Intelligent brain deep electrical stimulation parameter optimization method for Parkinson patients

Abstract

The invention relates to the technical field of deep brain stimulation, in particular to a deep brain electrical stimulation parameter intelligent optimization method for Parkinson patients. According to the method, feature measurement is carried out on neural residual signals in the deep brain stimulation process, waveform slope features and observation credibility are analyzed through the neural residual signals, and therefore an artifact form penalty term and path accumulation cost are determined; based on the path accumulation cost, a dynamic programming algorithm is adopted to obtain pathological signal estimation intensity, the pathological signal estimation intensity is utilized to perform characteristic measurement on the pathological outbreak density, and the stimulation pulse amplitude in the deep brain stimulation process is optimized based on the pathological outbreak density. According to the method, the intensity is estimated by measuring the pathological signals, the problem of signal fragmentation of the pathological signals is avoided, and therefore the stimulation pulse amplitude in the deep brain stimulation process is accurately optimized.

Description

Technical Field

[0001] This invention relates to the field of deep brain stimulation technology, specifically to an intelligent optimization method for deep brain electrical stimulation parameters for Parkinson's disease patients. Background Technology

[0002] Deep brain stimulation (DBS) is an effective treatment for neurodegenerative diseases such as Parkinson's disease by delivering high-frequency electrical pulses to specific brain nuclei via implanted electrodes. To improve treatment precision and reduce side effects, stimulation parameters need to be dynamically adjusted based on the patient's real-time pathological state during DBS treatment of Parkinson's patients. However, DBS can generate stimulation artifacts that mask and truncate weak neural signals.

[0003] Meanwhile, the amplitude of the stimulation pulse during deep brain stimulation is typically in the volt range, while the effective LFP (Local Field Potential) neural signal is only in the microvolt range, a difference of approximately [missing information]. Although the front-end amplifier has a certain common-mode rejection capability, the double-layer capacitance effect at the electrode-tissue interface produces an exponentially decaying artifact, which covers a wide frequency band and overlaps with the LFP neural signal. More seriously, at the moment of stimulation pulse delivery and for a short period thereafter, the front-end amplifier often enters a saturation or cutoff state due to excessive signal, resulting in the complete loss or flat-top truncation of the physical signal during this period.

[0004] Currently, existing technologies typically employ a blanking strategy, which involves directly discarding pathological signal data during the pulse. However, this breaks the continuous pathological signal into discontinuous fragments in the time domain, resulting in signal fragmentation. This makes it impossible to accurately optimize the amplitude of the stimulation pulse during deep brain stimulation, leading to unstable symptom control in Parkinson's patients. Summary of the Invention

[0005] To address the technical problem of existing technologies employing blanking strategies that fragment continuous pathological signals into discontinuous pieces in the time domain, thus hindering accurate optimization of stimulation pulse amplitude during deep brain stimulation (DBS), this invention aims to provide an intelligent optimization method for DBS parameters in Parkinson's disease patients. The specific technical solution adopted is as follows: This invention proposes an intelligent optimization method for deep brain stimulation parameters for Parkinson's disease patients, which includes the following steps: The mixed field potential signal during deep brain stimulation is acquired. Based on the signal saturation state at each sampling time, artifacts in the mixed field potential signal are eliminated to determine the neural residual signal at each sampling time. Based on the magnitude of the neural residual signal, the instantaneous residual energy is analyzed to determine the waveform slope factor at each sampling time. The residual instantaneous energy is compared with the preset maximum signal amplitude to determine the state observation value at each sampling time; each preset state index and corresponding preset state signal value are preset at each sampling time; the preset capacitor discharge condition is determined based on the waveform slope factor and the preset state signal value; and the artifact morphology penalty term of the preset state index is obtained based on the determination. For each preset state index at the current sampling time, compare it with the preset state index at the previous sampling time, analyze the first deviation between the state observation value and the preset state signal value, and combine it with the pseudo-trace morphology penalty term to obtain the path cumulative cost. Based on the path cumulative cost, a dynamic programming algorithm is used to obtain the optimal state index at each historical sampling time, and the pathological signal intensity at each historical sampling time is estimated using the optimal state index. The pathological burst density at each sampling time is obtained by estimating the temporal distribution of the intensity of the pathological signal, and the amplitude of the stimulation pulse during deep brain electrical stimulation is optimized based on the pathological burst density.

[0006] Preferably, the method for determining the signal saturation state includes: The difference between the maximum value of the preset range of the ADC analog-to-digital converter and the preset saturation judgment margin is calculated as the first difference. The sum of the minimum value of the preset range of the ADC analog-to-digital converter and the preset saturation judgment margin is used as the first sum value. Based on the comparison results between the mixed field potential signal and the first difference and the first sum, a saturation mask is determined for each sampling moment. A saturation mask of 0 indicates that the current sampling moment is in a non-saturated state; a saturation mask of 1 indicates that the current sampling moment is in a saturated state.

[0007] Preferably, the method for acquiring the neural residual signal includes: Within the initial preset first parameter of stimulation cycles after deep brain stimulation is initiated, the length of the stimulation cycle is the reciprocal of the preset stimulation frequency. A pseudo-waveform template vector with the same length as the stimulation cycle is constructed. At each sampling moment in the unsaturated state, the mixed field potential signal within the preset first parameter of stimulation cycles is phase-superimposed and averaged. Based on the result of the phase-superimposed and averaged, the initial template value is filled into the pseudo-waveform template vector to obtain the pseudo-waveform vector. The artifacts of the mixed field potential signal are eliminated based on the artifact waveform vector to determine the neural residual signal at each sampling time.

[0008] Preferably, the method for obtaining the waveform slope factor includes: The absolute value of the neural residual signal is calculated as the instantaneous residual energy at each sampling time. Based on the difference in neural residual signals at adjacent sampling times, and combined with the instantaneous residual energy, the waveform slope factor at each sampling time is obtained.

[0009] Preferably, the method for obtaining the state observation values ​​includes: Calculate the ratio of the residual instantaneous energy to the preset maximum signal amplitude, and use it as the first ratio; The first ratio is mapped using a preset state resolution, and the mapping result is rounded and compared with the preset state resolution to obtain the state observation value at each sampling time.

