Treatment monitoring method and system for percutaneous ear vagus nerve stimulation device

Abstract

The invention discloses a treatment monitoring method and system for a percutaneous ear vagus nerve stimulation device, and the method comprises the steps: carrying out the multi-source treatment situation monitoring of a target user in a treatment course, obtaining the multi-source treatment situation monitoring data, and constructing a multi-mode treatment situation feature; real-time treatment situation analysis is carried out according to the multi-mode treatment situation features, and real-time treatment situation analysis information is generated; performing dynamic state classification based on the real-time treatment situation analysis information, evaluating the treatment condition of the target user in the current treatment course, and generating a treatment regulation and control scheme to perform treatment management and control on the current treatment course when an abnormal treatment condition exists; and constructing a treatment situation prediction model, obtaining the regulated and controlled multi-mode treatment situation characteristics, generating a floating regulation and control boundary in combination with the treatment situation prediction model, and dynamically correcting the diagnosis and treatment regulation and control scheme by using the floating regulation and control boundary to ensure the curative effect and treatment safety. The treatment effect can be accurately perceived; and the treatment safety is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical devices, in particular to a treatment monitoring method and system for a transcutaneous auricular vagus nerve stimulation device. BACKGROUND

[0002] Vagus nerve stimulation, as an effective neuroregulation technology, has shown significant potential in treating epilepsy, depressive disorders, and inflammatory diseases. Traditional implanted vagus nerve stimulation devices require surgical implantation of electrodes in the neck, which has the problems of high surgical risk, high cost, and postoperative complications. The transcutaneous auricular vagus nerve stimulation device stimulates the auricular vagus nerve distribution area through a non-invasive method, providing a safer and more convenient treatment option for patients. However, most existing transcutaneous auricular vagus nerve stimulation devices use open-loop control mode, and lack effective real-time monitoring and feedback mechanisms during treatment, making it difficult to adjust parameters dynamically according to changes in the patient's physiological state.

[0003] Under current technical conditions, the setting of treatment parameters mainly relies on the experience of physicians and the subjective feelings of patients, lacking objective quantitative evaluation basis. During treatment, the device cannot real-time perceive the changes in the patient's physiological response, which may lead to insufficient stimulation or over-stimulation, affecting the treatment effect and even causing safety hazards. At the same time, due to significant individual physiological differences, uniform fixed stimulation parameters cannot meet the individualized treatment needs of different patients, resulting in large fluctuations in treatment efficacy. The existing equipment lacks real-time monitoring capability of the device operating state during treatment, and cannot timely discover device problems such as poor electrode contact and abnormal impedance. After treatment, it also cannot provide quantitative efficacy evaluation report, making it difficult for physicians to trace and analyze the treatment process and optimize the treatment plan. These limitations seriously restrict the clinical promotion and application effect of transcutaneous auricular vagus nerve stimulation technology.

[0004] Therefore, the present application provides a treatment monitoring method and system for a transcutaneous auricular vagus nerve stimulation device, which comprehensively monitors the real-time treatment situation of the target user to accurately perceive the treatment effect and ensure treatment safety, and generates individualized diagnosis and treatment assistance plans based on the user's treatment response to help doctors provide targeted treatment plans in the next treatment. SUMMARY

[0005] The present application overcomes the defects of the prior art and provides a treatment monitoring method and system for a transcutaneous auricular vagus nerve stimulation device, which accurately perceives the treatment effect and ensures treatment safety, and helps doctors provide targeted treatment plans in the next treatment.

[0006] To achieve the above-mentioned purpose, the first aspect of the present application provides a treatment monitoring method for a transcutaneous auricular vagus nerve stimulation device, comprising: Within a treatment course, multi-source treatment status monitoring is performed on the target user to obtain multi-source treatment status monitoring data, and deep feature extraction and neural state decoding are performed to construct multimodal treatment status features; Based on the multimodal treatment situation characteristics, real-time treatment situation analysis is performed to analyze the treatment status and efficacy of the target user in the current treatment course and generate real-time treatment situation analysis information. Based on the real-time treatment status analysis information, dynamic status classification is performed to assess the treatment status of the target user in the current treatment course. When there is an abnormal treatment status, a treatment control plan is generated to manage the current treatment course. A treatment status prediction model is constructed to obtain the multimodal treatment status characteristics after regulation. The floating regulation boundary is generated by combining the treatment status prediction model with the floating regulation boundary. The diagnosis and treatment regulation plan is dynamically corrected by the floating regulation boundary to ensure efficacy and treatment safety.

