing the pixel deviation of the image and the standard template, and the two are weighted and summed according to the preset ratio (such as 0.6:0.4). The abnormality degree index A is calculated by the formula , where exp(-Q) is an exponential function used to non-linearly adjust the influence of Q on A, ensuring that the sensitivity of A is reduced when the data quality is low. According to the A value (range 0 to 1), the preset mapping relationship (such as A<0.3 is normal, 0.3≤A≤0.6 is mild abnormality, 0.6≤A≤0.8 is moderate abnormality, and A≥0.8 is severe abnormality) is determined. The abnormality level. Finally, the A value, abnormality level, fiber bundle name, Z-score value and anatomical location (such as corpus callosum, internal capsule) are integrated into a dynamic evaluation report, and the A value is mapped to a three-dimensional brain template image using visualization tools (such as FSLView or MRtrix) to generate a heat map, which is stored in the patient database associated with personal information (such as patient number, age).

[0034] Taking the left corticospinal tract of a 65-year-old male Parkinson's disease patient as an example, assuming that the FD calculated in step S5 is 120 per cubic millimeter, the FCS is 2.5 square millimeters, and the FDP is 300 (FD x FCS). Through statistical analysis of the healthy control group (30 age- and gender-matched healthy individuals), the weight coefficients are determined as w1=0.4, w2=0.3, and w3=0.3, based on the random forest model training, in which the feature importance of FD, FCS, and FDP is 40%, 30%, and 30%, respectively. The pre-processing step evaluates the signal-to-noise ratio score as 8 (calculated by the gray level gradient variance), and the stereomicroscope calibration deviation is 0.2 millimeters. The two are weighted by 0.6:0.4 to obtain Q=8.2. Substituting the formula A=(0.4x120+0.3x2.5+0.3x300) / [1+exp(-8.2)], A≈0.65 is calculated, which is mapped to the moderate abnormality level. The dynamic assessment report is generated as follows: the fiber bundle name is "left corticospinal tract", the abnormality degree index A=0.65, the abnormality level is moderate, the Z-score value (obtained from step S6) is FA=-2.5 and RD=+2.8, and the anatomical location is "left cerebral hemisphere, from the precentral gyrus to the medulla oblongata". By using the FSLView tool, A=0.65 is mapped to orange and labeled on the left corticospinal tract area of the three-dimensional MNI space template to generate a heat map showing the abnormality degree and location. The report is stored in the patient database, associated with the patient number PD001 and personal information (such as age 65 years old, male), facilitating subsequent clinical diagnosis and long-term tracking.

[0035] By calculating the fiber bundle abnormality degree index and generating a dynamic assessment report, this method realizes the quantitative evaluation and visual presentation of white matter abnormalities in Parkinson's disease. Combined with the weighted calculation of FD, FCS, FDP, and data quality factor Q, the abnormality degree index can comprehensively reflect the structural characteristics of the fiber bundle, and reduce the risk of false judgment when the data quality is poor through nonlinear adjustment. The dynamic assessment report integrates multi-dimensional information (abnormality degree, Z-score value, anatomical location), and visually displays the abnormality distribution through a heat map, providing accurate diagnostic evidence for clinicians, significantly improving the accuracy and efficiency of white matter abnormality detection in Parkinson's disease, and supporting early screening and long-term disease monitoring.

[0036] Further, the method for automatic segmentation and quantitative labeling of fiber bundles of white matter abnormalities in Parkinson's disease further comprises: extracting diffusion magnetic resonance image data collected at consecutive time points from the database of the target Parkinson's disease patient, including fiber bundle parameter data at at least three time nodes; Preprocess the fiber bundle parameter data for each time node, ensure all data are unified to the same anatomical space, and use the TractSeg algorithm and fiber tracking algorithm in the S3 step to calculate the fiber density, fiber cross-section, and fiber density-cross-section product for each time node, respectively; Compare the fiber bundle parameters of each time node with the Z-score standardized database of the healthy control group, and calculate the Z-score value of each fiber bundle parameter at each time node; According to the Z-score value, identify the abnormal fiber bundle that exceeds the preset threshold range at each time node, and record its name, anatomical location and abnormality degree; Time series analysis is performed on the Z-score values of the same fiber bundle at different time nodes, and the rate of change of the Z-score value is calculated, which is calculated by the difference between the Z-score values of adjacent time nodes divided by the time interval, with the unit of Z-score change per month; According to the rate of change of the Z-score value, determine the dynamic trend of the fiber bundle abnormality; Combine the dynamic trend of each fiber bundle with the corresponding Z-score value, fiber bundle name and anatomical location to generate a dynamic tracking report of fiber bundle abnormalities; In the dynamic tracking report, list the Z-score value, rate of change, dynamic trend and corresponding anatomical location of each fiber bundle at each time node, and show the trend of Z-score value change over time through a line chart; According to the results of dynamic trend analysis, generate a prediction model of abnormal fiber bundles, which is based on time series analysis and machine learning algorithms, combining historical Z-score values and rates of change to predict the abnormality degree of fiber bundles in the future time node; Combine the prediction results with the dynamic tracking report to generate a comprehensive trend analysis report, which includes the current abnormality degree, historical trend and future prediction results of each fiber bundle; Store the comprehensive trend analysis report in the patient database, associate it with the patient's personal information and historical fiber bundle parameters, and provide data support for long-term dynamic monitoring and clinical intervention of Parkinson's disease white matter abnormalities.

