Personalized electrical stimulation treatment scheme generation system and method

Abstract

The invention belongs to the field of auxiliary medical treatment, and provides a personalized electrical stimulation treatment scheme generation system and method, and the method comprises the steps: obtaining a voice signal of a patient, and extracting voice dynamic change information; acquiring neuromyoelectric signals, heart rate signals and motion information of the patient, and extracting physiological change information; performing association modeling on the voice change information and the physiological change information to generate fusion features for representing a multi-part tremor mode of the patient; performing dynamic generation of personalized electrical stimulation parameters based on the fused features; constraining the dynamic change of the current amplitude, the stimulation frequency and the pulse width within an individually set safety threshold range, and performing protective adjustment in combination with real-time feedback of a patient; and the electrical stimulation module is used for implementing electrical stimulation on a target nerve part in the radial nerve, the median nerve or the ulnar nerve according to the parameters issued by the decision module.

Description

Technical Field

[0001] This invention belongs to the field of assistive medical care, and specifically relates to a system and method for generating personalized electrical stimulation treatment plans. Background Technology

[0002] Tremor is a common symptom of movement disorders, particularly prominent in patients with Parkinson's disease and essential tremor. Tremors not only affect the hands but may also involve the head and voice, severely impacting patients' daily lives and communication abilities. As the disease progresses, tremors in single or multiple sites often exhibit complex rhythms and dynamic changes, making traditional treatments insufficient to address them comprehensively.

[0003] In existing technologies, common treatment methods include drug therapy, deep brain stimulation (DBS), and transcutaneous electrical stimulation (TES). Drug therapy can alleviate some symptoms in the early stages, but long-term use may lead to decreased efficacy or side effects. DBS can suppress tremors to some extent, but this method is invasive, carries high surgical risks, and is expensive. TES, due to its non-invasiveness and reusability, is gradually being applied in clinical and rehabilitation training. However, current TES methods typically use fixed stimulation parameters and lack a mechanism for dynamic adjustment based on individual differences and real-time conditions.

[0004] Because existing electrical stimulation methods largely rely on standardized parameter settings, it is difficult to accurately match them to the multi-site tremor manifestations of different patients. This can easily lead to problems such as large individual differences in efficacy, inadequate control, or overstimulation, thus limiting the effectiveness and comfort of treatment. Therefore, there is an urgent need for a technology that can combine multimodal information to generate personalized electrical stimulation programs to improve the accuracy and safety of tremor treatment. Summary of the Invention

[0005] To address the problems in the prior art, the present invention provides a personalized electrical stimulation therapy plan generation system, comprising: The voice acquisition module is used to acquire the patient's voice signal and extract information on dynamic changes in voice. The physiological data acquisition module is used to acquire the patient's neuromuscular signals, heart rate signals, and movement information, and to extract physiological change information. The feature fusion analysis module is used to correlate and model speech change information with physiological change information to generate fusion features that characterize tremor patterns in multiple parts of the patient. The decision module is used to dynamically generate personalized electrical stimulation parameters based on the fusion features. The decision module makes decisions in the following ways: Identify the dominant location and time period of the patient's tremor, and select the corresponding target nerve stimulation channel based on the dominant location; Based on the fluctuation trend and degree of abnormality in the fusion characteristics, adjust the combination of current amplitude, stimulation frequency and pulse width to suit the patient's current tremor characteristics. When an abnormal increase in a patient's voice or physiological state is detected, the sensitivity of the stimulation parameters is automatically increased, and then gradually reduced back to the comfort range after the state stabilizes. When it is predicted that the electrical stimulation effect is insufficient or there is a risk of discomfort, the stimulation channel will be automatically shut down or switched. The safety threshold management module is used to constrain the dynamic changes of the current amplitude, stimulation frequency and pulse width within an individually set safety threshold range, and to make protective adjustments based on real-time patient feedback. The electrical stimulation execution module is used to apply electrical stimulation to the target nerve sites of the radial nerve, median nerve, or ulnar nerve according to the parameters issued by the decision module.

[0006] Furthermore, the speech acquisition module includes a preprocessing unit and a feature extraction unit. The preprocessing unit is used to perform noise suppression, frame windowing, and amplitude normalization on the speech signal. The feature extraction unit is used to extract dynamic information on pitch fluctuations, energy fluctuations, speech interruptions, and clarity changes in the speech signal.

[0007] Furthermore, the physiological data acquisition module acquires neuroelectromyographic signals, heart rate signals, and kinematic information through various sensing devices deployed on the patient's limbs or body surface. The kinematic information is acquired by a triaxial accelerometer or gyroscope and is used to describe the amplitude and rhythm of hand, head, or trunk tremors.

[0008] Furthermore, the feature fusion analysis module adopts a unified time reference to align the speech segments with the physiological signal segments using timestamps and normalize their scales, and calculates the synchronicity, energy difference, and phase coupling indices within a window.

[0009] Furthermore, the feature fusion analysis module adopts a hierarchical fusion architecture, generating intermediate fusion features of speech and physiological signals at the feature layer, constructing a dominant part recognition model based on the fusion features at the decision layer, and outputting a comprehensive judgment of tremor pattern in combination with historical state evolution curves.

[0010] Furthermore, the decision-making module includes a graded response mechanism that divides the tremor state into three levels: mild, moderate, and severe. In the mild case, a low level of background stimulation is maintained; in the moderate case, the frequency and amplitude are increased; and in the severe case, a reinforcement strategy including pulse width extension and multi-channel coordination is implemented.

[0011] Furthermore, when the decision module detects that the tremor is intensifying, it triggers a rapid response mechanism to increase the current amplitude and frequency, and executes a parameter easing strategy after the condition eases, so that the current amplitude and pulse width gradually return to the comfort range.