[0007] In this solution, the step of performing multi-source treatment status monitoring on the target user within a treatment course to obtain multi-source treatment status monitoring data, and then performing deep feature extraction and neural state decoding to construct multimodal treatment status features specifically includes: During a treatment course, the treatment status of the percutaneous vagus nerve stimulation device used in the treatment and the treatment user is monitored by a set multi-source sensor array to obtain multi-source treatment status monitoring data, including physiological status data and device operation data. Physiological state data is imported into an adaptive filter, and power frequency interference and baseline drift are dynamically eliminated by the nearest small mean square algorithm. Motion artifact detection algorithm is used to identify and remove signal distortion segments to obtain purified physiological state data. Wavelet transform was performed on the purified physiological state data, and Daubechies wavelet basis was selected for multi-scale decomposition. High-frequency noise was filtered out by soft thresholding while retaining effective physiological signal components to generate preprocessed physiological state data. The device's operating data is compared with a preset operating benchmark to calculate the output error. The electrode impedance is calculated using the voltage difference measured by the constant current source circuit. When the impedance value exceeds the preset safety range, an impedance abnormality flag is generated. Combined with the pre-processed physiological state data, a multi-source treatment status monitoring dataset is constructed.

[0008] In this solution, the step of acquiring multi-source treatment situation monitoring data for the target user, performing deep feature extraction and neural state decoding to construct multimodal treatment situation features, further includes: The multi-source treatment situation monitoring dataset is imported into a pre-trained deep convolutional neural network, and a three-branch architecture is used to construct multimodal treatment situation features. The three-branch architecture consists of a physiological signal branch, a device operation branch, and a cross-modal attention fusion branch. The physiological signal branch extracts physiological state features corresponding to physiological state data through multiple parallel one-dimensional convolution channels; the device operation branch maps device operation data into higher-order representations through fully connected layers to obtain device operation state features. Physiological state features and device operation features are input into the cross-modal attention fusion branch. Based on the multi-head attention mechanism, the device operation features are used as the query, and the physiological state features are used as the key and value. The cross-correlation weights between the physiological state features and the device operation features are dynamically calculated. A temporal attention mechanism is introduced at the network end, a gated recurrent unit is used to capture long-range dependencies, multimodal feature fusion is performed using the calculated cross-association weights, and finally multimodal treatment status features are output through a fully connected layer.

[0009] In this solution, the step of performing real-time treatment status analysis based on the multimodal treatment status characteristics, analyzing the target user's treatment status and efficacy in the current treatment course, and generating real-time treatment status analysis information specifically includes: The multimodal treatment status features are acquired, and the continuous multimodal treatment status features are divided into several feature sequence segments by a sliding time window. Each segment represents a snapshot of the treatment status over a specific time span. The segments are then input into a pre-built dual-channel status analysis network to analyze the treatment status and efficacy of the target user in the current treatment course. The dual-channel situation analysis network includes a treatment status analysis channel and an efficacy analysis channel. The treatment status analysis channel is constructed by a support vector machine, and the width parameter and regularization coefficient of the radial basis kernel function are optimized by grid search cross-validation to obtain the optimal hyperplane parameters. The efficacy analysis channel uses a gradient boosting decision tree regression model as a framework, adopts squared error as the loss function, and iteratively constructs a regression decision tree sequence through a forward step-by-step algorithm. In the treatment status analysis channel, the treatment status snapshot is mapped to a high-dimensional feature space through the radial basis kernel function. The relative distance between the treatment status snapshot and the optimal classification hyperplane is calculated. The relative distance is used to characterize the similarity between the current treatment status snapshot and the abnormal state. The relative distance is converted into a physiological risk probability value using the Sigmoid function. In the efficacy analysis channel, based on the input treatment status snapshot, a forward stepwise algorithm is used to construct multiple regression decision trees as weak learners. Each weak learner fits the negative gradient of the prediction result of the preceding weak learner. The predicted value of the treatment effect within the current time window is obtained by weighted summation of the output results of all weak learners. Finally, real-time treatment status analysis information is generated based on the physiological risk probability value and treatment effect prediction value output by the dual channels.