[0010] Preferably, the method for obtaining the artifact morphology penalty term includes: Obtain the pulse trigger time of the stimulus pulse to which each sampling time belongs, and calculate the difference between each sampling time and the pulse trigger time of the stimulus pulse to which it belongs, as the stimulus phase offset of each sampling time; For each sampling moment, if the stimulus phase offset belongs to an element within the sensitive time window of the preset capacitor discharge, and the waveform slope factor is less than the preset waveform smoothing threshold, and the preset state signal value corresponding to each preset state index is greater than the preset safety noise threshold, then the preset capacitor discharge triggering condition is met. Based on the preset strong suppression coefficient and the preset state signal value, the artifact morphology penalty term for each preset state index at each sampling moment is determined. If the preset capacitor discharge triggering condition is not met, then the artifact morphology penalty term for each preset state index at each sampling time is set to 0.

[0011] Preferably, the method for obtaining the path cumulative cost includes: For the first sampling time, the cumulative path cost for each preset state index at the first sampling time is set to 0; For each sampling time other than the first sampling time, the observation confidence of each sampling time is analyzed, and the observation confidence is used as the weight of the first deviation to construct an observation fitting term; The artifact morphological penalty term for each preset state index at each sampling time is used as the morphological penalty term; A set of state indices to be analyzed is constructed with each preset state index at the current sampling time as the center; the state indices to be analyzed at the previous sampling time are compared with the corresponding preset state indices at the current sampling time, and the path cumulative cost under the state indices to be analyzed at the previous sampling time is combined to construct a state transition term; Based on the observation fitting term, the morphological penalty term, and the state transition term, the path cumulative cost of each preset state index at each of the remaining sampling times is obtained.

[0012] Preferably, the method for obtaining the observation confidence level includes: For each sampling time, if the signal is in a saturated state at that sampling time, then the observation confidence of that sampling time is set to 0. If the sampling time is in a signal saturation state, the observation reliability at that sampling time is calculated by comparing the neural residual signal and the mixed field potential signal within a preset short time window.

[0013] Preferably, the method for obtaining the estimated intensity of the pathological signal includes: Calculate the second ratio using the optimal state index and the preset state resolution; The product of the second ratio and the preset maximum signal amplitude is calculated as the estimated intensity of the pathological signal at each historical sampling time.

[0014] Preferably, the method for obtaining the pathological outbreak density includes: A preset long-term sliding window and a preset short-term statistical window are pre-set for each sampling time, and the preset pathological judgment threshold for each sampling time is obtained using the preset long-term sliding window; The number of sampling points in which the estimated intensity of the pathological signal at all historical sampling times within the preset short-term statistical window is greater than the preset pathological judgment threshold is recorded as the target number of sampling points. The ratio of the target number of sampling points to the total number of sampling points within the preset short-term statistical window is recorded as the pathological outbreak density at each sampling time.

[0015] The present invention has the following beneficial effects: This invention acquires mixed field potential signals during deep brain stimulation, providing a reliable data foundation for accurate extraction of pathological features. It analyzes signal saturation to eliminate artifacts in the mixed field potential signals and determine neural residual signals. Signal saturation features more clearly reflect the loss of pathological feature information, helping to eliminate exponentially decaying trailing artifacts and more clearly revealing the pathological features of Parkinson's patients. Based on the magnitude of the neural residual signals, the invention analyzes the instantaneous energy of the residuals and determines the waveform slope factor, providing a reliable physical constraint for subsequent dynamic programming algorithms and helping to solve the problem of signal fragmentation in pathological signals. By determining the state observation values ​​and combining the waveform slope factor and preset state signal values, the invention judges whether preset capacitor discharge conditions are met, and obtains artifact morphology penalties based on the judgment. The penalty term facilitates the active suppression of false signals. Furthermore, obtaining the path accumulation cost based on the artifact morphology penalty term allows for a balance between the reliability of observed data and the rationality of the physical waveform, helping to address the problem of signal fragmentation in pathological signals. Through a dynamic programming algorithm based on path accumulation cost, the estimated intensity of the pathological signal at each historical sampling moment is obtained, representing a temporally continuous index smoothed by physical morphological constraints and physiological inertia, thus solving the problem of signal fragmentation in pathological signals and enabling accurate optimization of the stimulation pulse amplitude during deep brain stimulation. The pathological burst density at each sampling moment is obtained based on the estimated intensity of the pathological signal, which more realistically reflects the current pathological load duty cycle, allowing for accurate optimization of the stimulation pulse amplitude during deep brain stimulation. This invention avoids signal fragmentation in pathological signals by measuring the estimated intensity of the pathological signal, thereby accurately optimizing the stimulation pulse amplitude during deep brain stimulation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a method for intelligent optimization of deep brain stimulation parameters for Parkinson's disease patients, provided as an embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for intelligent optimization of deep brain stimulation parameters for Parkinson's disease patients proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent optimization method for deep brain stimulation parameters for Parkinson's disease patients provided by this invention.

[0021] Please see Figure 1 The diagram illustrates a flowchart of a method for intelligent optimization of deep brain stimulation parameters for Parkinson's disease patients, according to an embodiment of the present invention. The method includes the following steps: Step S101: Acquire mixed field potential signals during deep brain stimulation; eliminate artifacts in the mixed field potential signals based on the signal saturation state at each sampling time; determine the neural residual signal at each sampling time; analyze the instantaneous residual energy based on the magnitude of the neural residual signal; and determine the waveform slope factor at each sampling time.

[0022] In order to accurately optimize the amplitude of stimulation pulses during deep brain stimulation, it is necessary to collect the mixed field potential signal at each sampling moment during deep brain stimulation. This will help to accurately extract pathological feature information from the mixed field potential signal.

[0023] In general, in unsaturated mixed field potential signals, due to the double-layer capacitance effect between the electrodes and brain tissue interface during signal acquisition, exponentially decaying trail artifacts will be generated in the mixed field potential signals. Although the amplitude of these exponentially decaying trail artifacts is smaller than that in the saturation region, they will still cover the neural signal frequency band, thereby interfering with the pathological characteristics of Parkinson's patients.

[0024] Therefore, to eliminate artifact interference in the mixed field potential signal, it is necessary to eliminate artifacts in the mixed field potential signal based on the signal saturation state at each sampling time, thereby determining the neural residual signal at each sampling time. This neural residual signal can more clearly reflect the pathological characteristics of Parkinson's patients, eliminating the trailing artifacts of exponential decay, which is beneficial for subsequent accurate analysis of pathological characteristics. However, if a sampling time is in a signal saturation state, the voltage amplitude at the moment of stimulation pulse delivery during deep brain stimulation often exceeds the linear input range of the front-end amplifier, easily causing the mixed field potential signal to exhibit a flat-top truncation signal characteristic. In this case, there is a greater possibility of loss of pathological characteristic information.