[0012] Furthermore, the safety threshold management module sets safety boundaries for current amplitude, frequency, and pulse width based on the individual patient's tolerance during system initialization, and dynamically adjusts these boundaries in conjunction with body surface impedance and patient feedback during daily treatment.

[0013] Furthermore, the electrical stimulation execution module includes a wearable electrode assembly and a multi-channel output circuit, which can independently control the radial nerve, median nerve, and ulnar nerve.

[0014] The present invention also provides a method for generating personalized electrical stimulation treatment plans, which uses the aforementioned personalized electrical stimulation treatment plan generation system to generate personalized electrical stimulation treatment plans.

[0015] This invention, through the fusion analysis of voice information and physiological data, can more comprehensively reflect the multi-site characteristics of a patient's tremor, avoiding the biases caused by single-signal analysis. Synchronous modeling of voice, electromyography, heart rate, and kinematic signals on a time scale enables the system to accurately identify the dominant site and evolution trend of the tremor, thus providing a reliable basis for the generation of individualized parameters.

[0016] During the generation of treatment parameters, the system incorporates a dynamic decision-making mechanism. It adjusts the combination of current amplitude, stimulation frequency, and pulse width in real time based on the intensity and changing state of the tremor. This allows for rapid enhancement of control when the tremor worsens and automatic return to a comfortable range when the condition subsides, ensuring a balance between therapeutic efficacy and tolerability. This differentiated response enables the electrical stimulation process to truly meet the individualized needs of each patient.

[0017] Furthermore, this invention incorporates a safety threshold management and multi-channel execution mechanism, which can constrain changes in stimulation parameters within individually set limits and flexibly switch stimulation channels when needed. This design ensures the safety of the stimulation process while enhancing adaptability to complex tremor manifestations, resulting in treatment that is significantly superior to existing technologies in terms of stability, precision, and continuity. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments 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.

[0019] Figure 1 This is the overall schematic diagram of the present invention; Figure 2 This is a schematic diagram of a layered convergence architecture; Figure 3 This is a state diagram of a hierarchical response mechanism. Detailed Implementation

[0020] The personalized electrical stimulation treatment plan generation system of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. However, this embodiment should not be construed as limiting the scope of protection of the present invention.

[0021] Example 1: This example provides an implementation method for a personalized electrical stimulation treatment plan generation system for patients with vocal tremor as the main symptom. Its overall structure is as follows: Figure 1 As shown, this system acquires the patient's voice signals during vocalization through a voice acquisition module, and combines this with neuromuscular signals, heart rate signals, and motion information simultaneously acquired by a physiological data acquisition module to perform correlation analysis between voice abnormalities and physiological fluctuations. Based on these fusion features, the system can identify the dominant role of vocal tremor in the patient's overall tremor pattern, thereby dynamically generating targeted electrical stimulation parameter combinations in the decision-making module and implementing them on the target nerve sites through the electrical stimulation execution module, in order to improve vocal tremor while taking into account both overall treatment effectiveness and patient comfort.

[0022] The specific implementation of the system is as follows: The voice acquisition module is used to acquire continuous voice signals during daily communication, reading aloud, voice training, or other language expression activities. This module can be implemented in various ways, such as using a built-in microphone on a mobile terminal, an external high-sensitivity pickup device, or a wearable sound acquisition device that fits over the face and throat area. Through the coordination of different hardware configurations, accurate capture of the patient's voice signals can be ensured in various environments, including home, clinic, or noisy public places, guaranteeing the usability of subsequent analysis.

[0023] The voice acquisition module includes a preprocessing unit for processing the acquired raw voice signal to improve signal quality and stability. The preprocessing includes the following steps: Noise suppression: Environmental noise estimation and filtering algorithms are used to reduce background noise. For example, in speech signals collected in public places, noise from air conditioners, conversations, or vehicles is removed to highlight the main components of the patient's speech.

[0024] Framing and windowing: The continuous speech signal is segmented into segments of fixed time length, such as 20 milliseconds per frame, and a Hamming window is introduced at the frame boundaries to reduce spectral leakage and improve the accuracy of short-time analysis.

[0025] Smoothing and normalization: Smoothing is performed on speech signals with excessive amplitude variations, while adjusting the signal energy to a uniform amplitude range. For example, the volume is kept at a standardized level across different patients or different acquisition devices to facilitate comparative analysis of different batches of data.

[0026] The speech acquisition module also includes a feature extraction unit for characterizing the speech after signal preprocessing. The purpose of this characterization is to extract dynamic change information reflecting the patient's tremor state, thereby characterizing subtle fluctuations in the speech in both the time and amplitude domains. Specifically, the dynamic change information includes: Temporal fluctuations in speech pitch: used to reveal the instability of vocal cord vibration. For example, when reading "one, two, three, four, five", the fundamental frequency of a normal person's speech is stable, while the fundamental frequency of a tremor patient's speech shows periodic tremors.

[0027] The amplitude of changes in vocal energy over time: used to characterize abnormal fluctuations in the intensity of sound. For example, when a patient continuously utters "ah—", the volume of the sound should be stable, but in a tremor state, there will be periodic fluctuations in intensity.

[0028] Interruptions and continuity of speech: This reflects the discontinuity in the process of phonation. For example, when reading the short sentence "The weather is very nice today", the patient's speech may have involuntary short pauses in the middle.

[0029] Stability of speech intelligibility: used to reveal distortion and irregularity of speech signals. For example, the word "mother" may sound muffled or prolonged in the mouth of a tremor patient due to unstable pronunciation.

[0030] Through the above preprocessing and feature processing, the speech acquisition module can output a set of stable, comparable and intuitive information that reflects the dynamic changes in voice tremor, providing reliable data support for the modeling of the feature fusion analysis module.