[0010] In this solution, dynamic status classification is performed based on the real-time treatment status analysis information to assess the target user's treatment status in the current treatment course. When abnormal treatment conditions exist, a treatment control plan is generated to manage and control the current treatment course, specifically including: Acquire real-time treatment status analysis information, import the real-time treatment status analysis information into a state classifier trained based on preset rules to classify the treatment status within the current treatment course, and generate treatment status classification information; The treatment status of the target user in the current treatment course is determined by the treatment status classification information, and different strategies and degrees of treatment control are applied to the target percutaneous ear vagus nerve stimulation device based on different treatment status determination results. Among them, the different treatment states include Class I treatment state, Class II treatment state and Class III treatment state. Class I treatment state is normal physiological state and normal treatment effect. Class II treatment state is normal physiological state and abnormal treatment effect. Class III treatment state is abnormal physiological state and abnormal or normal treatment effect. Several historical control strategies are obtained by retrieving historical data. A control strategy generator is constructed using the whale optimization algorithm and trained using the historical control strategies. When the treatment status classification information is a type II or type III treatment status, the optimal control strategy that matches the current treatment status is obtained through the control strategy generator to control the target percutaneous ear vagus nerve stimulation device, so as to ensure the treatment safety and efficacy of the target user.

[0011] In this scheme, the step of obtaining the optimal control strategy that conforms to the current treatment situation through the control strategy generator to control the target percutaneous vagus nerve stimulation device specifically includes: The treatment status type corresponding to the current treatment course is extracted through the treatment status classification information. If it is a third-class treatment status, the physiological risk probability value of the target user is extracted and compared with the preset risk threshold. If it is greater than the preset risk threshold, a stop treatment command is generated and an early warning is issued. If the risk threshold is less than the preset risk threshold, the multimodal treatment status characteristics are obtained, and the optimal treatment delay control strategy is obtained by using the control strategy generator to reduce the treatment control parameters set by the target percutaneous ear vagus nerve stimulation device. If it is a Class II treatment state, then based on the characteristics of the multimodal treatment state, the optimal treatment course change strategy is obtained through the regulation strategy generator to optimize the diagnostic and treatment control parameters set for the target percutaneous ear vagus nerve stimulation device.

[0012] In this scheme, the step of using a control strategy generator to generate strategies based on the treatment status type of the current treatment course specifically includes: An objective function is generated based on the treatment posture type. The objective function is set to maximize the treatment response index and minimize the physiological risk probability, and is constrained by the comfort index. Several initial control strategies are obtained by similarity retrieval in a preset strategy space in combination with the input multimodal treatment posture features. The preset strategy space is constructed by historical control strategies. Subsequently, each initial regulation strategy is encoded as a whale individual, and the stimulation parameters of the corresponding initial regulation strategy are used as position vectors to form an initial whale population. The regulation parameter features corresponding to each initial strategy are extracted and randomly sampled. Perturbed whale individuals are generated by cross-combination and Gaussian noise addition, and then merged into the initial whale population for individual enrichment to generate a perturbed whale population. The perturbed whale population was clustered using the K-means clustering algorithm. In the first clustering, a whale individual was randomly selected from the perturbed whale population as the cluster center, and K-1 cluster centers were selected according to the maximum distance principle. In subsequent clustering, the cluster centers were recalculated based on the mean value of whale individuals in each current cluster, and each whale individual was assigned to the nearest cluster according to the minimum distance principle. Obtain the clustered whale populations after clustering, solve the fitness problem for each clustered whale population using the set objective function, and select the globally optimal whale individual and the optimal whale individual in each cluster based on the fitness values ​​obtained from the solution. During the position update phase, for each whale individual, the current global best whale individual or the best whale individual corresponding to its own cluster is randomly selected as the leader whale. The position update strategy of the whale individual is determined by random probability and coefficient vector, including the prey encirclement phase, the bubble attack phase, and the prey search phase. After the location update is completed, it is determined whether the preset stopping criterion is met. If the preset stopping criterion is not met, the clustering operation and location update operation are repeated iteratively based on the whale population after the location update until the preset stopping criterion is reached and the final clustered whale population is output. The optimal treatment control strategy is generated by obtaining the final global optimal solution through the final clustering of whale populations, and the treatment control parameters set by the target percutaneous vagus nerve stimulation device are managed.