[0025] To provide a reliable physical constraint basis for subsequent dynamic programming algorithms, it is necessary to analyze the instantaneous energy of the residual based on the magnitude of the neural residual signal and determine the waveform slope factor at each sampling time. The instantaneous energy of the residual represents the absolute intensity of the neural residual signal and is used as the observation benchmark for subsequent state estimation. The waveform slope factor represents the normalized rate of change of the waveform in the neural residual signal and is used to identify smooth capacitor discharge artifacts.

[0026] In one specific implementation of this invention, the global sampling time is defined as k. Mixed field potential signals from Parkinson's patients are acquired in real time via implanted electrode contacts and converted by a front-end analog amplifier and an ADC analog-to-digital converter to obtain the mixed field potential signals at each sampling time during the real-time acquisition of deep brain stimulation. .

[0027] Step S102: Compare the residual instantaneous energy with the preset maximum signal amplitude to determine the state observation value at each sampling time; pre-set each preset state index and corresponding preset state signal value at each sampling time; determine whether the preset capacitor discharge condition is met based on the waveform slope factor and the preset state signal value; obtain the artifact morphology penalty term of the preset state index based on the determination; for each preset state index at the current sampling time, compare it with the preset state index at the previous sampling time, and analyze the first deviation between the state observation value and the preset state signal value; combine the artifact morphology penalty term to obtain the path cumulative cost.

[0028] To address the issues of nonlinear artifact interference and loss of pathological feature information in mixed field potential signals, it is necessary to transform the physical characteristics of the mixed field potential signals into an optimal path search problem at the mathematical level. This is achieved by constructing a cost function that incorporates artifact morphology penalties and physiological inertia constraints, thereby finding the pathological intensity trajectory that best conforms to physiological laws in the discrete state space. Therefore, comparing the residual instantaneous energy with the preset maximum signal amplitude to determine the state observation value at each sampling moment, and using this comparative method, can more clearly reflect the state observation characteristics caused by the residual instantaneous energy.

[0029] Because the separated neural residual signal may still retain nonlinear capacitance waveforms with smooth characteristics, and amplitude jumps may occur at the moment the front-end analog amplifier exits saturation, in order to actively suppress non-neural signals from being misjudged as pathological outbreaks, a preset state index and corresponding preset state signal value are pre-set for each sampling moment. The preset state signal value is in a dimensionless state space, which helps solve the dimensional difference problem in subsequent calculations. The waveform slope factor and the preset state signal value are used to determine whether a preset capacitance discharge condition is met, and the artifact morphology penalty term of the preset state index is obtained based on the determination. The preset state signal value helps to accurately determine whether the current sampling moment is in a phase prone to capacitance discharge, while the waveform slope factor helps to accurately identify whether the signal waveform exhibits abnormally smooth characteristics. Therefore, the artifact morphology penalty term represents a higher cost applied to high-amplitude state estimation if the current sampling moment is in a phase prone to capacitance discharge and the waveform exhibits abnormally smooth characteristics, for subsequent active suppression of false signals.

[0030] Therefore, in order to accurately construct a cost function that includes artifact morphology penalties and physiological inertia constraints, and thus find the pathological intensity trajectory that best conforms to physiological laws in the discrete state space, for each preset state index at the current sampling time, the path cumulative cost is obtained by comparing it with the preset state index at the previous sampling time and analyzing the first deviation between the observed state value and the preset state signal value, combined with the artifact morphology penalty term. Specifically, by comparing the preset state index at the previous sampling time to ensure changes conforming to physiological inertia, and by analyzing the first deviation between the observed state value and the preset state signal value, the reliability of the observed data is weighed. Combined with the artifact morphology penalty term, the physical waveform becomes more reasonable. Therefore, this path cumulative cost can balance the reliability of the observed data with the reasonableness of the physical waveform, helping to solve the problem of signal fragmentation in pathological signals.

[0031] Step S103: Based on the path cumulative cost, a dynamic programming algorithm is used to obtain the optimal state index for each historical sampling time, and the pathological signal intensity is estimated for each historical sampling time using the optimal state index.

[0032] To output a continuous, smooth, and repaired optimal pathological state trajectory in the temporal domain, thus addressing the signal fragmentation problem in pathological signals, a dynamic programming algorithm is needed to obtain the optimal state index for each historical sampling moment based on path accumulation cost. This index is used to infer the intermediate state that should transition to a higher amplitude at each historical sampling moment, thereby helping to fill in signal gaps in the pathological signal and achieving a smooth reconstruction of the optimal pathological state trajectory. Therefore, by utilizing the optimal state index to obtain the estimated intensity of the pathological signal at each historical sampling moment, this estimated intensity is a continuous temporal index smoothed by physical morphological constraints and physiological inertia. This solves the problem of signal fragmentation in pathological signals and is used to accurately optimize the amplitude of stimulation pulses during deep brain stimulation.

[0033] Step S104: Based on the pathological signal intensity distribution over time, obtain the pathological burst density at each sampling time, and optimize the stimulation pulse amplitude during deep brain electrical stimulation based on the pathological burst density.

[0034] To more accurately optimize the amplitude of stimulation pulses during deep brain stimulation, taking into account the severity of motor symptoms in Parkinson's patients... The duration of the frequency band neural oscillation is positively correlated. Therefore, the pathological burst density at each sampling moment is obtained by estimating the intensity based on the pathological signal. This pathological burst density can accurately reflect the current pathological load duty cycle. Through this pathological burst density, the amplitude of the stimulation pulse during deep brain stimulation can be accurately optimized, thereby avoiding the problem of unstable symptom control in Parkinson's patients.

[0035] Preferably, in some implementations of the present invention, the method for determining the signal saturation state includes: In order to accurately optimize the amplitude of stimulation pulses during deep brain stimulation, the saturation mask of each sampling moment is obtained using the mixed field potential signal, thereby more accurately indicating whether the mixed field potential signal at the current sampling moment has been distorted.