[0031] The physiological data acquisition module is used to collect multi-source physiological signals at different stages of the treatment process through sensing devices placed on the patient's limbs or body surface. These signals include neuromuscular signals, heart rate signals, and kinematic information to comprehensively reflect the patient's physiological responses and motor performance during tremor.

[0032] The neuromyomyography (EMG) signals are collected by attaching surface EMG electrodes to the muscles of the patient's forearm, back of hand, or neck. These signals are used to characterize the contraction frequency and intensity of muscle fibers during tremor. For example, when the patient maintains a stationary outstretched arm posture, normal muscle signals should exhibit stable low-frequency discharges, while the EMG signals of a tremor patient will show periodic high-amplitude discharges of 4Hz to 8Hz, thus reflecting abnormal rhythmic muscle contractions.

[0033] The heart rate signal is acquired via a photoplethysmography (PPG) sensor or electrocardiogram (ECG) electrodes to analyze the regulatory state of the patient's autonomic nervous system during tremor. When the patient enters a state of emotional tension or when the tremor worsens, the heart rate often experiences a brief increase, while the heart rate variability index decreases. For example, when the patient transitions from a quiet sitting posture to a talking state, the heart rate rises from 72 beats / minute to 90 beats / minute, accompanied by a narrowing of the heart rate interval fluctuation range, indicating abnormally enhanced autonomic nervous activity.

[0034] The kinematic information is collected using accelerometers or gyroscopes fixed to the patient's wrist, head, or torso to describe the amplitude and rhythmic characteristics of the tremor. For example, in hand tremor monitoring, triaxial acceleration signals show repetitive oscillations of ±3 degrees; in head tremor monitoring, gyroscopes detect periodic rotational motions of approximately 5 Hz; and during patient walking, kinematic information can reveal asymmetrical tremor interference in the gait.

[0035] The physiological data acquisition module further performs joint time-domain and frequency-domain analysis on the aforementioned multi-source signals. Time-domain analysis is used to identify signal fluctuation trends and irregular changes, such as detecting changes in the root mean square value of electromyographic signals; frequency-domain analysis is used to reveal the dominant frequency and harmonic distribution of tremor, such as identifying the energy peaks of acceleration signals in the 4Hz to 12Hz frequency band. Through these analyses, the system can extract physiological change information reflecting the patient's tremor pattern and provide reliable input data for subsequent feature fusion analysis.

[0036] The feature fusion analysis module is used to model the correlation between dynamic changes in speech and physiological changes under a unified time reference, forming a fused feature that reflects the overall state of tremor in multiple sites. Although speech and physiological signals originate from different sources, they are both influenced by the same neural motor control pathway, often exhibiting similar rhythmic changes and mutually reinforcing fluctuation trajectories during the occurrence and exacerbation of tremor. Based on this common cause relationship, by performing time alignment, quality assessment, and correlation analysis on the two types of signals, it is possible to simultaneously characterize the strength relationship and dominant sites of vocal tremor and limb tremor within a single time period, and provide a more robust description of the evolutionary trend of the state.

[0037] In one implementation, the feature fusion analysis module first establishes a unified timeline. The speech dynamics information output by the speech acquisition module and the neuromyomyography, heart rate, and kinematics information output by the physiological data acquisition module are resampled to a unified step size after timestamp alignment and sampling rate tuning. In scenarios with speech activity detection, time slices are segmented using speech activity segments as anchor points, and electromyography and kinematic segments are simultaneously extracted within each time slice. This ensures that multimodal information within the same segment corresponds to the same physiological state, reducing bias caused by cross-segment splicing.

[0038] Subsequently, quality and reliability assessments were conducted for both types of signals. For speech segments, a segment quality score was calculated based on environmental noise levels, recording amplitude stability, and continuity. For physiological segments, a segment reliability score was determined based on indicators such as electrode contact impedance, heart rate signal stability, and motion sensor saturation. During the fusion process, segments with lower quality scores were downweighted or compensated to prevent any single modality from dominating the overall assessment under low-quality conditions.

[0039] For example, in speech segments collected in noisy dining environments, if there is still a high residual noise after noise suppression, the weight of that speech segment in the fusion judgment is reduced, and more reliance is placed on the electromyographic and kinematic information collected at the same time.

[0040] After completing time alignment and quality assessment, the module performs scale unification and time windowing on each modal feature. To eliminate amplitude differences between individuals and devices, speech and physiological features are normalized to make them comparable on the same scale. Then, the time axis is segmented using a sliding window, with the window length and step size preferably covering one to several tremor cycles to ensure the complete presentation of rhythmic information. Within each window, the degree of synchronization and phased trends of speech and physiological features are calculated to capture short-term linkage patterns and slow drifts.

[0041] The module further calculates the correlation and consistency metrics between multimodalities to identify synchronous tremors and dominant sites. For the same window, if speech pitch fluctuations and forearm electromyography (EMG) intensity simultaneously show periodic increases within similar frequency bands and their phase relationship is stable, then a sound-upper limb linked tremor is identified within that window. If the head gyroscope shows significant rhythmic oscillations under low-volume speech or silence conditions, while speech changes remain stable, then the dominant tremor within that window is identified as originating from the head. In segments with short-term stress, if the heart rate increases and is accompanied by a decrease in heart rate variability, while speech energy shows intermittent declines, even if the kinematic amplitude has not yet increased significantly, this state can be marked as a possible precursor to tremor exacerbation, providing an advance warning for subsequent decision-making.

[0042] To improve the stability of the fusion results, the module performs smoothing and consistency checks on the outputs of temporally adjacent windows to suppress momentary misjudgments caused by occasional noise. When different modalities give contradictory judgments, segment quality scores and historical consistency are introduced as arbitration criteria. Only when a consistent trend appears within several consecutive windows will the change of the dominant part and the upgrade of the state level be written into the fusion result to avoid control instability caused by frequent switching.

