Pd motor subtype classification method based on transcranial doppler and bayesian analysis

Abstract

The application discloses a PD movement subtype classification method based on transcranial Doppler and Bayesian analysis, and relates to the technical field of medical diagnosis, and comprises the following steps: S1, collecting the cerebral blood flow velocity signals of a subject under different physiological conditions through a transcranial Doppler device, and synchronously collecting the arterial blood pressure signals through a continuous noninvasive blood pressure monitoring device; S2, processing the signals collected in S1, and extracting characteristic indexes reflecting the dynamic cerebral blood flow automatic regulation function; S3, combining the characteristic indexes with basic hemodynamic parameters to form a characteristic vector; S4, inputting the characteristic vector into a pre-trained Bayesian discriminant model; and S5, according to the output of the Bayesian discriminant model, classifying the subject into different Parkinson's disease movement subtypes. The application has the advantages of improving classification objectivity, capturing dynamic physiological signal characteristics, and realizing precise subtype identification.

Description

Technical Field

[0001] This invention relates to the field of medical diagnostic technology, specifically to a method for classifying motor subtypes of PD based on transcranial Doppler and Bayesian analysis. Background Technology

[0002] Parkinson's disease (PD) is a complex neurodegenerative disease with significant clinical heterogeneity. Currently, clinical practice primarily relies on the Movement Disorders Association Unified Parkinson's Disease Rating Scale (MDS-UPDRS) for subtype classification. However, this method has significant limitations: firstly, the assessment process is time-consuming and dependent on physician subjective judgment, making it difficult to capture subtle changes in motor characteristics; secondly, the requirement for regular follow-up visits makes it difficult to track the dynamic evolution of patients' symptoms in a timely manner; and thirdly, existing methods cannot reflect underlying neurovascular dysfunction.

[0003] Recent studies have revealed characteristic cerebrovascular dysfunction in patients with different motor subtypes of Parkinson's disease. Dynamic cerebral autoregulation (dCA) function, a crucial mechanism for maintaining cerebral perfusion stability, is closely related to the pathological progression of Parkinson's disease. However, current technologies have not yet established effective objective indicators to characterize this association, resulting in a lack of reliable biomarkers in clinical practice to aid in subtype identification.

[0004] While transcranial Doppler (TCD) technology can non-invasively monitor cerebral hemodynamic parameters, traditional analysis methods struggle to effectively extract features related to motor subtypes. The dynamic relationship between blood pressure and cerebral blood flow velocity contains important physiological regulatory information, but current technologies have failed to fully exploit the potential classification features within these signals. Furthermore, conventional statistical methods suffer from insufficient model adaptability when processing multidimensional physiological parameters, failing to meet the accuracy requirements of clinical classification.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] In view of this, the present invention provides a PD motor subtype classification method based on transcranial Doppler and Bayesian analysis, which has the advantages of improving classification objectivity, capturing dynamic physiological signal characteristics, and achieving accurate subtype identification.

[0007] This invention provides a method for classifying motor subtypes of PD based on transcranial Doppler and Bayesian analysis, including:

[0008] S1. The transcranial Doppler device was used to collect cerebral blood flow velocity signals of the subjects under different physiological conditions, and the arterial blood pressure signals were collected simultaneously using a continuous non-invasive blood pressure monitoring device.

[0009] S2. Process the signals collected in S1 and extract characteristic indicators that reflect the dynamic cerebral blood flow autoregulation function.

[0010] S3. Combine the aforementioned feature indicators with the basic hemodynamic parameters to form a feature vector;

[0011] S4. Input the feature vector into the pre-trained Bayesian discriminant model;

[0012] S5. Based on the output of the Bayesian discriminant model, classify the subjects into different motor subtypes of Parkinson's disease.

[0013] In one optional implementation, S1 includes:

[0014] While the subject was in a resting state, cerebral blood flow velocity signals were continuously collected using a transcranial Doppler device for a first preset time period, and arterial blood pressure signals were simultaneously collected using a continuous non-invasive blood pressure monitoring device.

[0015] While the subject underwent postural changes, cerebral blood flow velocity and arterial blood pressure signals were collected during a second preset time period.

[0016] In one optional implementation, the acquisition conditions of the transcranial Doppler device are as follows:

[0017] The probe of a 1.6MHz transcranial Doppler ultrasound device was fixed to the bilateral temporal windows of the subject, with a detection depth of 45–60 mm;

[0018] Subjects were required to fast for ≥2 hours and discontinue autonomic nervous system medications for ≥12 hours prior to the data collection.

[0019] In one alternative implementation, S2 includes:

[0020] The time-domain signals of the cerebral blood flow velocity signal and the arterial blood pressure signal are converted into frequency-domain signals using Fourier transform;

[0021] Based on the converted frequency domain signal, the transfer function analysis method is used to calculate the frequency domain characteristic parameters between the cerebral blood flow velocity signal and the arterial blood pressure signal within one or more preset frequency bands; wherein, the frequency domain characteristic parameters include phase, gain and consistency function;

[0022] Simultaneously, based on the average frequency domain characteristic parameters of both hemispheres, the average correlation coefficient between average cerebral blood flow velocity and arterial blood pressure was calculated.

[0023] The frequency domain feature parameters and the average correlation coefficient are used together as feature indicators to characterize the dynamic cerebral blood flow autoregulation function.

[0024] In one alternative implementation, the baseline hemodynamic parameters are cerebral artery blood flow parameters directly obtained from raw cranial Doppler signals monitored by a transcranial Doppler device; the cerebral artery blood flow parameters include peak systolic velocity, end-diastolic velocity, pulsatility index, and resistance index.

[0025] In one alternative implementation, the Bayesian discriminant model is learned from training samples of known subtypes to obtain a classification function for distinguishing between tremor-dominant and postural instability-difficulty Parkinson's disease.

[0026] As can be seen from the above, the PD motor subtype classification method provided in this application, based on transcranial Doppler and Bayesian analysis, constructs a multidimensional feature vector by integrating dynamic cerebral blood flow autoregulation features and hemodynamic parameters, and achieves objective quantitative classification by combining a Bayesian discriminant model. It has the advantages of improving classification objectivity, capturing dynamic physiological signal features, and achieving accurate subtype identification. Attached Figure Description

[0027] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0028] Figure 1 This is a flowchart illustrating a method for classifying motor subtypes of PD based on transcranial Doppler and Bayesian analysis according to an embodiment of the present invention.