[0013] In this scheme, the construction of a treatment status prediction model, the acquisition of multimodal treatment status characteristics after regulation, and the generation of a floating regulation boundary based on the treatment status prediction model, along with the dynamic correction of the diagnosis and treatment regulation scheme using the floating regulation boundary to ensure efficacy and treatment safety, specifically includes: A treatment status prediction model was built based on the generative adversarial network framework. Several historical treatment cases were obtained by big data retrieval. The treatment status characteristics of each historical treatment case were extracted and generated into a historical treatment status feature sequence by time-series processing. Based on the historical treatment status feature sequence, the Markov algorithm is used to take the treatment status feature corresponding to each time stamp as a state node, calculate the state transition probability between the state node and the next state node to generate a state transition matrix; the state transition matrix is ​​used as a discriminator constraint, and a training dataset is generated through the historical treatment status feature sequence to train the constructed treatment status prediction model. After the target percutaneous vagus nerve stimulation device is modulated, continuous treatment monitoring is performed to obtain the multimodal treatment status characteristics after modulation. These characteristics are then input into a trained treatment status prediction model to obtain a sequence of predicted physiological risk probabilities and a sequence of predicted treatment effects for future time periods. Based on the physiological risk probability value sequence, the extreme value theory method is used to calculate the distribution characteristics of the maximum points of the physiological risk probability value sequence through the kernel density estimation method, and the high quantile of the maximum point distribution is taken to generate the floating upper limit boundary. Simultaneously, based on the predicted value sequence of treatment effect, the first-order rate of change at each time point is calculated through difference operation, and the sequence is decomposed into trend component, periodic component and residual component using trend decomposition method; the average rate of change is calculated through trend component, and combined with the expected efficacy of the preset treatment plan, a floating lower limit boundary is generated; a floating control boundary is generated based on the floating upper limit boundary and the floating lower limit boundary. In subsequent treatment adjustments, the obtained real-time optimal treatment control strategy is compared with the floating control boundary. The control parameters that exceed the floating control boundary are projected into the boundary range using the least squares projection algorithm to generate a corrected real-time optimal treatment control strategy for treatment management.

[0014] A second aspect of the present invention provides a treatment monitoring system for a percutaneous ear vagus nerve stimulation device. The system includes a memory, a processor, and a communication interface. The memory contains a treatment monitoring method program for the percutaneous ear vagus nerve stimulation device. When executed by the processor, the treatment monitoring method program for the percutaneous ear vagus nerve stimulation device performs the following steps: Within a treatment course, multi-source treatment status monitoring is performed on the target user to obtain multi-source treatment status monitoring data, and deep feature extraction and neural state decoding are performed to construct multimodal treatment status features; Based on the multimodal treatment situation characteristics, real-time treatment situation analysis is performed to analyze the treatment status and efficacy of the target user in the current treatment course and generate real-time treatment situation analysis information. Based on the real-time treatment status analysis information, dynamic status classification is performed to assess the treatment status of the target user in the current treatment course. When there is an abnormal treatment status, a treatment control plan is generated to manage the current treatment course. A treatment status prediction model is constructed to obtain the multimodal treatment status characteristics after regulation. The floating regulation boundary is generated by combining the treatment status prediction model with the floating regulation boundary. The diagnosis and treatment regulation plan is dynamically corrected by the floating regulation boundary to ensure efficacy and treatment safety.

[0015] A third aspect of the present invention provides a computer-readable storage medium comprising a treatment monitoring method program for a percutaneous ear vagus nerve stimulation device, wherein when the treatment monitoring method program for a percutaneous ear vagus nerve stimulation device is executed by a processor, the program implements the steps of the treatment monitoring method for a percutaneous ear vagus nerve stimulation device as described in any of the preceding claims. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples 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 according to these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a first method for monitoring the treatment of a percutaneous vagus nerve stimulation device according to an embodiment of the present invention; Figure 2 This is a flowchart of a second method for monitoring the treatment of a percutaneous vagus nerve stimulation device according to an embodiment of the present invention; Figure 3 This is a block diagram of a treatment monitoring system for a percutaneous vagus nerve stimulation device provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0020] Figure 1 This is a flowchart of a first method for monitoring the treatment of a percutaneous vagus nerve stimulation device according to an embodiment of the present invention; likeFigure 1 As shown, the present invention provides a first method flowchart for a treatment monitoring method for a percutaneous vagus nerve stimulation device, comprising: S102, within a treatment course, multi-source treatment status monitoring is performed on the target user to obtain multi-source treatment status monitoring data, and deep feature extraction and neural state decoding are performed to construct multimodal treatment status features; S104, perform real-time treatment status analysis based on the multimodal treatment status characteristics, analyze the treatment status and efficacy of the target user in the current treatment course, and generate real-time treatment status analysis information; S106, Based on the real-time treatment status analysis information, perform dynamic status classification, assess the treatment status of the target user in the current treatment course, and when there is an abnormal treatment status, generate a treatment control plan to manage the current treatment course. S108, Construct a treatment status prediction model, obtain the multimodal treatment status characteristics after regulation, and generate a floating regulation boundary in combination with the treatment status prediction model. Use the floating regulation boundary to dynamically correct the diagnosis and treatment regulation plan to ensure efficacy and treatment safety.