[0043] In a preferred embodiment, the fusion features output by the module are presented in a structured description, including but not limited to the following elements: time stamp, quality score of each modality segment, dominant tremor location in the current time period, intensity contrast between vocal and limb tremors, synchronicity score, state level (stable, fluctuating, aggravated), and short-term trend indication (slowly rising, flat, slowly declining). This structured result is convenient for direct use by the decision-making module and also facilitates visualization and tracking in medical scenarios.

[0044] The following are some specific examples to illustrate the role and effect of the fusion process.

[0045] Example 1: When reading a 30-second segment of numbers, the speech segment shows a stable rhythmic fluctuation around 5 Hz. At the same time, the forearm electromyography segment also shows a similar rhythmic increase in discharge. The synchronization score of the two is high. Based on this, the module determines that the period is a linkage tremor between voice and upper limb and gives a fusion label of "upper limb-voice coordination".

[0046] Example 2: The patient is sitting quietly without making a sound. The speech segments are of high quality but change slowly. The head gyroscope shows repetitive oscillations in the range of four to six Hz. No significant enhancement of electromyography is observed. The module outputs a fusion result of "head-dominant" and marks the state level as "fluctuation".

[0047] Example 3: During a conversation in a noisy coffee shop, the voice segment has a low quality score due to the loud background noise; at the same time, the wrist acceleration and electromyography show a slight synchronous increase, and the heart rate rises from more than 70 beats / minute to more than 90 beats / minute. The module reduces the weight of the voice and increases the weight of the physiological modality, outputting a trend indication of "potential aggravation", so that subsequent parameter decisions can take the inhibitory strategy in advance.

[0048] When handling missing or abnormal data, the module has a robustness strategy. If a modality has missing or insufficient data in a single window, the fusion judgment is temporarily completed based on the other modalities, and the data is recalibrated when the window is restored. When multiple modalities fail to meet the quality standards at the same time, the module outputs a "low confidence" flag and prompts the decision-making module to adopt a conservative strategy or maintain the predetermined parameter upper limit to ensure the safety and controllability of the overall process.

[0049] Through the above design, the fusion results can maintain consistency under multiple scenarios, environments, and tasks, providing a reliable basis for the target selection, intensity setting, and timing control of subsequent electrical stimulation parameters.

[0050] The decision-making module, after acquiring the fusion features output by the feature fusion analysis module, dynamically generates personalized electrical stimulation parameters and enables real-time adjustment and safety management of these parameters throughout the treatment process. This module is designed to allow the stimulation path and parameters to dynamically adapt to changes in the patient's tremor performance, addressing the technical problems of unstable efficacy and insufficient response to multi-site tremors in traditional fixed-parameter modes for different patients.

[0051] In one specific implementation, the decision module first identifies the dominant site of tremor in the patient at different time periods. This process is based on the fusion of dynamic speech changes and physiological changes, comparing the intensity of each modality within a specific time window to determine whether the tremor originates primarily from the voice, hands, or head. If a sustained increase in speech pitch fluctuations within the 4-6 Hz range is detected, while electromyography and acceleration signals remain stable, the period is determined to be dominated by vocal tremor. If, under sedentary conditions, the head gyroscope signal shows periodic oscillations exceeding ±5 degrees, while speech and hand signals show no significant fluctuations, the period is determined to be dominated by head tremor. Through this identification mechanism, the system can concentrate stimulation on neural pathways related to the affected area. For example, when vocal tremor is dominant, the median and ulnar nerve pathways are preferentially stimulated; when hand tremor is dominant, the radial nerve pathway is preferentially stimulated, thereby improving the targeting and efficiency of stimulation.

[0052] After determining the dominant site, the decision module determines the combination relationship between current amplitude, stimulation frequency, and pulse width based on the fluctuation trend and abnormality degree in the fusion features. Current amplitude directly affects stimulation intensity, frequency determines the rhythm of nerve excitation, and pulse width determines the pulse duration. These three are coupled and need to be dynamically coordinated according to different situations. When the fusion features show a slow upward trend in tremor amplitude, the system appropriately increases the current amplitude to enhance the inhibitory effect; when the detected tremor frequency increases from 4Hz to 8Hz, the stimulation frequency is appropriately increased to achieve frequency-following inhibition; when the patient's condition tends to improve, the pulse width is shortened and the amplitude is reduced, allowing the stimulation to gradually return to a comfortable daily level. For example, if a rapid increase in hand electromyographic signal energy is detected during a patient's writing process, the system's output electrical stimulation parameters can be adjusted to a current amplitude of 3.5mA, a stimulation frequency of 130Hz, and a pulse width of 250μs; if the patient's speech fluctuates intermittently while reading aloud, the system automatically adjusts to a current amplitude of 2.0mA, a frequency of 100Hz, and a pulse width of 200μs to better suit vocal tremors.

[0053] When a significant worsening of a patient's speech or physiological state is detected, the decision-making module further executes a rapid response mechanism to enhance the sensitivity of stimulation parameters. Such worsening manifests as a short-term, sharp decrease in speech energy, an increase in heart rate, or a sudden increase in hand acceleration amplitude. To prevent the rapid spread of tremor, the system increases the stimulation intensity within a very short time. For example, if a sudden interruption occurs in the speech signal while the patient is talking, and the heart rate rises from 75 beats / minute to 95 beats / minute, the system immediately increases the current amplitude from 2.5 mA to 4.0 mA and extends the pulse width by 50 μs. During walking, when the peak-to-peak value of the wrist acceleration signal suddenly increases from ±2 degrees to ±6 degrees, the system immediately increases the frequency from 90 Hz to 140 Hz to enhance the neuromodulation effect. After the patient's condition gradually stabilizes, the decision-making module automatically executes a parameter easing strategy, gradually reducing the current amplitude and pulse width to the comfort threshold to ensure a balance between therapeutic efficacy and tolerability.