[0036] Because the voltage amplitude at the moment of stimulation pulse delivery during deep brain stimulation often exceeds the linear input range of the front-end amplifier, the mixed field potential signal can easily exhibit a flat-top truncation characteristic. This increases the likelihood of loss of pathological information. Therefore, saturation detection is necessary for the mixed field potential signal at each sampling time.

[0037] Based on the above analysis, the preset range of the ADC analog-to-digital converter is set as follows: In one specific implementation of the present invention and They are respectively and .

[0038] The difference between the maximum value of the preset range of the ADC analog-to-digital converter and the preset saturation judgment margin is calculated as the first difference. In a specific implementation of this invention, the preset saturation judgment margin is 1% of the preset range of the ADC analog-to-digital converter.

[0039] The sum of the minimum value of the preset range of the ADC analog-to-digital converter and the preset saturation judgment margin is calculated as the first sum value.

[0040] Based on the comparison results between the mixed field potential signal and the first difference and the first sum, a saturation mask is determined for each sampling moment. A saturation mask of 0 indicates that the current sampling moment is in a non-saturated state; a saturation mask of 1 indicates that the current sampling moment is in a saturated state. Simultaneously, when the saturation mask is 1, it indicates that the current mixed field potential signal is distorted and does not contain valid neural information; when the saturation mask is 0, it indicates that the current mixed field potential signal is not distorted and contains valid neural information.

[0041] In one specific implementation of this invention, the saturation mask is calculated as follows: In the formula, The saturation mask at the k-th sampling time. The mixed field potential signal at the k-th sampling time. This is the maximum value within the preset range of the ADC analog-to-digital converter. To pre-determine the saturation margin, This is the minimum value of the preset range of the ADC analog-to-digital converter. The first difference, This is the first sum.

[0042] Preferably, in some implementations of the present invention, the method for obtaining the neural residual signal includes: In general, in unsaturated mixed field potential signals, due to the double-layer capacitance effect between the electrodes and brain tissue interface during signal acquisition, exponentially decaying trail artifacts will be generated in the mixed field potential signals. Although the amplitude of these exponentially decaying trail artifacts is smaller than that in the saturation region, they will still cover the neural signal frequency band, thereby interfering with the pathological characteristics of Parkinson's patients.

[0043] Therefore, in order to eliminate interference with the pathological characteristics of Parkinson's patients, a artifact waveform template vector with the same length as the stimulation cycle is constructed within the initial preset first parameter stimulation cycle after deep brain stimulation is initiated. The length of the stimulation cycle is the reciprocal of the preset stimulation frequency. In one specific implementation of this invention, the preset first parameter is 50.

[0044] Then, at each sampling moment in the unsaturated state, the mean value of the mixed field potential signal corresponding to the same stimulus phase offset within the preset first parameter stimulation cycles is calculated and recorded as the phase superposition mean value of each stimulus phase offset. Then, the corresponding position of the phase superposition mean value of each stimulus phase offset on the artifact waveform template vector is filled with the initial value of the template. If there are missing values ​​in the filled artifact waveform template vector, 0 is filled into the corresponding missing positions to obtain the artifact waveform vector. This reflects the characteristic information of artifact waveforms in the mixed field potential signal.

[0045] Therefore, in order to eliminate interference with the pathological characteristics of Parkinson's patients, artifacts in the mixed field potential signal are eliminated based on the artifact waveform vector, and the neural residual signal at each sampling time is determined. This neural residual signal can more clearly reflect the pathological characteristics of Parkinson's patients, eliminates the trailing artifacts of exponential decay, and is beneficial for subsequent accurate analysis of pathological characteristics.

[0046] In one specific implementation of this invention, since the artifacts generated by deep brain stimulation are primarily manifested at the time of stimulation pulse occurrence, establishing a phase index based on the time of stimulation pulse occurrence serves as the time reference for signal processing. Therefore, by monitoring the pulse trigger index of the stimulation pulse at each sampling time using a logic level detection circuit, for example, under the nth stimulation pulse at the kth sampling time, the pulse trigger index of the current sampling time is recorded as follows: The pulse trigger index represents the sampling time corresponding to the pulse trigger time.

[0047] To establish a phase index based on the occurrence time of the stimulus pulse, the difference between each sampling time and the pulse trigger index of its corresponding stimulus pulse is calculated as the stimulus phase offset for each sampling time. This stimulus phase offset characterizes the phase shift between each sampling time and its pulse occurrence time index. If the stimulus phase offset is greater than the length of the stimulus period, subsequent processing is stopped until the next pulse triggers. The length of the stimulus period is the reciprocal of a preset stimulus frequency; in one specific implementation of this invention, the preset stimulus frequency is 130Hz.

[0048] If the saturation mask at the k-th sampling time is 0, the neural residual signal is calculated as follows: , The neural residual signal at the k-th sampling time. The mixed field potential signal at the k-th sampling time. The stimulus phase shift at the k-th sampling time in the artifact waveform vector The corresponding element value in This represents the stimulus phase offset at the k-th sampling time.

[0049] If the saturation mask at the k-th sampling time is 1, it indicates that the current mixed field potential signal has been distorted and does not contain effective neural information. In this case, the neural residual signal is directly set to 0 to avoid introducing erroneous pathological information.

[0050] Preferably, in some implementations of the present invention, after determining the neural residual signal at the k-th sampling time, the mixed field potential signal at the k-th sampling time is used to analyze the artifact waveform vector. The corresponding elements in the data are updated to track impedance drift in real time. In the formula, The preset forgetting factor is set to 0.01. Here, is the value of . The calculation result replaces the stimulus phase offset at the k-th sampling time in the artifact waveform vector. The corresponding element value This enables the manipulation of the artifact waveform vector. Update the corresponding elements in the table.

[0051] Preferably, in some implementations of the present invention, the method for obtaining the waveform slope factor includes: To provide a reliable physical constraint for subsequent dynamic programming algorithms, the absolute value of the neural residual signal is calculated as the instantaneous residual energy at each sampling time. This instantaneous residual energy characterizes the absolute strength of the neural residual signal and serves as the observation benchmark for subsequent state estimation.

[0052] Based on the differences in neural residual signals at adjacent sampling times, and combined with the instantaneous energy of the residuals, a waveform slope factor is obtained for each sampling time. This waveform slope factor characterizes the normalized rate of change of the waveform in the neural residual signal and is used to identify smooth capacitive discharge artifacts. If the waveform slope factor is significantly lower than the statistical mean of the neural residual signal, it indicates that the waveform in the neural residual signal is smooth at this time, and is suspected to be a smooth capacitive discharge artifact.