[0029] Figure 2 This is a Bayesian model discriminant analysis diagram according to an embodiment of the present invention;

[0030] Figure 3 The above are the results of the Bayesian model diagnostic performance analysis according to an embodiment of the present invention; wherein, (a) is a distribution diagram of the analysis values ​​of each research object feature in the dataset; (b) is an internal validation diagram of the model diagnostic performance; and (c) is an external validation diagram of the model diagnostic performance. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] In current technologies, the classification of motor subtypes of Parkinson's disease (PD) mainly relies on clinical scale assessments, which have limitations such as high subjectivity, long assessment cycles, and inability to track symptom changes in real time. Traditional methods depend on manual observation of patients' motor performance, which is easily affected by the assessor's experience and the patient's cooperation, and the scale scores cannot reflect differences in cerebral blood flow dynamic regulation function (dCA). Although imaging techniques can detect changes in cerebral blood flow, objective standards for association with motor subtypes have not yet been established, posing a technical bottleneck for clinical application.

[0033] To address the aforementioned issues, the inventors discovered a potential link between abnormal dynamic cerebral blood flow autoregulation and subtypes of Parkinson's disease, but existing technologies have not incorporated this into their classification systems. Analysis revealed that transcranial Doppler-acquired cerebral blood flow velocity (MCA) signals combined with blood pressure fluctuation data (BP) can quantify indicators of dynamic regulation function. Further consideration was given to how to transform multidimensional physiological signals into classification features and establish a classification model adapted to individual differences. Ultimately, a technical approach was proposed: extracting frequency domain features through signal processing, constructing feature vectors by combining them with basic blood flow parameters, and utilizing a Bayesian model to achieve probabilistic classification.

[0034] Therefore, as Figure 1 As shown, this invention provides a method for classifying motor subtypes of PD based on transcranial Doppler and Bayesian analysis, including:

[0035] Step S1: Collect cerebral blood flow velocity signals of the subject under different physiological conditions using a transcranial Doppler device, and simultaneously collect arterial blood pressure signals using a continuous non-invasive blood pressure monitoring device.

[0036] Step S2: Process the signals collected in step S1 and extract characteristic indicators that reflect the dynamic cerebral blood flow autoregulation function.

[0037] Step S3: Combine the feature indicators with the basic hemodynamic parameters to form a feature vector;

[0038] Step S4: Input the feature vector into the pre-trained Bayesian discriminant model;

[0039] Step S5: Based on the output of the Bayesian discriminant model, classify the subjects into different motor subtypes of Parkinson's disease.

[0040] Among them, signal acquisition under different physiological states refers to acquiring data in the resting state and the postural change-induced state, such as comparing signals in the lying state and the standing state, in order to capture the dynamic response of cerebrovascular regulation function.

[0041] The characteristic indicators of dynamic cerebral blood flow autoregulation function refer to frequency domain parameters obtained through transfer function analysis, such as phase and gain coefficients, which are used to characterize the coupling relationship between blood pressure fluctuations and cerebral blood flow velocity.

[0042] Basic hemodynamic parameters refer to time-domain parameters directly extracted from the raw signal, such as peak systolic velocity and pulsatility index, which are used to reflect the basic functional state of cerebrovascular system.

[0043] Bayesian discriminant models are classifiers built on probability and statistics theory. They achieve classification by calculating the posterior probability of a feature vector belonging to each subtype and can handle nonlinear relationships between parameters and individual differences.

[0044] Specifically, cerebral blood flow velocity and blood pressure signals were first collected simultaneously while the subjects were resting and changing positions to ensure spatiotemporal consistency of the data. Then, Fourier transform was used to convert the time-domain signals to the frequency domain, and transfer function analysis was used to calculate frequency-domain parameters such as phase and gain to quantify the dynamic regulation capacity of cerebrovascular systems. Simultaneously, baseline blood flow parameters were extracted as static features and combined with dynamic regulation parameters to form a multidimensional feature vector. This vector was input into a Bayesian model, which calculated the probability distribution of each subtype based on the classification boundaries established during the training phase, and finally output the classification result of the motor subtype corresponding to the highest probability.

[0045] Compared to existing technologies, current classification methods rely on manual assessment of motor symptoms and cannot obtain cerebrovascular function parameters, resulting in a single classification dimension. This proposed solution integrates dynamic regulatory function indicators and basic blood flow parameters to construct multidimensional classification features, overcoming the limitation of traditional scales that rely solely on motor manifestations. Furthermore, by employing a Bayesian probability model instead of traditional threshold judgments, it effectively handles individual variability and nonlinear characteristics of physiological parameters, significantly improving classification accuracy.

[0046] Through the above technical solutions, this application achieves a classification of Parkinson's disease motor subtypes based on objective physiological signals, eliminating the subjective bias of manual assessment. By integrating dynamic cerebral blood flow regulation parameters, it can timely reflect changes in the patient's cerebrovascular function, providing real-time classification basis for clinical practice. The use of a Bayesian probability model to process multidimensional feature data effectively solves the problem of sensitivity to parameter thresholds in traditional classification methods, improving the stability and accuracy of subtype discrimination.

[0047] In one optional implementation, step S1 includes:

[0048] While the subject was in a resting state, cerebral blood flow velocity signals were continuously collected using a transcranial Doppler device for a first preset time period, and arterial blood pressure signals were simultaneously collected using a continuous non-invasive blood pressure monitoring device.

[0049] While the subject underwent postural changes, cerebral blood flow velocity and arterial blood pressure signals were collected during a second preset time period.

[0050] The first preset time period refers to the continuous collection duration for acquiring basic physiological data in a resting state. Specifically, a time window of 3-5 minutes can be used. The design of this time period ensures the stability and integrity of cerebral blood flow velocity and blood pressure signals in a resting state, providing a reliable benchmark reference for subsequent analysis.

[0051] Among them, postural change induction refers to an intervention method that actively changes the subject's position to trigger a cerebral blood flow regulation response. Specifically, it can be achieved by changing from a supine position to an upright position. This operation can simulate scenarios that may cause fluctuations in cerebral blood flow in real life, thereby exposing potential defects in dynamic regulation function.

[0052] The second preset time period refers to the continuous acquisition duration for capturing dynamic adjustment responses after a change in body position. Specifically, a time window of 5-10 minutes can be used. This time period covers the key adjustment phase after a change in body position, ensuring that real-time data on the impact of blood pressure fluctuations on cerebral blood flow velocity are acquired.