[0054] When the system predicts that electrical stimulation may not achieve the expected therapeutic effect, or detects that the stimulation exceeds the individual's safety limits, the decision-making module executes protective measures. These protective measures include automatically shutting down electrical stimulation or switching to a backup stimulation channel. If, within multiple consecutive time windows, fusion characteristics show no decrease in tremor amplitude, and the patient subjectively reports a stinging or burning sensation, the system determines that the current stimulation is ineffective or unsuitable for continued maintenance and immediately triggers a shutdown command. When the electrode contact impedance of a stimulation channel increases above a preset threshold, indicating poor electrode adhesion or reduced local tissue tolerance, the system executes a channel switching strategy, switching to a backup channel to output stimulation parameters. For example, if the contact impedance in the right radial nerve channel increases to 20kΩ due to electrode loosening, the system automatically stops outputting from that channel and switches to the median nerve channel to maintain treatment continuity.

[0055] Through the aforementioned mechanism, the decision-making module achieves targeted selection of stimulation channels at the spatial level and dynamic adjustment of current amplitude, stimulation frequency, and pulse width at the temporal level. Combined with rapid response and protective measures, it constructs a complete parameter generation and control process. Compared to traditional stimulation methods that rely on fixed parameters, this module significantly improves individualization, enabling different patients in different states to achieve more precise treatment effects, while reducing the adverse risks of overstimulation, thus balancing efficacy, safety, and comfort.

[0056] The safety threshold management module is used to constrain the real-time adjustment of current amplitude, stimulation frequency and pulse width to ensure that the treatment process is carried out within an individualized tolerable range and to avoid discomfort or tissue damage caused by excessive parameters.

[0057] In a specific embodiment, the system completes threshold setting by medical staff or patients under guidance during the initialization phase. This threshold setting includes the maximum acceptable value of the current amplitude, the upper limit range of the stimulation frequency, and the safety limit of the pulse width. The setting process is usually carried out in combination with clinical evaluation and the subjective feelings of the patient. For example, by gradually increasing the current amplitude, the first perception, comfort threshold, and maximum tolerable value of the patient are recorded, and the comfort threshold and the maximum tolerable value are used as the constraint boundaries for the system operation. On this basis, the system forms individualized parameter upper limits as the initial safety reference for subsequent treatments.

[0058] During the daily treatment process, the safety threshold management module conducts secondary verification on the electrical stimulation parameters generated by the decision-making module. The verification process not only compares with the initialization threshold but also combines the changes in the surface impedance measured in real time and the immediate feedback from the patient. When the electrode attachment state is good and the impedance is within the normal range, the system allows dynamic adjustment of parameters within the threshold range; when poor electrode contact causes an increase in impedance, the system determines that the local tissue tolerance has decreased, and even if the parameters do not exceed the preset upper limit, a reduction measure will be triggered to avoid local overstimulation. For example, when the impedance of the electrodes attached to a patient's wrist increases from 5 kΩ to 15 kΩ, the system automatically reduces the current amplitude from 3.0 mA to 2.0 mA and prompts the user to check the electrode fitting condition.

[0059] If it is detected that the parameters generated in real time exceed the individualized threshold range, the module immediately triggers a protective adjustment mechanism to forcibly limit the parameters within the safe range. For example, when the decision-making module generates a combination of a current amplitude of 5.0 mA and a pulse width of 400 μs during a sudden exacerbation of the patient's tremor, and the safety upper limit for this patient is 4.0 mA and 300 μs, the safety threshold management module will correct the parameters to 4.0 mA and 300 μs and feedback them to the decision-making module, prompting the subsequent strategy to optimize within this range.

[0060] This module not only sets upper limits for single parameters but also can achieve linkage constraints on parameter combinations. For example, for patients with relatively low skin tolerance, when the current amplitude approaches the maximum upper limit, the system automatically limits the pulse width not to exceed a certain threshold to avoid excessive charge per unit area and cause skin tingling. Another example is in a long-term treatment scenario, the module can dynamically lower the upper limit of the frequency according to the cumulative stimulation amount to ensure comfort and tissue safety during the long-term treatment process.

[0061] In a preferred embodiment, the subjective feedback of the patient is also an important basis for safety threshold management. The system provides a feedback interface at the application end. When the patient actively reports a tingling sensation, a burning sensation, or a feeling of excessive muscle contraction, the module immediately triggers a protective adjustment, lowers the current parameters as a whole to the next lower level, and records this feedback in the patient's individual file for optimizing the subsequent threshold range.

[0062] Through the aforementioned mechanism, the safety threshold management module establishes a multi-layered protection chain of "preset boundary—real-time verification—feedback correction." On the one hand, it ensures that treatment parameters operate within an individual's safe range, avoiding adverse reactions caused by overstimulation; on the other hand, through dynamic correction and linkage constraints, the system can maintain stable treatment effects under different environments and conditions. Compared with traditional fixed upper limit safety controls, this module can flexibly adjust according to the patient's real-time condition, significantly improving individualized adaptability and treatment safety.

[0063] The electrical stimulation execution module is used to apply electrical stimulation to at least one target nerve site among the radial nerve, median nerve, or ulnar nerve according to the current amplitude, stimulation frequency, pulse width, and channel selection signals issued by the decision module, so as to achieve targeted intervention on the dominant tremor site.

[0064] In one specific embodiment, the electrical stimulation execution module includes a wearable electrode assembly and a multi-channel output circuit. The electrode assembly is fixed to the patient's forearm region, and the electrode pads are pre-calibrated to correspond to the surface distribution locations of the radial, median, and ulnar nerves, respectively. This arrangement allows the system to act on multiple nerve pathways individually or simultaneously without frequent electrode repositioning. The multi-channel output circuit enables independent control of the three channels, with each channel receiving independent current amplitude, frequency, and pulse width parameters from the decision module, thereby ensuring precise and differentiated stimulation output.