[0053] In one specific implementation of this invention, the waveform slope factor is calculated as follows: In the formula, Let be the waveform slope factor at the k-th sampling time. and These are the neural residual signals at the k-th and (k+1)-th sampling times, respectively. Let the residual instantaneous energy be the value at the k-th sampling time. This is a preset small constant, its purpose is to avoid the denominator from being 0, and to prevent it from being 0. .

[0054] Preferably, in some implementations of the present invention, the method for obtaining the state observation value includes: To address the issues of nonlinear artifact interference and loss of pathological feature information in mixed field potential signals, it is necessary to transform the physical features in the mixed field potential signals into an optimal path search problem at the mathematical level. By constructing a cost function that includes artifact morphology penalties and physiological inertia constraints, the pathological intensity trajectory that best conforms to physiological laws can be found in the discrete state space.

[0055] Therefore, the ratio of the residual instantaneous energy to the preset maximum signal amplitude is calculated as the first ratio, representing the normalized result of the maximum value of the residual instantaneous energy. The first ratio is mapped using a preset state resolution, and the mapped result is rounded and compared with the preset state resolution to obtain the state observation value at each sampling time. This state observation value eliminates the physical dimension of the residual instantaneous energy, more clearly reflecting the state observation characteristics caused by the residual instantaneous energy.

[0056] In one specific implementation of this invention, the method for calculating the state observation value is as follows: In the formula, This represents the state observation value at the k-th sampling time. The function is for finding the minimum value, where L is the preset state resolution. The preset maximum signal amplitude is set to 5, where the preset state resolution is 100 and the preset maximum signal amplitude is 5. , This is the first ratio. During subsequent dynamic programming algorithm calculations, all terms involving the instantaneous energy of the residuals are converted into corresponding state observations.

[0057] Preferably, in some implementations of the present invention, the method for obtaining the artifact morphological penalty term includes: To address the dimensionality discrepancy between the microvolt-level neural residual signal and the high-weight penalty term during computation, a preset state index and its corresponding preset state signal value are pre-defined for each sampling time. The i-th preset state index is denoted as _i_, and the preset state signal value corresponding to the i-th preset state index is _i_. A discrete state set can be formed by all the preset state signal values. Where L is the preset state resolution. Therefore, the preset state signal value elements in the set do not directly correspond to physical voltage values, but rather represent dimensionless index values ​​after mapping. .

[0058] Preferably, in some implementations of the present invention, a physical voltage range is preset. The linear mapping relationship between each preset state index and the physical voltage: In the formula, The physical voltage mapped to the i-th preset state index. The preset maximum signal amplitude is set to 5. .

[0059] Since the separated neural residual signal may still retain a nonlinear capacitive waveform with smooth characteristics, and there may be a change in amplitude jump at the moment when the front-end analog amplifier exits saturation, a nonlinear artifact morphology penalty term needs to be constructed in order to actively suppress the misjudgment of non-neural signals as pathological outbreaks.

[0060] Therefore, for each sampling moment, the pulse trigger time of the stimulus pulse to which each sampling moment belongs is obtained, and the difference between each sampling moment and the pulse trigger time of its corresponding stimulus pulse is calculated as the stimulus phase offset for each sampling moment, representing the phase offset between each sampling moment and its pulse trigger time. If the stimulus phase offset belongs to an element within the sensitive time window of the preset capacitor discharge, and the waveform slope factor is less than the preset waveform smoothing threshold, and the preset state signal value corresponding to each preset state index is greater than the preset safety noise floor threshold, then the artifact morphology penalty term for each preset state index at each sampling moment is determined according to the preset strong suppression coefficient and the preset state signal value. The sensitive time window of the preset capacitor discharge is the interval from 2ms to 10ms after the pulse ends, the preset waveform smoothing threshold is 0.1, the preset safety noise floor threshold is 30, and the preset strong suppression coefficient is [value missing]. .

[0061] Among them, the stimulus phase offset and the preset state signal value help to accurately determine whether the current sampling time is in a phase that is prone to capacitor discharge, while the waveform slope factor helps to accurately identify whether the signal waveform exhibits abnormal smoothness. Therefore, the artifact morphology penalty term represents that if the current sampling time is in a phase that is prone to capacitor discharge and the waveform exhibits abnormal smoothness, a higher cost is imposed on the high amplitude state estimation, which is used to subsequently achieve active suppression of false signals.

[0062] If the preset capacitor discharge triggering condition is not met, then the artifact morphology penalty term for each preset state index at each sampling time is set to 0.

[0063] In one specific implementation of this invention, the method for calculating the artifact morphological penalty term is as follows: In the formula, The artifact morphology penalty term is the i-th preset state index at the k-th sampling time. The preset strong suppression coefficient is a dimensionless weight set based on the normalized space, and its value is [value missing]. The triggering condition is that the stimulus phase shift at the k-th sampling time belongs to an element within the sensitive time window of capacitor discharge, i.e. And the waveform slope factor is less than a preset waveform smoothing threshold, i.e. Furthermore, the preset state signal value corresponding to each preset state index is greater than the preset safety noise threshold, i.e. ,in, It is the set of all elements within the sensitive time window of capacitor discharge. To preset the waveform smoothing threshold, Set a preset safety noise threshold.

[0064] The artifact morphological penalty term characterizes the imposition of a higher cost on high-amplitude state estimation when the current sampling time is in a phase prone to capacitor discharge and the waveform exhibits abnormally smooth characteristics. This cost is then used to actively suppress spurious signals. The subsequent dynamic programming algorithm forces the optimal path to avoid high-amplitude regions and select low-amplitude states, thereby achieving active suppression of spurious signals.

[0065] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the path cumulative cost includes: In order to accurately construct a cost function that includes artifact morphology penalty and physiological inertia constraint, and thus find the pathological intensity trajectory that best conforms to physiological laws in the discrete state space, in particular, for the first sampling time, since there are no sampling times before the first sampling time, the path cumulative cost of each preset state index of the first sampling time is directly set to 0, indicating that the initial state is unknown.

[0066] For each sampling time other than the first sampling time, the observation reliability of each sampling time is analyzed, and the observation reliability is used as the weight of the first deviation to construct an observation fitting term to weigh the reliability of the observation data.