[0053] Specifically, signal acquisition in the resting state establishes a baseline hemodynamic parameter database for the subject by simultaneously recording cerebral blood flow velocity and arterial blood pressure signals. Following induction of positional changes, the cerebral autoregulation system responds dynamically to blood pressure fluctuations. Continuously acquiring signals over a second preset time period captures real-time changes in regulatory function. For example, when a subject changes from a supine to an upright position, gravity causes a drop in blood pressure. At this time, the dynamic cerebral blood flow regulation mechanism maintains cerebral perfusion through vasoconstriction. Changes in parameters such as phase and gain during this process reflect differences in regulatory capacity. By combining data from the resting state and the dynamic regulation phase, potential random errors from data acquired in a single state can be eliminated, comprehensively covering the baseline level and stress response capacity of the cerebral autoregulation function.

[0054] Compared to existing technologies, traditional methods typically collect data only in a single resting state, failing to capture the dynamic regulatory processes induced by postural changes. This application, by introducing a postural change-induced and segmented data acquisition mechanism, not only acquires stable data under basal conditions but also actively induces physiological state changes to observe the real-time response of regulatory functions. For example, existing methods using continuous 10-minute monitoring in a supine position may miss regulatory defects caused by orthostatic hypotension, while this application, through data acquisition in a second time period after a postural change, can effectively identify such dynamic regulatory abnormalities, providing differentiated characteristic data for subtype classification.

[0055] Through the above technical solution, this application solves the problem of incomplete signal acquisition under different physiological states. By combining data acquisition from the resting state and the postural change phase, it ensures dual coverage of basic regulatory capacity and dynamic response characteristics. The design of the first preset time period avoids the deviation of baseline parameters caused by insufficient resting data acquisition time, while the setting of the second preset time period targets the key regulatory window after postural change, fully recording the impact of blood pressure fluctuations on cerebral blood flow velocity. The postural change-induced operation effectively exposes potential regulatory function defects by simulating real-life scenarios, enabling the classification model to distinguish different motor subtypes of Parkinson's disease based on more comprehensive dynamic features.

[0056] In one optional implementation, the acquisition conditions of the transcranial Doppler device are as follows:

[0057] The probe of a 1.6MHz transcranial Doppler ultrasound device was fixed to the bilateral temporal windows of the subject, with a detection depth of 45–60 mm;

[0058] Subjects were required to fast for ≥2 hours and discontinue autonomic nervous system medications for ≥12 hours prior to the data collection.

[0059] The 1.6MHz probe frequency refers to the oscillation frequency of the ultrasound transmitter. Specifically, a piezoelectric ceramic crystal can be used to convert electromagnetic energy into acoustic energy. This frequency strikes a balance between penetrating the skull and maintaining signal resolution. The bilateral temporal window fixation position refers to the area of ​​thinner squamous bone in the temporal bone. An elastic headband combined with medical coupling gel can be used to ensure probe adhesion. This position covers the anatomical projection area of ​​the middle cerebral artery. The detection depth of 45–60 mm refers to the propagation distance of the ultrasound beam in the tissue. This can be set via the echo reception time window on the device control panel. This range allows for precise capture of blood flow signals from the main trunk of the middle cerebral artery. Fasting for ≥2 hours means restricting the subject's intake of solid or liquid food. Standardized pre-experiment preparation procedures can be used to eliminate interference from blood flow redistribution caused by digestion. Discontinuation of autonomic nervous system drugs for ≥12 hours means suspending the use of drugs affecting sympathetic or parasympathetic nerve activity. Clinical drug management protocols can be followed. This measure avoids interference with vascular tone regulation.

[0060] Specifically, when the probe frequency is set to 1.6MHz, the energy attenuation of ultrasound waves penetrating the temporal bone is controlled within the range of 20-30dB, while maintaining a signal-to-noise ratio of Doppler shift signal above 6dB. When the probe is fixed to the bilateral temporal windows, the incident angle is adjusted using a three-dimensional positioning device to keep the angle between the ultrasound beam and the blood vessel course within 30°, ensuring that the error in blood flow velocity measurement is less than 10%. When the detection depth is set to 45-60mm, it can cover the blood flow signals from the M1 to M2 segments of the middle cerebral artery, and interference signals from perforating vessels are eliminated through spectral envelope analysis. During the fasting period, the subjects maintain their basal metabolic state, reducing the proportion of gastrointestinal blood flow from 25% postprandial to 15% at rest, thereby reducing the competitive influence of visceral circulation on cerebral hemodynamics. After discontinuation of autonomic nervous system drugs, the sensitivity of vascular smooth muscle cells to changes in carbon dioxide concentration returns to baseline levels, allowing the coupling relationship between blood pressure fluctuations and cerebral blood flow velocity to truly reflect the physiological regulatory mechanism.

[0061] Through the above technical solutions, this application solves the problems of baseline drift and spectral distortion in cerebral blood flow signals caused by equipment parameter mismatch or subject condition interference, ensuring the repeatability and physiological relevance of the time-frequency domain characteristics required for dynamic cerebral blood flow autoregulation function assessment. Standardized settings for probe fixation position and depth avoid operator-dependent errors, enabling spatial comparability of bilateral hemisphere blood flow signals. Controlled subject preparation conditions effectively isolate the modulation effect of external factors on autonomic nervous function, ensuring that the acquired blood pressure-blood flow coupling signals truly reflect the cerebrovascular regulatory capacity under pathological conditions, providing high-quality input data for subsequent subtype classification.

[0062] In one optional implementation, step S2 includes:

[0063] Fourier transform was used to convert the time-domain signals of cerebral blood flow velocity and arterial blood pressure signals into frequency-domain signals;

[0064] Based on the converted frequency domain signal, the transfer function analysis method is used to calculate the frequency domain characteristic parameters between the cerebral blood flow velocity signal and the arterial blood pressure signal in one or more preset frequency bands; wherein, the frequency domain characteristic parameters include gain, phase, and coherence function; wherein, gain includes normalized gain (Gain1(%)) and absolute gain (Gain2).

[0065] Simultaneously, based on the average frequency domain characteristic parameters of both hemispheres, the average correlation coefficient Mxa between the average cerebral blood flow velocity and arterial blood pressure was calculated.

[0066] The frequency domain characteristic parameters and the average correlation coefficient are used together as characteristic indicators to characterize the dynamic cerebral blood flow autoregulation function.