[0065] The design principle of this module lies in the fact that different nerve-innervated muscle groups play different roles in tremor manifestations. The radial nerve mainly controls the wrist and dorsum of the hand extensor muscles, the median nerve mainly controls the forearm flexor muscles and some thumb muscles, and the ulnar nerve mainly innervates the little finger and some intrinsic hand muscles. By precisely selecting and independently controlling different channels, stimulation energy can be concentrated on the most relevant neural pathways, avoiding ineffective intervention on non-dominant areas. For example, when the fusion characteristics indicate that the tremor is dominated by the hand flexor muscles, the system preferentially activates the median nerve pathway and increases its stimulation intensity, while maintaining a lower level or temporarily closing the radial and ulnar nerve pathways.

[0066] The advantage of this module lies in its ability to flexibly switch stimulation targets and maintain stable output. Through independent multi-channel programming, the system can act on a single neural pathway or combine stimulation of multiple channels when tremors are complex. For example, when a patient has both hand and vocal tremors, the system can simultaneously activate the median and ulnar nerve pathways to achieve synergistic intervention on vocalization-related muscle groups and small muscle groups. Furthermore, the multi-channel circuit, through a constant current source and impedance compensation mechanism, can maintain a stable output current when electrode contact states change or skin impedance fluctuates, preventing a decrease in stimulation energy due to increased impedance, thus ensuring treatment continuity.

[0067] Specific application examples include: When a patient is performing a writing task, if increased hand tremor is detected, the decision module sends parameters of 3.0mA, 120Hz, and 250μs to the median nerve channel, while maintaining a background stimulus of 1.0mA to the radial and ulnar nerve channels. At this time, the electrical stimulation execution module will focus its output on the median nerve to suppress flexor muscle tremor. When a patient is reading a sentence, if significant vocal tremor is detected while the hand is stable, the system adjusts the parameters to 2.5mA, 100Hz, and 200μs to the ulnar nerve channel, and closes other channels, thereby focusing stimulation on the muscles related to vocalization. When a patient is walking, if tremor occurs simultaneously in the hand and head, the electrical stimulation execution module can simultaneously activate the radial and median nerve channels, outputting a combination of parameters of 2.0mA, 90Hz, and 220μs to achieve dual-channel coordinated control.

[0068] Through the above methods, the electrical stimulation execution module can flexibly switch or combine target neural channels while maintaining current stability, enabling personalized treatment for different dominant tremor sites. Compared with traditional single-channel, fixed-parameter stimulation methods, this module not only improves the targeting and precision of the stimulation effect but also enhances the flexibility and continuity of the treatment process, thereby significantly improving the overall treatment effect for patients with multi-site tremors.

[0069] Example 2: In this example, the implementation of the feature fusion analysis module is further improved to enhance the comprehensive utilization of flutter information from different modes. For example... Figure 2 As shown, the core of the improvement lies in adopting a hierarchical fusion architecture, which fuses dynamic speech change information and physiological change information at two levels: the feature layer and the decision layer, respectively.

[0070] At the feature layer, the system first performs unified processing on speech signals, electromyography (EMG), heart rate, and kinematic signals to ensure comparability and consistency of multimodal data on the same time reference. Speech signals are typically acquired at a high sampling rate, while EMG and acceleration signals have different sampling frequencies. To eliminate this difference, the system uses timestamp alignment and sampling rate reshaping to map speech frames to the same time window as EMG sampling points, heart rate waveforms, and acceleration data. After this processing, different modalities of data can be analyzed simultaneously on a unified time axis.

[0071] To further improve the comparability of data, the system performs normalization processing on the features of each modality. After amplitude normalization, the volume differences between different patients and under different equipment conditions in speech signals are eliminated; after standardization, electromyographic signals can reflect the true amplitude of muscle contraction without being affected by electrode impedance; heart rate data is normalized through cardiac cycle normalization, which facilitates comparison with other modalities on the same scale; after amplitude normalization, kinematic data can accurately present the relative strength of tremor amplitude under different movements. The multimodal data processed in this way retains its own characteristics and can be cross-analyzed on the same dimension.

[0072] Based on this, the system calculates multiple complementary indicators to generate multi-dimensional intermediate fusion features. The synchronicity correlation coefficient between the fundamental frequency fluctuation of speech and the amplitude of electromyographic discharge is used to reveal whether vocal cord tremor and hand muscle contraction exhibit synchronous oscillations during sound production. When both exhibit periodic activity of approximately 5Hz within the same window, coordinated tremor can be identified. The energy distribution difference between heart rate variability and kinematic amplitude reflects the relationship between the degree of autonomic nervous system tension and the amplitude of limb tremor. For example, when a patient's heart rate variability significantly decreases and the amplitude of hand tremor increases, it indicates that the tremor state is accompanied by abnormal sympathetic nerve excitation. The phase coupling relationship between speech pitch jitter and head rotation signals can reveal whether phonation and head movement exhibit periodic coordination. If the head oscillates at a rhythm of 4Hz~6Hz, and speech jitter increases in the same frequency band, it indicates that both may be driven by the same abnormal center. Through the joint calculation of the above indicators, the system can generate structured intermediate fusion features at the feature layer, without losing the independent expression of each modality, while revealing the interaction relationships between them.