[0037] Working principle: Extract diffusion magnetic resonance image data of target Parkinson's disease patients at at least three time nodes, calculate fiber bundle parameters (FD, FCS, FDP) at each time node, and compare with Z-score standardized database of healthy control group to generate Z-score value, identify abnormal fiber bundle and its dynamic trend. Time node data is extracted from patient database, which needs to include diffusion magnetic resonance image at at least three time points (such as 6 month interval), preprocessing steps (head motion correction, noise reduction, distortion correction, bias field correction) are realized by FSL tool, and ensure that the data is unified to MNI standard anatomical space. Using S3 step TractSeg algorithm and fiber tracking algorithm, calculate FD (number of fibers per cubic millimeter, track along the main diffusion direction and divide by the volume of the region), FCS (square millimeter, number of statistical voxels multiplied by voxel volume) and FDP (FDxFCS) at each time node. Z-score value is calculated by comparing patient parameters with mean and standard deviation of healthy control group, formula is Z=(x-μ) / σ, where x is patient parameter, μ and σ are mean and standard deviation of healthy control group. Abnormal fiber bundle is identified by comparing Z-score value with first threshold (|Z|>2.0) and second threshold (|Z|>1.96), record name (such as corpus callosum), anatomical location (such as interthalamic adhesion) and abnormality degree (high / low confidence). Time series analysis calculates Z-score change rate, formula is ΔZ / Δt=(Zt2-Zt1) / (t2-t1), unit is Z-score change per month, reflecting abnormal progression speed. Dynamic trend is determined according to the change rate (such as ΔZ / Δt>0.1 for deterioration, -0.1≤ΔZ / Δt≤0.1 for stability, ΔZ / Δt<-0.1 for improvement). Based on historical Z-score value and change rate, use machine learning algorithm (such as ARIMA or LSTM) to train prediction model, predict Z-score value of next time node, specific steps are: take historical Z-score value and change rate as input, optimize model parameters (such as LSTM hidden layer unit number), evaluate prediction accuracy by cross-validation. Comprehensive trend analysis report integrates current abnormality degree (Z-score value), historical change rate, dynamic trend and prediction result, shows Z-score change with time through line chart, stores to patient database, associated with personal information (such as patient number, gender).

[0038] Take the same 65-year-old male Parkinson's disease patient as an example, extract his diffusion magnetic resonance image data at three time nodes of March 2024, September 2024, and March 2025. Preprocessing uses FSL tools for head motion correction (based on rigid registration), noise reduction (wavelet denoising), distortion correction (TOPUP algorithm), and bias field correction (N4 algorithm), and unifies to MNI space. For the left corticospinal tract, the TractSeg algorithm segments the fiber bundle, and the fiber tracking algorithm (step size 0.5 mm, FA termination threshold 0.1) calculates: T1 (March 2024) FD=120, FCS=2.5, FDP=300; T2 (September 2024) FD=115, FCS=2.4, FDP=276; T3 (March 2025) FD=110, FCS=2.3, FDP=253. Z-score values are calculated by comparing with the healthy control group (μFD=130, σFD=10, μFCS=2.8, σFCS=0.2, μFDP=364, σFDP=30), resulting in T1: ZFD=-1.0, ZFCS=-1.5, ZFDP=-2.13; T2: ZFD=-1.5, ZFCS=-2.0, ZFDP=-2.93; T3: ZFD=-2.0, ZFCS=-2.5, ZFDP=-3.7. ZFDP exceeds the first threshold (|Z|>2.0) at T2 and T3, and is marked as high confidence abnormality. The change rate is calculated as: T1-T2 (6 months) ΔZFDP=(-2.93-(-2.13)) / 6=-0.133, T2-T3 ΔZFDP=-0.123, average change rate-0.128, and is determined to be a deterioration trend. Using the LSTM model (hidden layer 100 units, trained for 100 rounds), the Z-score value and the change rate are used to predict T4 (September 2025) ZFDP≈-4.0. The comprehensive trend analysis report lists: fiber bundle name "left corticospinal tract", T1-T3 ZFDP values (-2.13, -2.93, -3.7), change rate (-0.128 / month), trend (deterioration), predicted ZFDP=-4.0, and location "left cerebral hemisphere, from the precentral to the medulla oblongata". The line chart is generated by Matplotlib, showing the downward trend of ZFDP over time. The report is stored in the patient database, associated with patient number PD001, supporting long-term monitoring and clinical intervention.