[0067] Fourier transform refers to converting time-domain physiological signals into frequency-domain signals using a fast Fourier transform algorithm, specifically discrete Fourier transform, to eliminate the influence of transient interference in the time-domain signal on feature extraction. Transfer function analysis calculates the amplitude-frequency and phase-frequency characteristics of cerebral blood flow velocity and blood pressure signals in the frequency domain, specifically using a complex domain transfer function model, to quantify the dynamic coupling relationship between the two signals in a specific frequency band. Phase refers to the time delay angle between blood pressure fluctuations and changes in cerebral blood flow velocity, specifically obtained through transfer function phase spectrum calculation, reflecting the temporal characteristics of signal transmission during dynamic regulation. Gain refers to the transmission efficiency of changes in cerebral blood flow velocity caused by changes in blood pressure signal amplitude, specifically obtained through transfer function amplitude spectrum calculation, characterizing the signal energy transmission characteristics during regulation. Consistency function refers to the degree of linear correlation between two signals in the frequency domain, specifically obtained through the ratio of cross-spectral density to self-spectral density, used to verify the reliability of the transfer function analysis results. The average value of the frequency domain characteristic parameters of both hemispheres refers to the arithmetic mean of the phase, gain, and consistency functions of the corresponding frequency bands of the left and right cerebral hemispheres. This can be achieved by simultaneously acquiring data from both probes, calculating the values ​​separately, and then averaging them. This method is used to eliminate individual physiological differences that may exist in unilateral measurements. The average correlation coefficient refers to the average value of the phase, gain, and consistency function parameters of both hemispheres after performing Pearson correlation analysis with the blood pressure signal. This can be achieved by calculating the dynamic correlation coefficient using the sliding window method and then averaging the results. This method is used to enhance the ability of characteristic indicators to represent the synergistic regulation of bilateral cerebral blood flow.

[0068] Specifically, by converting the original time-domain signal to the frequency domain using Fourier transform, frequency band components corresponding to different physiological rhythms can be effectively separated. For example, 0.07-0.20Hz is defined as the low-frequency band for analyzing autonomic nervous system regulation mechanisms. Transfer function analysis calculates three complementary parameters—phase, gain, and consistency function—within a preset frequency band. Phase reflects the time delay characteristic of blood pressure changes transmitted to cerebral blood flow, gain reflects the energy transfer efficiency during regulation, and the consistency function verifies the significance of the linear relationship between the two signals. Bilateral parameter averaging eliminates unilateral data bias caused by probe position offset or individual vascular anatomy differences. For example, averaging the phases of the left and right middle cerebral arteries forms a feature value with higher physiological reliability. The calculation of the average correlation coefficient further quantifies the dynamic correlation between frequency domain features and blood pressure signals in the time domain. For example, a 30-second sliding window is used to calculate the correlation coefficient, and the average is taken to form a supplementary indicator reflecting the stability of the regulation process. Finally, the frequency domain feature parameter set and the correlation coefficient together constitute a multi-dimensional feature vector, providing input data containing joint time-frequency domain features for subsequent classification models.

[0069] Compared to existing technologies, traditional PD subtype classification relies on manual observation of motor symptoms, while this approach establishes an objective physiological parameter evaluation system by quantitatively analyzing the dynamic coupling relationship between blood pressure and cerebral blood flow velocity. Existing methods often employ time-domain parameters such as the blood pressure coefficient of variation, while this approach innovatively combines frequency-domain features with bilateral synergistic parameters, enabling the simultaneous capture of the periodic characteristics and bilateral symmetry of the regulatory mechanism. Conventional techniques use only a single parameter to assess regulatory function, while this approach comprehensively characterizes the time-frequency characteristics and spatial distribution features of dynamic cerebral blood flow autoregulation through a multi-dimensional combination of phase, gain, consistency functions, and correlation coefficients.

[0070] Through the above technical solutions, this application achieves a quantitative assessment of the dynamic cerebral blood flow autoregulation function, solving the problem of traditional classification methods relying on subjective scales. The construction of multidimensional feature indicators can objectively reflect the differences in physiological regulatory mechanisms corresponding to different motor subtypes of PD patients, eliminating assessment biases that may be caused by manual observation. The techniques of frequency domain analysis and bilateral parameter fusion effectively improve the stability and discriminative power of feature indicators, providing high signal-to-noise ratio input data for machine learning models, and ultimately realizing automatic classification of PD subtypes based on objective physiological parameters.

[0071] In one alternative implementation, the baseline hemodynamic parameters are cerebral artery blood flow parameters directly obtained from raw cranial Doppler signals monitored by a transcranial Doppler device; the cerebral artery blood flow parameters include peak systolic velocity, end-diastolic velocity, pulsatility index, and resistance index.

[0072] The peak systolic velocity refers to the maximum blood flow velocity in cerebral vessels during cardiac systole, which can be identified by transcranial Doppler ultrasound of the blood flow velocity waveform, reflecting the maximum blood supply capacity of cerebral vessels. The end-diastolic velocity refers to the stable value of blood flow velocity in cerebral vessels at the end of cardiac diastole, which can be identified by transcranial Doppler ultrasound of the stable end-diastolic segment, characterizing the peripheral vascular circulation status. The pulsatility index is the ratio of the difference between the peak systolic velocity and the end-diastolic velocity to the average velocity, which can be calculated by measuring the dynamic range of the blood flow velocity waveform, reflecting vascular elasticity and peripheral resistance. The resistance index is the ratio of the difference between the peak systolic velocity and the end-diastolic velocity to the peak systolic velocity, which can be calculated by observing the relationship between the systolic and diastolic phases of the blood flow velocity waveform, quantifying blood flow resistance characteristics.

[0073] Specifically, after acquiring raw transcranial Doppler (TCD) signals, four parameters—peak systolic velocity, end-diastolic velocity, pulsatility index, and resistance index—are directly extracted. Peak systolic velocity is extracted by identifying the periodically occurring highest points in the blood flow velocity waveform, while end-diastolic velocity is extracted by identifying the lowest stable value during diastole. The pulsatility index and resistance index are obtained by calculating the ratio of the velocity difference to its corresponding baseline value. These four parameters construct quantitative indicators from four dimensions: peak blood flow velocity, peripheral circulation status, vascular elasticity, and resistance characteristics, forming a complete hemodynamic feature description system. By directly using the raw signals from the device to extract parameters, errors introduced by manual intervention or subjective judgment are avoided, ensuring the objectivity and repeatability of the data source and providing stable and reliable feature input for subsequent Bayesian classification models.