[0073] At the decision-making level, the system utilizes intermediate fusion features to construct a dominant site identification model and makes judgments based on the strength and synchronicity of each modality. When the fundamental frequency fluctuation of speech and the electromyographic energy of the forearm show a strong correlation within the same window, and both frequency peaks are concentrated around 5Hz, the system can determine that the patient is in a state of coordinated tremor of voice and hand. When the speech signal is stable, but the head gyroscope signal continuously increases in the range of 4Hz to 6Hz, it can be identified as head tremor dominating. When the heart rate briefly increases and is accompanied by an increase in the amplitude of the hand acceleration signal, while the speech signal shows no obvious abnormalities, the system determines it as limb tremor dominating and indicates accompanying autonomic nerve activation.

[0074] To reduce misjudgments caused by transient noise, the decision-making layer also incorporates historical state evolution curves for dynamic comprehensive judgment of tremor patterns. The system smooths the intermediate fusion features across multiple consecutive time windows to ensure temporal continuity in pattern recognition. As tremor gradually worsens, the system can identify risks in advance. For example, if a patient's speech tremor amplitude gradually increases and electromyographic energy synchronously enhances over three consecutive time windows while the patient is reading aloud, the system outputs a "coordinated tremor worsening trend" and provides an early warning for the subsequent decision-making module to generate enhanced stimulation parameters.

[0075] By employing a hierarchical fusion approach, the system preserves the independent representation of different modalities at the feature layer and achieves a unified balance at the decision layer, ensuring that the fusion result is both comprehensive and stable. This approach avoids excessive interference from single-modal anomalies in the overall judgment. In noisy environments, speech signals may be distorted due to background noise, but the system can reduce the weight of the speech modality, relying on electromyography and kinematic data to maintain judgment accuracy. Simultaneously, this approach enhances the interpretability of the results, allowing clinicians to trace whether a particular judgment primarily relied on speech, electromyography, or kinematic features, thereby increasing the system's transparency and credibility.

[0076] For example, during a patient's walking test, environmental noise caused distortion in the speech data, but both electromyography (EMG) and kinematic signals showed a significant 6Hz tremor. The system ultimately output "lower limb dominant tremor" and automatically reduced the weight of the speech modality. In another patient's writing task, both speech and EMG showed rhythmic abnormalities. The system output "voice-hand coordinated tremor" and suggested that the electrical stimulation execution module should simultaneously act on the median and ulnar nerve pathways to improve the intervention effect. Furthermore, when a patient was emotionally stressed, their heart rate increased from 70 beats / minute to 95 beats / minute, and the amplitude of hand tremor increased. The system determined this to be "limb tremor dominant with autonomic nervous system abnormality" and suggested increasing the current amplitude to enhance control.

[0077] Example 3: In this example, the processing mechanism of the decision module is further improved to enhance the system's adaptability and stability under different flutter severity levels. For example... Figure 3 As shown, this embodiment proposes a graded response mechanism that divides tremor into three levels: mild, moderate, and severe, and generates differentiated combinations of electrical stimulation parameters for different levels, thereby achieving a graded control strategy that better meets clinical needs.

[0078] In the mild state, the system maintains only a low level of background stimulation. This stimulation, delivered with a small current amplitude and low frequency, is used to maintain the basic excitability of the nerves, preventing neural circuits from entering unstable rhythms, while continuously monitoring subsequent changes in tremor characteristics. This low-level stimulation setting ensures that the patient is not significantly disturbed in their daily life, while still maintaining a stable nervous system. For example, when the patient is quietly reading aloud or engaging in daily conversation, if the amplitude of speech fluctuations is only near the upper limit of normal, the system maintains background stimulation with a current amplitude controlled at 1.0–1.5 mA, a frequency of approximately 60 Hz, and a pulse width of 150 μs, thus allowing the patient to remain comfortable even with mild tremor.

[0079] In the moderate state, the system detects that the amplitude and frequency of tremor have exceeded the individualized threshold, and simple background stimulation is insufficient to suppress the performance. Therefore, it automatically increases the stimulation frequency and moderately increases the current amplitude. The increased frequency can cover the abnormal rhythm, making muscle contraction more synchronized; the increased current amplitude further reduces tremor output. The control focus at this stage is to effectively alleviate tremor while taking into account tolerability, so as to maintain the therapeutic effect during long-term use. For example, during the patient's writing task, if the amplitude of the hand acceleration signal increases to ±4 degrees, and the electromyographic energy is significantly enhanced, the system automatically adjusts the current amplitude to 2.0-2.5 mA, increases the frequency to 100-120 Hz, and maintains the pulse width at 200 μs, thereby enhancing the control of the hand flexor muscles.

[0080] In severe cases, tremors significantly interfere with the patient's movement or speech, making single-parameter adjustments insufficient for control. In such situations, the system employs a reinforcement strategy, including further increasing the current amplitude, extending the pulse width, and activating multi-channel synergistic stimulation as needed. Extending the pulse width increases the charge per unit pulse, thus strengthening the effect on the nerves; multi-channel synergy allows the system to act on multiple neural pathways simultaneously, creating a synergistic effect to rapidly quell the abnormality. For example, if a patient experiences increased head and hand tremors and significant speech interruptions while walking, the system increases the current amplitude to 3.5–4.0 mA, extends the pulse width to 300 μs, and simultaneously activates the radial and median nerve pathways, achieving powerful intervention for multiple tremors within seconds.

[0081] Through a tiered response mechanism, the system maintains comfort and monitoring in mild cases, provides stable and sustainable control in moderate cases, and allows for rapid intervention to ensure immediate treatment in severe cases. A progressive relationship is established between different stimulation intensities, avoiding discomfort and energy waste caused by overstimulation in mild cases while providing sufficient inhibition when tremors worsen. The entire process combines flexible regulation with strong intervention, enabling patients to receive treatment tailored to their individual conditions, both in daily life and during tremors.