[0039] Through multi-time point fiber bundle parameter analysis and time series modeling, the method realizes the dynamic tracking and future trend prediction of white matter abnormalities in Parkinson's disease. The rate of Z-score change and dynamic trend analysis reveals the abnormal progression rate, and the prediction model provides a quantitative estimate of the future abnormality degree, providing a basis for the selection of clinical intervention timing. The comprehensive trend analysis report and the broken line chart intuitively present the dynamic changes of fiber bundle abnormalities, and the combination of patient personal information is stored in the database to support long-term disease monitoring and personalized treatment plan formulation, which significantly improves the accuracy and forward-looking of white matter abnormality diagnosis in Parkinson's disease.

[0040] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any actual relationship or order between or among such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0041] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, alternatives, and variations can be made in the embodiments without departing from the spirit and scope of the present application as defined by the appended claims and their equivalents.

Claims

1. A method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease, characterized in that, Includes the following steps: S1. Collect diffusion magnetic resonance imaging data with normal fiber bundles and unify all data into the same anatomical space to form a reference database; S2. Extract a portion of the data from the reference database to serve as a healthy control group; S3. Automatically segment fiber bundles using the TractSeg algorithm and extract the main fiber bundles; The fiber bundle parameters of the healthy control group were calculated using a fiber tracking algorithm. The fiber bundle parameters included fiber density, fiber cross-section, and fiber density-cross-section product. Statistical analysis was performed on the fiber bundle parameters of each individual in the healthy control group to calculate their mean and standard deviation. S4. The fiber bundle parameters of the healthy control group were standardized using Z-score to detect abnormal fiber bundles. S5. Collect diffusion magnetic resonance imaging data of the target Parkinson's disease patient, preprocess the diffusion magnetic resonance imaging data, and use step S3 to calculate the fiber bundle parameters of the target Parkinson's disease patient. S6. Compare the fiber bundle parameters of the target Parkinson's disease patients with the Z-score standardized database of the healthy control group, calculate the Z-score value of each fiber bundle parameter, and mark the fiber bundles whose Z-score values ​​exceed the preset threshold range as abnormal fiber bundles. S7. Based on the location and characteristics of the abnormal fiber bundles, label them as specific fiber bundles related to Parkinson's disease, use visualization tools to map the Z-score values ​​of the abnormal fiber bundles onto a heat map, and generate detailed labeling information for each abnormal fiber bundle.

2. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 1, characterized in that, In S6, fiber bundles with Z-score values ​​exceeding a preset threshold range are marked as abnormal fiber bundles, including: Data samples were extracted from the Z-score normalized database of the healthy control group to analyze the Z-score distribution of different fiber bundle parameters; Based on the statistical analysis results, set an initial threshold; Based on the initial threshold, and in conjunction with the research objectives and clinical needs, a first threshold and a second threshold are set respectively. The first threshold is used to reduce false positive detections, and the second threshold is used to improve the sensitivity of detection.

3. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 2, characterized in that, S6 further includes classifying fiber bundles with Z-score values ​​exceeding a preset threshold as abnormal fiber bundles, and also includes: Obtain fiber bundle parameters from Parkinson's disease patients, as well as research objectives and clinical needs, and determine the selection of the first threshold, the second threshold, or both based on research objectives and clinical needs; When selecting the first threshold, if the Z-score value of any parameter of any fiber bundle exceeds the first threshold, then that parameter of the fiber bundle is marked as a high-confidence anomaly. When selecting the second threshold, if the Z-score value of any parameter of any fiber bundle exceeds the second threshold but does not exceed the first threshold, then that parameter of the fiber bundle is marked as a low-confidence anomaly. When both are used, each parameter of each fiber bundle is compared with the first threshold and the second threshold respectively to obtain a comprehensive judgment result for each parameter of each fiber bundle.

4. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 2, characterized in that, S1 collects diffusion magnetic resonance imaging data with normal fiber bundles and unifies all data into the same anatomical space to form a reference database, including: The Z-score value of each fiber bundle parameter in the reference database is traversed and compared with the first threshold and the second threshold respectively to obtain the judgment result of each fiber bundle parameter. Obtain data samples from the reference database whose Z-score values ​​of fiber bundle parameters exceed the first threshold range, and use them as the first dataset; Data samples from the reference database whose Z-score values ​​of fiber bundle parameters exceed the second threshold range but do not exceed the first threshold range are used as the second dataset. Create two classification storage structures in the reference database, including: a first threshold storage structure and a second threshold storage structure; The first and second datasets are matched to their respective data sample cache areas, and the information of each stored fiber bundle parameter is labeled, including: fiber bundle name, parameter type, Z-score value, and corresponding threshold category; S1 collects diffusion magnetic resonance imaging data with normal fiber bundles and unifies all data into the same anatomical space to form a reference database, which also includes: The update frequency of the reference database should be set according to research needs and the actual situation of data collection. The Euclidean distance method is used to calculate the similarity between new data samples and historical data samples; Based on the similarity measurement results, calculate the overlap rate between the new data sample and the historical data sample; Identify data samples with high repetition rates and delete historical data samples with high repetition rates with new data samples from the reference database. New data samples are stored in the corresponding cache area to form an updated reference database.

5. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 1, characterized in that, S5 uses step S3 to calculate fiber bundle parameters for the target Parkinson's disease patient, including: Acquire preprocessed diffusion magnetic resonance imaging data of target Parkinson's disease patients, match them with corresponding healthy control groups, and obtain fiber tracking algorithm parameters for the healthy control groups; The fiber tracking algorithm parameters of the healthy control group were adjusted, and a tensor model was fitted to its diffusion magnetic resonance image data to obtain the diffusion tensor parameters of each voxel, including: anisotropy fraction and radial diffusion rate. In a standard spatial brain atlas, the target region of the fiber tract to be studied is defined; Starting from the seed point of interest in the target region, trace along the main diffusion direction until the termination condition is met; During the tracking process, the voxel positions along the fiber path are recorded in real time.

6. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 5, characterized in that, S5 uses step S3 to calculate fiber bundle parameters for target Parkinson's disease patients, and also includes: Select the region of interest, count the number of fiber paths passing through the region, and divide by the volume of the region to obtain the fiber density value; Select a fiber path, count the number of voxels that intersect with the fiber path, and multiply the number by the volume of the voxels to obtain the fiber cross-section value. Multiply the fiber density value by the fiber cross-sectional area value to obtain the fiber density-cross-sectional area product; Record the personal information of the target Parkinson's disease patient and store the calculated fiber bundle parameters in the personal information.

7. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 1, characterized in that, Based on the location and characteristics of abnormal fiber bundles, S7 identifies them as specific fiber bundles associated with Parkinson's disease, including: Obtain the fiber path obtained through the fiber tracing algorithm and clearly label the names of abnormal fiber bundles; Determine the specific location of the abnormal fiber bundle in the brain based on its name; The degree of abnormality of the fiber bundle is determined based on the fiber density, fiber cross-section, and Z-score of the product of fiber density and cross-section. Create a 3D brain template image and map the Z-score value of each fiber bundle to its corresponding location; The Z-score values ​​of abnormal fiber bundles are converted into specific colors, and a heatmap is generated. In the heat map, the name, Z-score value, and corresponding location information of each abnormal fiber bundle are labeled; Based on the anomalies in fiber bundle parameters, a comprehensive conclusion is generated; Based on the anomalies in fiber bundle parameters, a comprehensive conclusion is generated, including: In the heat map, different colors are used to mark the degree and distribution of abnormal fiber bundles; Insert the generated heatmap into the diagnostic report and list the annotation information for each abnormal fiber bundle for clinical use.

8. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 1, characterized in that, In S3, the fiber bundle parameters of each healthy control individual were statistically analyzed, and their mean and standard deviation were calculated, including: Obtain the same fiber bundle parameter value for each individual in all healthy control groups, add up the same fiber bundle parameter values ​​for all individuals, and then divide by the number of individuals to obtain the average value of the parameter. Calculate the square of the difference between each parameter value and the mean, sum them, divide by the number of individuals, and then take the square root to obtain the standard deviation of the parameter.

9. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 7, characterized in that, Also includes: Fiber density, fiber cross-section, and fiber density-cross-section product were extracted from fiber bundle parameters of target Parkinson's disease patients. The fiber bundle anomaly index is calculated based on fiber density, fiber cross-section, and the product of fiber density and cross-section. The calculation formula is as follows: Where A is the fiber bundle abnormality index, ranging from 0 to 1, with higher values ​​indicating greater fiber bundle abnormality; FD is the fiber density parameter, representing the number of fibers per unit volume within the fiber bundle of the target Parkinson's disease patient calculated through step S5, in millimeters per cubic meter, reflecting the density of the fiber bundle; FCS is the fiber cross-sectional area parameter, representing the average cross-sectional area of ​​the fiber bundle path calculated through step S5, in millimeters per square meter, reflecting the spatial distribution characteristics of the fiber bundle; FDP is the fiber density-cross-sectional area product parameter, representing the product of FD and FCS, in dimensionless units, reflecting the comprehensive structural characteristics of the fiber bundle; w1, w2, and w3 are weighting coefficients, respectively corresponding to... The weighted effects of fiber density, fiber cross-section, and the fiber density-cross-section product, ranging from 0 to 1 and satisfying w1+w2+w3=1, were determined by statistical analysis of fiber bundle parameters from healthy control groups and training of machine learning models. Q is the data quality factor, representing the signal-to-noise ratio score of the diffusion magnetic resonance imaging data of the target Parkinson's disease patients, ranging from 0 to 10. A higher value indicates higher data quality, and it is calculated by weighting the image sharpness and stereomicroscope calibration parameters assessed in the preprocessing step. exp(-Q) is an exponential function used to adjust the nonlinear effect of the data quality factor on the fiber bundle abnormality index, ensuring that the sensitivity of the abnormality index is reduced when the data quality is poor. Based on the calculated fiber bundle anomaly index, and by comparing it with the pre-set mapping relationship between the anomaly index and the anomaly level, the anomaly level of each fiber bundle is determined. The determined abnormality level is combined with the corresponding fiber bundle name, location information and Z-score value to generate a dynamic assessment report of the degree of fiber bundle abnormality; the dynamic assessment report lists the abnormality index, abnormality level, fiber bundle name, Z-score value and corresponding anatomical location of each fiber bundle. The abnormality index in the dynamic assessment report is mapped onto a three-dimensional brain template image; The generated dynamic assessment report and 3D brain template image are then transmitted to a visualization tool for clinical diagnosis and subsequent research. Storing dynamic assessment reports in a patient database and linking them to the patient's personal information and historical fiber bundle parameters provides data support for long-term tracking of the progression of white matter abnormalities in Parkinson's disease.

10. The method for automatic segmentation and quantitative labeling of fiber bundles for white matter abnormalities in Parkinson's disease according to claim 1, characterized in that, Also includes: Diffusion magnetic resonance imaging data collected at consecutive time points were extracted from the database of target Parkinson's disease patients, including fiber bundle parameter data at at least three time points; The fiber bundle parameter data at each time point are preprocessed to ensure that all data are unified to the same anatomical space. The TractSeg algorithm and fiber tracing algorithm in step S3 are used to calculate the fiber density, fiber cross-section and fiber density-cross-section product at each time point, respectively. The fiber bundle parameters at each time point were compared with the Z-score normalized database of the healthy control group, and the Z-score value of each fiber bundle parameter at each time point was calculated. Based on the Z-score, abnormal fiber bundles exceeding the preset threshold range at each time point are identified, and their names, anatomical locations, and degrees of abnormality are recorded. Time series analysis was performed on the Z-score values ​​of the same fiber bundle at different time points, and the rate of change of the Z-score values ​​was calculated. The rate of change was obtained by dividing the difference of the Z-score values ​​at adjacent time points by the time interval, and the unit was the monthly Z-score change. The dynamic trend of fiber bundle anomalies is determined based on the rate of change of Z-score values; The dynamic trend of each fiber bundle is combined with the corresponding Z-score value, fiber bundle name and anatomical location to generate a dynamic tracking report of fiber bundle abnormalities; The dynamic tracking report lists the Z-score value, rate of change, dynamic trend, and corresponding anatomical location of each fiber bundle at each time point, and displays the trend of Z-score value over time through a line graph. Based on the results of dynamic trend analysis, a predictive model for abnormal fiber bundles is generated. The predictive model is based on time series analysis and machine learning algorithms, combined with historical Z-score values ​​and rates of change, to predict the degree of abnormality of fiber bundles at a future time point. The prediction results are combined with the dynamic tracking report to generate a comprehensive trend analysis report, which includes the current degree of anomaly, historical trends and future predictions for each fiber bundle. The comprehensive trend analysis report is stored in the patient database and linked to the patient's personal information and historical fiber bundle parameters, providing data support for long-term dynamic monitoring and clinical intervention of white matter abnormalities in Parkinson's disease.

Images