[0074] Compared to existing technologies, traditional classification methods rely on manual observation and scale scoring, making them susceptible to subjective judgment by assessors and unable to directly obtain objective hemodynamic parameters. This approach directly extracts blood flow parameters from raw signals monitored by transcranial Doppler ultrasound, using peak systolic velocity, end-diastolic velocity, pulsatility index, and resistance index as classification criteria to establish an objective classification system based on quantitative physiological indicators. Existing technologies lack direct utilization of cerebral hemodynamic parameters, while this approach effectively overcomes the shortcomings of traditional methods in terms of data objectivity and comprehensiveness by integrating multi-dimensional blood flow characteristic parameters.

[0075] Through the above technical solution, this application solves the problem that traditional classification methods rely on subjective judgment and cannot directly obtain objective hemodynamic parameters, thus improving the accuracy of Parkinson's disease motor subtype classification. By directly extracting parameters from the raw signals of the device, subjective biases introduced by manual observation are avoided. Simultaneously, the pulsatility index and resistance index are used to quantify vascular characteristics, enhancing the comprehensiveness and reliability of the classification basis. The feature vector constructed based on objective blood flow parameters provides stable and repeatable data input for the Bayesian classification model, thereby improving the scientific validity and consistency of the classification results.

[0076] In one alternative implementation, the Bayesian discriminant model learns from training samples of known subtypes to obtain a classification function for distinguishing between tremor-dominant and postural instability-difficulty Parkinson's disease.

[0077] The Bayesian discriminant model is a probabilistic classification model built on Bayes' theorem. Specifically, it can be implemented using the maximum a posteriori probability estimation method, classifying samples by calculating the conditional probability of each sample belonging to a different category. This model can integrate prior knowledge and maintain classification stability under finite sample conditions. Training samples refer to clinical datasets containing labeled motor subtypes of Parkinson's disease. These can be constructed by collecting patients' cerebral hemodynamic parameters and dynamic cerebral blood flow autoregulation function indicators, used for model parameter optimization. The classification function is a mathematical expression that maps multidimensional feature vectors to category labels. It can be implemented using a linear or quadratic discriminant function, and subtype differentiation is achieved by setting a probability threshold.

[0078] Specifically, this technical solution collects cerebral blood flow velocity signals, arterial blood pressure signals, and basic hemodynamic parameters from patients with tremor-dominant and postural instability-related dyssteadiness types to construct a feature vector set containing indicators of dynamic cerebral blood flow autoregulation. The feature vectors labeled with subtypes are input into a Bayesian discriminant model for supervised learning. The model establishes a discriminant function describing the distribution characteristics of the two types of samples by calculating the mean vector and covariance matrix of samples from different categories. In the classification stage, the feature vectors of new patients are preprocessed and input into the discriminant function to calculate the posterior probability of belonging to the two subtypes. The final classification result is determined based on the maximum probability principle. This process effectively utilizes the noise tolerance characteristics of the Bayesian framework, reducing the impact of individual physiological differences on the classification results through probabilistic modeling.

[0079] Compared to existing technologies, traditional methods rely on subjective scale scores for subtype classification, which suffers from significant inter-observer variability and difficulty in capturing subtle symptoms. This approach, however, eliminates human judgment bias and standardizes the classification process through objective physiological signal analysis and probabilistic model calculations. Existing machine learning methods are prone to overfitting when sample sizes are insufficient; this approach employs a Bayesian method to fuse prior distribution information, enhancing the model's generalization ability on small clinical datasets. Compared to using hemodynamic parameters alone, this approach combines dynamic cerebral blood flow autoregulation indicators to more comprehensively reflect the pathological differences among different Parkinson's disease subtypes.

[0080] Through the above technical solutions, this application achieves automated classification of motor subtypes of Parkinson's disease, solving the problem of poor classification consistency caused by the reliance on subjective assessment in traditional methods. Probabilistic modeling effectively handles measurement noise and individual differences in clinical data, improving the reliability of classification results. The established classification function provides quantifiable decision-making basis for clinical diagnosis, supports doctors in timely tracking of patient symptom evolution, and lays the technical foundation for developing individualized treatment plans.

[0081] In summary, Parkinson's disease (PD) is a heterogeneous neurodegenerative disease characterized by resting tremor, bradykinesia, rigidity, and postural instability, typically presenting with these features. According to Parts II and III of the Movement Disorders Association Unified Parkinson's Disease Rating Scale (MDS-UPDRS), PD can be classified into three subtypes: Postural Instability and Gait Difficulty (PIGD), Tremor-Dominant (TD), or Indeterminate (ID). Each subtype has unique motor characteristics and prognosis.

[0082] However, traditional classification methods are limited by the observer's subjective judgment and are difficult to comprehensively and timely track changes in patient symptoms. Dynamic autoregulation of cerebral blood flow (dCA) may be related to PD subtypes. Therefore, this invention aims to utilize transcranial Doppler (TCD) to dynamically monitor cerebral blood flow velocity and apply Bayesian analysis to explore the classification of PD.

[0083] Therefore, this invention screened 126 PD patients and 46 healthy volunteers. All participants were assessed using the MDS-UPDRS scale and the modified Hoehn-Yahr staging scale (mH&Y). Cerebral blood flow velocity was monitored using transcranial Doppler (TCD), and blood pressure fluctuations were considered; transfer function analysis (TFA) was used to reflect dCA function. A Bayesian discriminant model for classifying PD patients was established, and subsequently, 60 PD patients and healthy volunteers were randomly assigned to externally validate the model.

[0084] The results showed that the blood flow velocity in the middle cerebral artery of PD patients was significantly lower than that in healthy controls, and there were significant differences in dCA function impairment among different subtypes (P<0.001). Comparative analysis between TD and PIGD patients showed that PIGD patients had poorer dCA function. A Bayesian discriminant model was established using TFA parameters, with an accuracy of 73.3% and an AUC of 0.925. This model was further validated externally, with an accuracy of 68% and an AUC of 0.786.

[0085] The above results confirm that TCD technology can distinguish differences in dCA between different PD subtypes, and a Bayesian analysis equation for classifying PD patients has been established. This provides new insights into Parkinson's disease, helping clinicians identify different subtypes and thus provide individualized treatment.

[0086] To verify the technical effects of the present invention, a total of 126 PD patients who visited the Department of Neurology between June 2021 and September 2024 were recruited as the experimental group, and 46 healthy volunteers matched for gender and age were recruited as the healthy control group.