[0067] Using the artifact morphological penalty term of each preset state index at each sampling time as a morphological penalty term helps to make the physical waveform more reasonable.

[0068] A set of state indices to be analyzed is constructed with each preset state index at the current sampling time as the center; the state indices to be analyzed at the previous sampling time are compared with the corresponding preset state indices at the current sampling time, and the path cumulative cost under the state indices to be analyzed at the previous sampling time is combined to construct a state transition term, which conforms to the change in physiological inertia.

[0069] Based on the observation fitting term, morphological penalty term, and state transition term, the path cumulative cost for each preset state index at each of the remaining sampling times is obtained. This path cumulative cost non-uniformly balances the reliability of the observation data with the rationality of the physical waveform, and also utilizes physiological inertia to search for smooth transition paths during signal loss gaps, which helps to solve the problem of signal fragmentation in pathological signals.

[0070] In one specific implementation of this invention, the method for calculating the cumulative path cost for each preset state index at each of the remaining sampling times is as follows: In the formula, The cumulative cost of the path for the i-th preset state index at the k-th sampling time. for The set consisting of all the preset state indices, The cumulative cost of the path for the j-th preset state index at the (k-1)-th sampling time. The preset physiological inertia weight is set to 1. and These are the preset state signal values ​​corresponding to the i-th and j-th preset state indices, respectively. This is a function that minimizes the value. It should be noted that the formula for calculating the cumulative path cost includes... .

[0071] In the formula for calculating the cumulative path cost, the observation fitting term is derived using the observation confidence level. Weighting is applied, and the observation reliability decreases when hardware saturation occurs. This makes the observation fitting term zero, indicating that the dynamic programming algorithm completely ignores the current state observations and relies solely on physiological inertia for prediction. The morphological penalty term introduces physical constraints to suppress morphological artifacts. The state transition term is for local search optimization; however, to reduce computational complexity, this embodiment does not traverse all preset state indices from the previous time step. Instead, it adopts a local neighborhood search strategy, defining the search radius. The search scope is limited to The preset physiological inertia weight is used to penalize drastic changes in state observations. The state transition term operation is beneficial for subsequent dynamic programming algorithms to find the optimal predecessor index that minimizes the sum of the previous time step cost and the jump cost. This is then recorded in the path pointer matrix. The calculation of the cumulative path cost employs a local search strategy to reduce computational complexity from... Reduce to This meets the low power consumption requirements of implantable chips.

[0072] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the observation confidence level includes: To force subsequent dynamic programming algorithms to rely entirely on physiological inertia for state prediction (Blind Prediction) and thus accurately eliminate the interference of erroneous observations, for each sampling time, if the saturation mask at that sampling time is 1, it is in a hardware saturation state, at which point effective neural information is completely lost. This forces the observation confidence at that sampling time to be 0, making the observation weight at this time zero, thus forcing subsequent dynamic programming algorithms to rely entirely on physiological inertia for state prediction (Blind Prediction) and thus accurately eliminating the interference of erroneous observations. If the saturation mask at the sampling moment is 0, the observation reliability at that sampling moment is calculated by comparing the neural residual signal and the mixed field potential signal within a preset short time window. The closer the magnitudes of the neural residual signal and the mixed field potential signal are, the higher the observation reliability at the current sampling moment. Therefore, this observation reliability characterizes the reliability of the observation signal data at each sampling moment and is used as a key interface connecting the physical layer and the mathematical layer in the subsequent dynamic programming algorithm.

[0073] In one specific implementation of this invention, the method for calculating the observation confidence level is as follows: In the formula, The observation confidence level at the k-th sampling time is... To find the minimum value function, The mean of the neural residual signals is the sum of the M nearest sampling times before the k-th sampling time. The mean of the mixed field potential signal is the sum of the M most recent sampling times before the k-th sampling time. A preset small constant is used to avoid the denominator being 0, and its value is 0.1. The formula for observation confidence uses a ratio comparison. At each sampling time in the unsaturated state, the closer the magnitude of the neural residual signal and the magnitude of the mixed field potential signal are, the closer the observation confidence at the current sampling time is to 1.

[0074] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the optimal state index at each historical sampling time based on the path cumulative cost using a dynamic programming algorithm includes: To output a continuous, smooth, and repaired optimal pathological state trajectory in the time domain, thereby addressing the problem of signal fragmentation in pathological signals, it is necessary to output such a trajectory by backtracking along the time axis, based on the path cumulative cost and path pointer information. Since the state transition term in the path cumulative cost calculation formula relies on local search optimization, a fixed-lag strategy is required to achieve a significant improvement in estimation accuracy with minimal delay.

[0075] To support backtracking operations, a path pointer matrix needs to be maintained in real time during the dynamic programming recursion process of calculating the cumulative cost of the execution path. The data structure of this path pointer matrix is ​​defined as a two-dimensional circular buffer with dimensions of . ,in, The length of the backtracking buffer corresponds to the number of lag steps. Its size is the amount of data within 50ms. This is the preset state resolution.

[0076] At the k-th sampling time, for the i-th preset state index The optimal predecessor index was found when calculating the state transition term. : In the formula, The optimal predecessor index for the i-th preset state index. for The function represents finding the function that makes the function The independent variable that achieves its minimum value.

[0077] Then, the optimal predecessor index of each preset state index is stored in the corresponding position in the path pointer matrix: ,in, This represents the modulo operation. express right Take the remainder.

[0078] If the path pointer matrix records the index of the current state at the i-th preset state, then the previous most likely state was the i-th preset state. The topological relationship of a preset state index.

[0079] Since the Viterbi path planning algorithm typically requires waiting for the entire data record to complete, it cannot meet the real-time requirements of closed-loop control. Therefore, this embodiment of the invention employs a fixed-lag backtracking strategy, utilizing the number of lag steps. The future information is used to correct the estimates of past moments. It should be noted that the lag caused by backtracking using the Viterbi path planning algorithm is extremely short relative to the cycle of changes in Parkinson's pathological state and does not affect the clinical efficacy of closed-loop control.

[0080] At the k-th sampling time, perform the following backtracking steps to determine the historical sampling time. Status: (1) Determine the endpoint of the backtracking search: Among the cumulative cost vector composed of the cumulative path costs of all preset state indices at the k-th sampling time, find the preset state index with the smallest value as the endpoint of the backtracking search. : The endpoint of the backtracking search. It represents the preset state index that is most likely to be in the current moment after comprehensively considering all state observations and physical constraints so far.