[0082] For example, in a patient's reading training, even slight vocal tremors in the early stages triggered background stimuli, and the patient barely noticed the external intervention. During a subsequent signing task, as hand tremors significantly increased, the system entered a moderate response, automatically adjusting parameters to allow for smooth writing. When the patient was emotionally stressed, their speech became intermittent, accompanied by increased head and hand tremors, triggering a severe response. This involved administering multi-channel, high-intensity stimulation for a short period, effectively suppressing tremor propagation. Through this differentiated response process, the treatment achieved dynamic balance and individualized, precise control over different stages of tremor.

[0083] Example 4 provides a method for generating personalized electrical stimulation treatment plans, based on the aforementioned personalized electrical stimulation treatment plan generation system. The method acquires the patient's voice signal through a voice acquisition module, combines it with neuromuscular signals, heart rate signals, and kinematic information acquired by a physiological data acquisition module, and forms fused features reflecting the patient's tremor patterns in multiple locations in a feature fusion analysis module. Under the dynamic generation strategy of the decision module, it outputs an individualized combination of current amplitude, stimulation frequency, and pulse width parameters. After constraint verification by a safety threshold management module, the electrical stimulation execution module finally applies electrical stimulation to the target neural channel, thereby achieving personalized treatment for different tremor states.

[0084] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0085] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0086] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. For some module structures not specifically defined in this invention, the content described in the prior art shall prevail. The prior art mentioned in the foregoing background and specific embodiments sections can be considered as part of this invention and used to understand the meaning of some technical features or parameters.

Claims

1. A personalized electrostimulation therapy regimen generation system, characterized in that, The system includes: The voice acquisition module is used to acquire the patient's voice signal and extract information on dynamic changes in voice. The physiological data acquisition module is used to acquire the patient's neuromuscular signals, heart rate signals, and movement information, and to extract physiological change information. The feature fusion analysis module is used to correlate and model speech change information with physiological change information to generate fusion features that characterize tremor patterns in multiple parts of the patient. The decision module is used to dynamically generate personalized electrical stimulation parameters based on the fusion features. The decision module makes decisions in the following ways: Identify the dominant location and time period of the patient's tremor, and select the corresponding target nerve stimulation channel based on the dominant location; Based on the fluctuation trend and degree of abnormality in the fusion characteristics, adjust the combination of current amplitude, stimulation frequency and pulse width to suit the patient's current tremor characteristics. When an abnormal increase in a patient's voice or physiological state is detected, the sensitivity of the stimulation parameters is automatically increased, and then gradually reduced back to the comfort range after the state stabilizes. When it is predicted that the electrical stimulation effect is insufficient or there is a risk of discomfort, the stimulation channel will be automatically shut down or switched. The safety threshold management module is used to constrain the dynamic changes of the current amplitude, stimulation frequency and pulse width within an individually set safety threshold range, and to make protective adjustments based on real-time patient feedback. The electrical stimulation execution module is used to apply electrical stimulation to the target nerve sites of the radial nerve, median nerve, or ulnar nerve according to the parameters issued by the decision module.

2. The personalized electro-stimulation therapy regimen generation system of claim 1, wherein, The speech acquisition module includes a preprocessing unit and a feature extraction unit. The preprocessing unit is used to perform noise suppression, frame windowing, and amplitude normalization on the speech signal. The feature extraction unit is used to extract dynamic information on pitch fluctuations, energy fluctuations, speech interruptions, and clarity changes in the speech signal.

3. The personalized electro-stimulation therapy regimen generation system of claim 1, wherein, The physiological data acquisition module acquires neuroelectromyographic signals, heart rate signals, and kinematic information through various sensing devices deployed on the patient's limbs or body surface. The kinematic information is acquired by a triaxial accelerometer or gyroscope and is used to describe the amplitude and rhythm of tremors in the hand, head, or trunk.

4. The personalized electrical stimulation therapy plan generation system according to claim 1, characterized in that, The feature fusion analysis module uses a unified time reference to align speech segments with physiological signal segments using timestamps and normalize their scales. It also calculates synchronicity, energy difference, and phase coupling indices within a window.

5. The personalized electrical stimulation therapy plan generation system according to claim 1, characterized in that, The feature fusion analysis module adopts a hierarchical fusion architecture. At the feature layer, it generates intermediate fusion features of speech and physiological signals. At the decision layer, it constructs a dominant part recognition model based on the fusion features and outputs a comprehensive judgment of tremor pattern by combining the historical state evolution curve.

6. The personalized electrical stimulation therapy plan generation system according to claim 1, characterized in that, The decision-making module includes a graded response mechanism that classifies tremor into three levels: mild, moderate, and severe. In mild cases, a low level of background stimulation is maintained; in moderate cases, the frequency and amplitude are increased; and in severe cases, a reinforcement strategy including pulse width extension and multi-channel coordination is implemented.

7. The personalized electrical stimulation therapy plan generation system according to claim 1, characterized in that, When the decision module detects that the tremor is intensifying, it triggers a rapid response mechanism to increase the current amplitude and frequency, and then executes a parameter easing strategy after the condition eases, so that the current amplitude and pulse width gradually return to the comfort range.

8. The personalized electrical stimulation therapy plan generation system according to claim 1, characterized in that, The safety threshold management module sets safety boundaries for current amplitude, frequency, and pulse width based on individual patient tolerance during system initialization, and dynamically adjusts these boundaries during routine treatment by combining body surface impedance and patient feedback.

9. The personalized electrical stimulation therapy plan generation system according to claim 1, characterized in that, The electrical stimulation execution module includes a wearable electrode assembly and a multi-channel output circuit, which can independently control the radial nerve, median nerve, and ulnar nerve.

10. A method for generating a personalized electrical stimulation therapy plan, characterized in that, Personalized electrical stimulation treatment plans are generated using the personalized electrical stimulation treatment plan generation system as described in any one of claims 1-9.

Images