[0087] The inclusion criteria for PD patients are as follows: (1) meeting the diagnostic criteria for Parkinson's disease established by the Movement Disorders Society (MDS); and (2) being able to cooperate with the examination.

[0088] At the same time, PD patients with other characteristic diseases need to be excluded. The exclusion criteria include: (1) patients with Parkinson's syndrome; (2) patients with a history of cerebral infarction, head trauma, intracranial tumor, encephalitis or other neurological diseases; (3) patients with a history of diabetes, simple autonomic dysfunction, multiple system atrophy or Lewy body dementia or other autonomic neuropathy; (4) patients with a history of arrhythmia, severe anemia, thyroid dysfunction or other diseases affecting cerebral hemodynamics; (5) patients with poor bilateral foci; (6) patients who cannot tolerate or cooperate with the examination.

[0089] It should be noted that the same exclusion criteria were used to select healthy volunteers in the healthy control group.

[0090] (1) Data Acquisition

[0091] Basic data such as age, sex, and blood pressure were required for all participants in both the experimental and healthy control groups; PD-related symptoms and severity were assessed using the Unified Parkinson's Disease Rating Scale (UPDRS); disease severity was assessed using the Modified Hoehn-Yahr Staging Scale (mH&Y); and nonmotor symptoms were assessed using the Nonmotor Symptom Scale for Parkinson's Disease (NMSS). All PD patients were assessed in an "on" state by trained physicians.

[0092] (2) Determination of PD subtype threshold

[0093] The TD / PIGD score ratio was calculated using the MDS-UPDRS scoring criteria from the 2015 MDS. Calculation method: Based on the TD and PIGD scores in the MDS-UPDRS scoring criteria. Each score ranged from 0 to 4 points. A TD / PIGD ratio ≥1.15 indicated the TD subtype, ≤0.90 indicated the PIGD subtype, and values ​​in between indicated the indeterminate type (ID). Because the clinical symptoms of ID PD patients are indeterminate, unbiased, and unstable, this invention only compared the differences in dynamic and static homeostasis function between PIGD and TD subtype PD patients.

[0094] (3) dCA detection

[0095] All examinations were conducted in a quiet room at 24°C. Participants were required to fast for at least 2 hours after meals to minimize postprandial hypotension. Strenuous exercise, alcohol, and caffeine intake were avoided, and medications affecting the autonomic nervous system were discontinued at least 12 hours prior to the examination. Subjects rested quietly in a lying position for 5 minutes. Then, a 1.6MHz TCD probe was placed in the bilateral temporal windows of the subject, secured with a head frame, at a detection depth of 45-60mm. Middle cerebral artery (MCA) blood flow signals were monitored for approximately 10 minutes, and a continuous non-invasive blood pressure monitoring device was connected. After blood flow and blood pressure signals stabilized, the CA200 software was activated to integrate the monitoring information, continuously recording high-quality blood flow and blood pressure data for at least 5 minutes. Subsequently, the patient was instructed to stand actively, and the instrument continued monitoring for 10 minutes, collecting and analyzing the data.

[0096] Following CARNet recommendations, transfer function analysis (TFA) was employed to store and process data used for dCA analysis. Fourier transform was used to convert the time-domain signals of cerebral blood flow velocity and arterial blood pressure into frequency-domain signals to reflect the oscillations of blood pressure and cerebral blood flow within specific frequency ranges. Transfer functions between cerebral blood flow velocity and mean blood pressure were collected at ultra-low frequencies (SLF: 0.02–0.07 Hz) and low frequencies (LF: 0.07–0.20 Hz) to obtain dCA parameters (gain, phase, coherence, normalized gain Gain1 (% / mmHg), absolute gain Gain2 (cm / s / mmHg)). Coherence was used as a data quality control parameter; data with coherence ≥ 0.51 were included in subsequent statistical analyses. Based on the TFA method, raw data from stable heart rate, BP, and TCD monitoring within 10 minutes of lying down, as well as induced standing blood pressure fluctuations, were selected for dCA analysis. The average values ​​of the dCA parameters from both hemispheres were used for further analysis. Mxa represents the average correlation coefficient between mean cerebral artery flow velocity (MCAv) and arterial blood pressure (ABP). Gain1 and Gain2 both represent the amplitude change of CBFV relative to arterial blood pressure; a smaller Gain value indicates better dCA function. Phase Phae represents the velocity characteristics of dCA, i.e., the time difference between changes in CBFV and arterial blood pressure; a higher phase indicates better dCA function.

[0097] (4) Statistical Analysis

[0098] All statistical tests were performed using SPSS statistical software. Normally distributed continuous variables are expressed as mean ± standard deviation. Independent samples t-tests were used for comparisons between groups; skewed continuous variables were expressed as median (interquartile range, IQR), and Mann-Whitney U tests were used for comparisons between groups. Categorical variables (e.g., gender) were expressed as percentages and analyzed using chi-square tests. Kruskal-Wallis tests were used to detect differences among multiple independent samples, such as dCA parameters and clinical characteristics between subgroups. If statistically significant differences existed between groups, Bonferroni-corrected Mann-Whitney U tests were performed. Spearman rank correlation analysis was used to analyze the correlation between HY stage and NMSS. PD patients were the experimental group, and healthy patients were the control group. A Bayesian discriminant equation was established with cerebral blood flow parameters as independent variables and PD motor type as the dependent variable. Sixty PD patients and healthy volunteers from 2021 to 2024 were randomly included as external validation samples. Self-validation, cross-validation, and external validation were used to test the discriminant equation. The significance level was set at α = 0.05, and a two-tailed test was used; P < 0.05 indicated statistical significance.

[0099] (5) Results Comparison

[0100] 1. Comparison of basic information and cerebral blood flow autoregulation parameters between PD patients and healthy controls

[0101] A total of 126 PD patients (mean age 61.74 ± 10.51 years; 28 males and 44 females) and 46 healthy controls (mean age 56.62 ± 9.52 years; 24 males and 22 females) successfully completed the dCA test. There were no statistically significant differences between PD patients and healthy controls in terms of age, sex, and peripheral arterial systolic and diastolic blood pressure (P > 0.05), as shown in Table 1.

[0102] Table 1

[0103]

[0104] Compared with the healthy control group, patients with PD had slower peak systolic velocity (PSV) and end-diastolic velocity (EDV) in the middle cerebral artery: PSV: (80.65±15.67) cm / s vs (87.79±25.65) cm / s, P=0.011; EDV: (34.03±8.77) cm / s vs (45.51±18.92) cm / s, P<0.001; pulsatility index (PI): 0.90±0.16 vs 1.23±0.48, P<0.001; resistance index (RI): 0.47±0.17 vs 0.54±0.08, P=0.007.