[0081] (2) Perform reverse tracing: in the path pointer matrix In, from the k-th sampling time, the first... Starting from a preset state index, iterate in reverse order. Step, trace back to historical sampling time Then, set a temporary pointer. Set the loop variable Incrementing from 0 to : ,in, express right Take the remainder. This process... After the next iteration, the temporary pointer Points to historical sampling time Optimal state index .

[0082] This allows us to obtain the optimal state index for each historical sampling moment. This backtracking mechanism can utilize historical sampling moments. after Use observational data within a given time period to correct historical sampling times. State estimation. For example, if historical sampling time In the signal loss region, i.e., the equipment saturation region, unidirectional filtering cannot accurately determine the state; however, if historical sampling times are available... after A clear high-amplitude signal was observed within a certain time period, and the backtracking mechanism will infer the historical sampling time based on physiological inertia. It should be in an intermediate state transitioning to a higher amplitude signal, thus filling the signal gap and achieving smooth reconstruction.

[0083] Preferably, in some implementations of the present invention, the method for obtaining the estimated intensity of the pathological signal includes: To output a continuous, smooth, and repaired optimal pathological state trajectory in the time domain, thereby addressing the signal fragmentation problem in pathological signals, a second ratio is calculated using the optimal state index and the preset state resolution. The optimal state index can infer intermediate states that should transition to higher amplitudes at historical sampling times, thus helping to fill signal gaps in the pathological signal and achieving smooth reconstruction of the optimal pathological state trajectory.

[0084] The product of the second ratio and the preset maximum signal amplitude is calculated as the estimated intensity of the pathological signal at each historical sampling time. This estimated intensity of the pathological signal is a time-domain continuous index that has been smoothed by physical morphological constraints and physiological inertia, solving the problem of signal fragmentation in pathological signals and enabling subsequent accurate optimization of the stimulation pulse amplitude during deep brain stimulation.

[0085] In one specific implementation of this invention, the method for calculating the estimated intensity of the pathological signal is as follows: In the formula, in the formula, Estimate the intensity of the pathological signal at the t-th historical sampling time. Let be the optimal state index at the t-th historical sampling time. This is the second ratio.

[0086] Preferably, in some implementations of the present invention, the method for obtaining the pathological outbreak density includes: To more accurately optimize the amplitude of stimulation pulses during deep brain stimulation, taking into account the severity of motor symptoms in Parkinson's patients... The duration of frequency band neural oscillations is positively correlated, therefore The duration of frequency band neural oscillations can be considered as pathological burst density, reflecting the severity of motor symptoms in Parkinson's patients.

[0087] To accurately characterize the density of pathological outbreaks, it is necessary to address baseline fluctuations across different patients and time periods, and to set reasonable pathological judgment thresholds. Therefore, a preset long-term sliding window and a preset short-term statistical window are pre-defined for each sampling time. The preset long-term sliding window is the most recent 30 seconds prior to each sampling time, used to set reasonable pathological judgment thresholds for each sampling time; the preset short-term statistical window is the most recent 3 seconds prior to each sampling time, used to characterize the density of pathological outbreaks in real time.

[0088] Therefore, the upper quartile of the estimated intensity of the pathological signal at all sampling times within the preset long-term sliding window is calculated as the preset pathological judgment threshold for each sampling time. The pathological judgment threshold is dynamically changed to ensure robustness under different signal gains.

[0089] The number of sampling points whose estimated intensity of pathological signals at all historical sampling times within the preset short-term statistical window is greater than the preset pathological judgment threshold is recorded as the target sampling point number.

[0090] The ratio of the target number of sampling points to the total number of sampling points within the preset short-term statistical window is recorded as the pathological burst density at each sampling moment. This pathological burst density accurately reflects the current pathological load duty cycle. By using this pathological burst density, the amplitude of the stimulation pulse during deep brain stimulation can be accurately optimized, thereby avoiding the problem of unstable symptom control in Parkinson's patients.

[0091] Preferably, in some implementations of the present invention, the method for optimizing the amplitude of stimulation pulses during deep brain electrical stimulation based on the pathological burst density includes: A linear integral control strategy is used to control the amplitude of the stimulus pulse. Dynamic adjustments are made to maintain pathological density within the ideal range.

[0092] Set the target treatment zone as ,in The value is 0.2. The value is 0.4. The adjustment logic is as follows: (1) Maintain: If ,in If the pathological outbreak density is the current sampling time, then the current stimulus intensity is deemed appropriate, and the stimulus pulse amplitude is maintained. constant.

[0093] (2) Enhancement: If If it is determined that the current pathology is not sufficiently suppressed, the amplitude of the stimulation pulse at the next sampling time will be enhanced. In the formula, The amplitude of the stimulus pulse at the next sampling time. The safe voltage limit is set at 5V. The amplitude of the stimulus pulse at the current sampling moment. The preset boost step size is set to 0.1V.

[0094] (3) Reduce: If If it is determined that the current state may be due to excessive inhibition or physiological rest, the amplitude of the stimulation pulse at the next sampling time will be reduced. In the formula, The preset step size is 0.05V.

[0095] Preferably, in some implementations of the embodiments of the present invention, a micro-shaking strategy is implemented during deep brain stimulation: Because frequency aliasing is prone to occur between the sampling rate and the stimulation frequency of the mixed field potential signal, it may cause continuous changes in the artifact morphology, resulting in a long-term low level of observation reliability.

[0096] Therefore, the reliability of real-time monitoring observations If the mean confidence level of all sampling times within the previous 10 seconds is less than the preset failure threshold of 0.1, it indicates that the deep brain stimulation process has been in a state of saturation or strong interference for a long time. In this case, the deep brain stimulation process is judged to have entered an abnormal state, triggering the self-healing process. (1) Micro-diffusing of stimulation frequency: Instead of making large adjustments to the stimulation frequency, a very small random perturbation is applied. The new stimulation frequency Set as: , To preset the stimulation frequency, The small deviation is randomly selected, and the value of the small deviation is... This minute deviation is so small that it is biologically insufficient to alter the therapeutic effect on the neural nuclei, but it is sufficient to physically disrupt the fixed phase-locked relationship between the sampling point and the stimulation pulse, thereby eliminating aliasing interference.