[0105] When lying down, compared with the healthy control group, PD patients had lower phase values ​​in the low-frequency band [39.86 (31.71, 49.73) vs 42.13 (35.705, 47.79), P = 0.044] and higher normalized gain Gain1 (%) values ​​[1.53 (1.27, 1.94) vs 1.3 (1.155, 1.535), P = 0.005]. The differences were statistically significant, as shown in Table 2.

[0106] Table 2

[0107]

[0108] In the ultra-low frequency band, the Phase value of the PD group was lower than that of the control group [64.12 (43.38, 76.17) vs 65.69 (57.745, 79.44), P = 0.007]; the normalized gain Gain1 (%) value was higher [1.19 (0.99, 1.51) vs 1.07 (0.9, 1.34), P = 0.021], and the difference was statistically significant.

[0109] After standing, the comparison of dCA parameters between PD patients and healthy controls showed that, in the low-frequency range, the Phase value of the PD group was lower than that of the control group [35.67 (27.5, 43.76) vs 40.58 (35.37, 49.24), P < 0.001]; the absolute gain Gain2 value was higher [0.72 (0.6, 0.945) vs 0.85 (0.69, 0.98), P = 0.009]; and the normalized gain Gain1 (%) value was higher [1.25 (1.055, 1.355) vs 1.63 (1.45, 1.99), P < 0.001]. The differences were statistically significant, as shown in Table 3.

[0110] Table 3

[0111]

[0112]

[0113] In the ultra-low frequency band, the PD group had a higher normalized gain Gain1 (%) value [0.99 (0.855, 1.19) vs 1.17 (1.01, 1.43), P < 0.001], and the difference was statistically significant.

[0114] 2. Differences in basic clinical data among PD patients of different motor subtypes

[0115] A total of 126 PD patients were included, including 51 with the TD subtype and 75 with the PIGD subtype. There were no statistically significant differences in gender, age, BMI, and disease duration between the two groups, as shown in Table 4. There were no statistically significant differences in the MDS-UPDRS total score or MDS-UPDRS III score between the two groups; the mean H&Y stage was higher in the PIGD group than in the TD group (2.2±0.36 vs 1.75±0.59, P=0.002). The NMSS score was positively correlated with the H&Y stage (r=0.244, P=0.011).

[0116] Table 4

[0117]

[0118] (6) Establishment of Bayesian discriminant function model

[0119] The dCA parameters of the normal group, TD group, and PIGD group were compared and analyzed. Parameters with p ≤ 0.1 in the transfer function analysis (TFA) method were included as variables in the Bayesian discriminant model. Based on TFA, this invention established a discriminant equation in healthy individuals and PD exercise classification: y = a1x1 + a2x2 + a3x3 + a4x4 + ... + constant. "a" represents the coefficient, and x represents the included parameter value. A total of 172 participants were included as the training set for this invention, and a Bayesian discriminant model was established, yielding two discriminant equations. The specific discriminant system is shown in Table 5. By substituting the independent variable data into the equations for each level group, the maximum y value obtained corresponds to the classification of the research subjects, such as... Figure 2 As shown.

[0120] Table 5

[0121]

[0122] Wherein, lying represents the relevant parameters measured in the lying position, and standing represents the relevant parameters measured in the standing position.

[0123] The classification results are shown in Table 6.

[0124] Table 6

[0125]

[0126] (7) Results and verification of the Bayesian discriminant function

[0127] 1. Internal validation and cross-validation

[0128] To verify the accuracy of the Bayesian discriminant formula, the independent variable data of the study subjects were substituted into the equation and compared with the patients' UPDRS motor classification. The function calculates the y-value for each patient to determine the classification, such as... Figure 3As shown in (a). The results show that the constructed comprehensive discriminant function has a correct recognition rate of 73.3% and a misclassification rate of 26.7%. The area under the curve (AUC) of the diagnostic model is 0.925 [95% CI (0.875-0.959)], the Youden index is 0.788, the sensitivity is 89.68% [95% CI (83.0%-94.4)], and the specificity is 89.13% [95% CI (76.4%-96.4%)]. Figure 3 As shown in (b).

[0129] To evaluate the effectiveness of the discriminant model, leave-one-out cross-validation was performed on the discriminant function established for 172 subjects, with a correct recognition rate of 64.5%.

[0130] 2. External Validation

[0131] Based on the established Bayesian equation, 47 newly admitted PD patients and 13 healthy volunteers were included. Data from lying and standing positions were measured and substituted into the diagnostic model to validate its diagnostic ability, yielding a correct recognition rate of 68% and a misclassification rate of approximately 32%. In the normal group of 13 individuals, the diagnostic accuracy was approximately 92.3%; for the TD group of 18 individuals, the correct classification rate was approximately 55.56%; and for the PIGD group of 29 individuals, the diagnostic accuracy was approximately 65.52%. The area under the curve (AUC) of the diagnostic model was 0.786 [95% CI 0.661–0.881], the sensitivity was 61.7% [95% CI (55.7%–83.6%)], and the specificity was 92.3% [95% CI (68.1%–99.8%)]. Figure 3 As shown in (c).

[0132] (7) Comprehensive Analysis

[0133] Motor dysfunction is a core clinical symptom of Parkinson's disease (PD), primarily caused by severe loss of dopaminergic neurons in the basal ganglia. The Unified Parkinson's Disease Rating Scale (UPDRS) is the most commonly used scale for assessing motor symptoms in PD patients. Based on UPDRS scores, PD is first divided into tremor-dominant (TD) and postural instability-difficulty (PIGD) subtypes. PD patients with different motor subtypes have different pathophysiological bases, disease progression, and prognoses. Clinically, TD phenotype patients typically present with resting tremor, normal gait, and mild disease progression, while PIGD phenotype is characterized by bradykinesia and rigidity. PIGD subtype patients also exhibit more significant brain structural damage and volume reduction. In this invention, the NMSS score and H&Y stage were higher in the PIGD group than in the TD group, and the NMSS score was positively correlated with the H&Y stage, consistent with previous studies. Therefore, improving the clinical classification of PD to achieve individualized treatment and identifying patient subgroups in PD are of great significance for defining PD heterogeneity, understanding the pathogenesis mechanism and formulating treatment hypotheses.