[0097] (2) Artifact waveform vector reset: Set a new stimulation frequency Then, clear the elements within the artifact waveform vector, and then use a new stimulus frequency. Within the initial preset first stimulation cycle of deep brain stimulation, the artifact waveform vector is recalculated. This allows for automatic recovery from environmental parameter drift without manual intervention, ensuring the long-term survivability and safety of the closed-loop control.

[0098] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0099] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for intelligent optimization of deep brain stimulation parameters for Parkinson's disease patients, characterized in that, The method includes the following steps: The mixed field potential signal during deep brain stimulation is acquired. Based on the signal saturation state at each sampling time, artifacts in the mixed field potential signal are eliminated to determine the neural residual signal at each sampling time. Based on the magnitude of the neural residual signal, the instantaneous residual energy is analyzed to determine the waveform slope factor at each sampling time. The residual instantaneous energy is compared with the preset maximum signal amplitude to determine the state observation value at each sampling time; each preset state index and corresponding preset state signal value are preset at each sampling time; the preset capacitor discharge condition is determined based on the waveform slope factor and the preset state signal value; and the artifact morphology penalty term of the preset state index is obtained based on the determination. For each preset state index at the current sampling time, compare it with the preset state index at the previous sampling time, analyze the first deviation between the state observation value and the preset state signal value, and combine it with the pseudo-trace morphology penalty term to obtain the path cumulative cost. Based on the path cumulative cost, a dynamic programming algorithm is used to obtain the optimal state index at each historical sampling time, and the pathological signal intensity at each historical sampling time is estimated using the optimal state index. The pathological burst density at each sampling time is obtained by estimating the temporal distribution of the intensity of the pathological signal, and the amplitude of the stimulation pulse during deep brain electrical stimulation is optimized based on the pathological burst density.

2. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for determining the signal saturation state includes: The difference between the maximum value of the preset range of the ADC analog-to-digital converter and the preset saturation judgment margin is calculated as the first difference. The sum of the minimum value of the preset range of the ADC analog-to-digital converter and the preset saturation judgment margin is used as the first sum value. Based on the comparison results between the mixed field potential signal and the first difference and the first sum, a saturation mask is determined for each sampling moment. A saturation mask of 0 indicates that the current sampling moment is in a non-saturated state; a saturation mask of 1 indicates that the current sampling moment is in a saturated state.

3. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for acquiring the neural residual signal includes: Within the initial preset first parameter of stimulation cycles after deep brain stimulation is initiated, the length of the stimulation cycle is the reciprocal of the preset stimulation frequency. A pseudo-waveform template vector with the same length as the stimulation cycle is constructed. At each sampling moment in the unsaturated state, the mixed field potential signal within the preset first parameter of stimulation cycles is phase-superimposed and averaged. Based on the result of the phase-superimposed and averaged, the initial template value is filled into the pseudo-waveform template vector to obtain the pseudo-waveform vector. The artifacts of the mixed field potential signal are eliminated based on the artifact waveform vector to determine the neural residual signal at each sampling time.

4. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for obtaining the waveform slope factor includes: The absolute value of the neural residual signal is calculated as the instantaneous residual energy at each sampling time. Based on the difference in neural residual signals at adjacent sampling times, and combined with the instantaneous residual energy, the waveform slope factor at each sampling time is obtained.

5. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for obtaining the state observation values ​​includes: Calculate the ratio of the residual instantaneous energy to the preset maximum signal amplitude, and use it as the first ratio; The first ratio is mapped using a preset state resolution, and the mapping result is rounded and compared with the preset state resolution to obtain the state observation value at each sampling time.

6. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for obtaining the artifact morphology penalty term includes: Obtain the pulse trigger time of the stimulus pulse to which each sampling time belongs, and calculate the difference between each sampling time and the pulse trigger time of the stimulus pulse to which it belongs, as the stimulus phase offset of each sampling time; For each sampling moment, if the stimulus phase offset belongs to an element within the sensitive time window of the preset capacitor discharge, and the waveform slope factor is less than the preset waveform smoothing threshold, and the preset state signal value corresponding to each preset state index is greater than the preset safety noise threshold, then the preset capacitor discharge triggering condition is met. Based on the preset strong suppression coefficient and the preset state signal value, the artifact morphology penalty term for each preset state index at each sampling moment is determined. If the preset capacitor discharge triggering condition is not met, then the artifact morphology penalty term for each preset state index at each sampling time is set to 0.

7. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for obtaining the cumulative cost of the path includes: For the first sampling time, the cumulative path cost for each preset state index at the first sampling time is set to 0; For each sampling time other than the first sampling time, the observation confidence of each sampling time is analyzed, and the observation confidence is used as the weight of the first deviation to construct an observation fitting term; The artifact morphological penalty term for each preset state index at each sampling time is used as the morphological penalty term; A set of state indices to be analyzed is constructed with each preset state index at the current sampling time as the center; the state indices to be analyzed at the previous sampling time are compared with the corresponding preset state indices at the current sampling time, and the path cumulative cost under the state indices to be analyzed at the previous sampling time is combined to construct a state transition term; Based on the observation fitting term, the morphological penalty term, and the state transition term, the path cumulative cost of each preset state index at each of the remaining sampling times is obtained.

8. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 7, characterized in that, The methods for obtaining the observation reliability include: For each sampling time, if the signal is in a saturated state at that sampling time, then the observation confidence of that sampling time is set to 0. If the sampling time is in a signal saturation state, the observation reliability at that sampling time is calculated by comparing the neural residual signal and the mixed field potential signal within a preset short time window.

9. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for obtaining the estimated intensity of the pathological signal includes: Calculate the second ratio using the optimal state index and the preset state resolution; The product of the second ratio and the preset maximum signal amplitude is calculated as the estimated intensity of the pathological signal at each historical sampling time.

10. The intelligent optimization method for deep brain stimulation parameters for Parkinson's patients according to claim 1, characterized in that, The method for obtaining the pathological outbreak density includes: A preset long-term sliding window and a preset short-term statistical window are pre-set for each sampling time, and the preset pathological judgment threshold for each sampling time is obtained using the preset long-term sliding window; The number of sampling points in which the estimated intensity of the pathological signal at all historical sampling times within the preset short-term statistical window is greater than the preset pathological judgment threshold is recorded as the target number of sampling points. The ratio of the target number of sampling points to the total number of sampling points within the preset short-term statistical window is recorded as the pathological outbreak density at each sampling time.

Images