[0134] Furthermore, previous studies have indicated that PD is considered a disease associated with cerebrovascular dysfunction. Insufficient cerebral perfusion plays a crucial role in the neuropathological changes of neurodegenerative diseases. Previous research has confirmed that dopaminergic neurons attach to cerebral microvessels, and the metabolic reduction caused by neuronal degeneration and death leads to changes in cerebral blood flow (CBF). This study used transcranial Doppler (TCD) to assess dCA function in each group and classified PD according to internationally recognized PD motor classification criteria. Results showed that, compared to healthy controls, PD patients had slower middle cerebral artery blood flow velocity and increased PI and RI when resting in a supine position, consistent with theoretical results. Animal experiments have found that an early PD mouse model with excessive α-synuclein expression affects neurovascular units and astrocytes, reducing cerebral blood flow velocity and affecting glucose uptake and metabolism. The slowed cerebral blood flow velocity in PD patients may be attributed to the deficiency of dopaminergic neurotransmitters, reducing the metabolic demand of brain tissue, while increased blood viscosity leads to increased peripheral small-volume vascular resistance and decreased vascular compliance, further exacerbating tissue ischemia. In the TFA study, gain, phase, coherence function, absolute gain (Gain2), and normalized gain (Gain1(%)) were used to represent the trend of CBFV with blood pressure (BP), reflecting cerebrovascular regulatory function. The study found that PD patients exhibited lower phase values ​​and higher gain values ​​in both supine and standing positions compared to the normal group. Further subgroup analysis showed that PD patients in the PIGD group had poorer dCA function, which corresponds to the previously observed poorer quality of life in PIGD-type PD patients. Previous pathological studies have shown that compared to PIGD-type PD patients, TD-type PD patients have less locus coeruleus neuron involvement and relatively preserved adrenergic function. However, in PIGD-type PD patients, locus coeruleus neurons are significantly involved, leading to decreased peripheral vascular resistance and cerebral perfusion pressure. Insufficient cerebral perfusion can lead to fatigue, drowsiness, cognitive decline, and reduced norepinephrine release from sympathetic neurons. Central norepinephrine deficiency is associated with postural instability and gait abnormalities in Parkinson's disease (PD), and these patients typically exhibit nonmotor characteristics such as depression, emotional blunting, attention deficit, and sleep disturbances. Therefore, the heterogeneity of PD subtypes may affect patients' dCA function through complex regulatory mechanisms, providing theoretical support for the discriminant analysis of PD subgroups in this invention.

[0135] Discriminant analysis (DFA) is widely used in many medical fields. This invention utilizes the relationship between spontaneous blood pressure changes and location-induced blood pressure fluctuations and cerebral blood flow changes to assess dCA function, more closely aligning with physiological regulatory conditions. Subsequently, a Bayesian discriminant model was established to classify and subtype PD patients based on objective data. The results of the discriminant function showed a correct classification rate of 73.3% in the internal test and 64.5% in cross-validation. Since internal testing and cross-validation methods often underestimate the misclassification rate and exaggerate the discriminant effect, we additionally used external validation samples to estimate the expected misclassification probability of the diagnostic model. This method produced a more objective misclassification probability, while still achieving a diagnostic accuracy of 68%, indicating that the established discriminant model has good stability and low variability, and that using TFA as the basis for distinguishing PD subtypes is clinically feasible.

[0136] This invention demonstrates differences in the ability of patients with different motor subtypes to regulate cerebral blood flow, providing new insights into the pathophysiological mechanisms of PD. Furthermore, the use of TCD technology and Bayesian discriminant analysis for effective classification of PD patients significantly reduces the impact of subjectivity and inter-rater consistency fluctuations on the discrimination results in clinical practice, saving time and effort, and assisting clinicians in timely and accurate identification and individualized treatment of different PD subtypes.

[0137] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for classifying motor subtypes of PD based on transcranial Doppler and Bayesian analysis, characterized in that, include: S1. The transcranial Doppler device was used to collect cerebral blood flow velocity signals of the subjects under different physiological conditions, and the arterial blood pressure signals were collected simultaneously using a continuous non-invasive blood pressure monitoring device. S2. Process the signals collected in S1 and extract characteristic indicators that reflect the dynamic cerebral blood flow autoregulation function. S2 includes: The time-domain signals of the cerebral blood flow velocity signal and the arterial blood pressure signal are converted into frequency-domain signals using Fourier transform; Based on the converted frequency domain signal, the transfer function analysis method is used to calculate the frequency domain characteristic parameters between cerebral blood flow velocity signals and arterial blood pressure signals within one or more preset frequency bands; wherein, the frequency domain characteristic parameters include phase, gain, and consistency function; Simultaneously, based on the average frequency domain characteristic parameters of both hemispheres, the average correlation coefficient between average cerebral blood flow velocity and arterial blood pressure was calculated. The frequency domain feature parameters and the average correlation coefficient are used together as feature indicators to characterize the dynamic cerebral blood flow autoregulation function. S3. Combine the aforementioned feature indicators with the basic hemodynamic parameters to form a feature vector; S4. Input the feature vector into the pre-trained Bayesian discriminant model; S5. Based on the output of the Bayesian discriminant model, classify the subjects into different motor subtypes of Parkinson's disease.

2. The method according to claim 1, characterized in that, S1 includes: While the subject was in a resting state, cerebral blood flow velocity signals were continuously collected using a transcranial Doppler device for a first preset time period, and arterial blood pressure signals were simultaneously collected using a continuous non-invasive blood pressure monitoring device. While the subject underwent postural changes, cerebral blood flow velocity and arterial blood pressure signals were collected during a second preset time period.

3. The method according to claim 2, characterized in that, The acquisition conditions for the transcranial Doppler device are as follows: The probe of a 1.6 MHz transcranial Doppler device was fixed to the bilateral temporal windows of the subject, with a detection depth of 45–60 mm; Subjects were required to fast for ≥2 hours and discontinue autonomic nervous system medications for ≥12 hours prior to the data collection.

4. The method according to claim 1, characterized in that, The basic hemodynamic parameters are cerebral artery blood flow parameters directly obtained from the raw cranial Doppler signals monitored by the transcranial Doppler device; the cerebral artery blood flow parameters include peak systolic velocity, end-diastolic velocity, pulsatility index, and resistance index.

5. The method according to claim 1, characterized in that, The Bayesian discriminant model is obtained by learning from training samples of known subtypes to obtain a classification function for distinguishing between tremor-dominant and postural instability-difficulty Parkinson's disease.

Images