[0021] Furthermore, in a preferred embodiment of the present invention, the step of performing multi-source treatment status monitoring on the target user within a treatment course to obtain multi-source treatment status monitoring data, and performing deep feature extraction and neural state decoding to construct multimodal treatment status features specifically includes: During a treatment course, the treatment status of the percutaneous vagus nerve stimulation device used in the treatment and the treatment user is monitored by a set multi-source sensor array to obtain multi-source treatment status monitoring data, including physiological status data and device operation data. Physiological state data is imported into an adaptive filter, and power frequency interference and baseline drift are dynamically eliminated by the nearest small mean square algorithm. Motion artifact detection algorithm is used to identify and remove signal distortion segments to obtain purified physiological state data. Wavelet transform was performed on the purified physiological state data, and Daubechies wavelet basis was selected for multi-scale decomposition. High-frequency noise was filtered out by soft thresholding while retaining effective physiological signal components to generate preprocessed physiological state data. The device's operating data is compared with a preset operating benchmark to calculate the output error. The electrode impedance is calculated using the voltage difference measured by the constant current source circuit. When the impedance value exceeds the preset safety range, an impedance abnormality flag is generated. Combined with the pre-processed physiological state data, a multi-source treatment status monitoring dataset is constructed.

[0022] It should be noted that during a treatment course, a pre-deployed multi-source sensor array comprehensively monitors the percutaneous vagus nerve stimulation device and its user, simultaneously collecting two types of data: physiological status and device operation data. Physiological status data originates from bioelectric sensors integrated into the device, continuously capturing the user's electrocardiogram (ECG), electroencephalogram (EEG), and electrodermal signal. Device operation data records real-time parameters such as stimulation current output, electrode voltage, and pulse frequency, forming an initial multi-source treatment status monitoring data stream. Subsequently, the collected physiological status data is imported into an adaptive filter for processing. A normalized least mean square algorithm is used as the core processing engine, dynamically tracking and eliminating 50Hz power frequency interference and low-frequency baseline drift by updating filter coefficients in real time. Simultaneously, motion signals from a triaxial accelerometer are combined, and a motion artifact detection algorithm based on morphological processing is used to identify and remove signal distortion segments caused by user movement, ultimately outputting purified physiological status data with a significantly improved signal-to-noise ratio. Next, wavelet transform processing is applied to the purified physiological status data. A five-level multi-scale decomposition was performed using the Daubechies wavelet basis function to decompose the signal into different frequency subbands. At each scale, a soft thresholding method based on Stein's unbiased risk estimation was used to quantify the detail coefficients, effectively filtering high-frequency noise while retaining the effective signal components reflecting the neurophysiological state. This resulted in the reconstruction of high-quality preprocessed physiological state data. Simultaneously, the device operation data was processed in parallel: real-time acquired stimulation parameters were compared with preset operating baseline values ​​to calculate the output error index; the voltage difference across the electrodes was obtained using a constant current source measurement circuit, and the electrode-skin contact impedance was calculated in real-time according to Ohm's law. When the impedance value exceeded the preset safety threshold range, an impedance anomaly flag was immediately generated. Finally, the preprocessed physiological state data and the device operation data with quality flags were aligned and fused in the time dimension to form a multi-source treatment status monitoring dataset, providing a data foundation for subsequent deep feature extraction and intelligent analysis.

[0023] Furthermore, in a preferred embodiment of the present invention, the step of acquiring multi-source treatment situation monitoring data by performing multi-source treatment situation monitoring on the target user, and constructing multimodal treatment situation features by performing deep feature extraction and neural state decoding, further includes: The multi-source treatment situation monitoring dataset is imported into a pre-trained deep convolutional neural network, and a three-branch architecture is used to construct multimodal treatment situation features. The three-branch architecture consists of a physiological signal branch, a device operation branch, and a cross-modal attention fusion branch. The physiological signal branch extracts physiological state features corresponding to physiological state data through multiple parallel one-dimensional convolution channels; the device operation branch maps device operation data into higher-order representations through fully connected layers to obtain device operation state features. Physiological state features and device operation features are input into the cross-modal attention fusion branch. Based on the multi-head attention mechanism, the device operation features are used as the query, and the physiological state features are used as the key and value. The cross-correlation weights between the physiological state features and the device operation features are dynamically calculated. A temporal attention mechanism is introduced at the network end, a gated recurrent unit is used to capture long-range dependencies, multimodal feature fusion is performed using the calculated cross-association weights, and finally multimodal treatment status features are output through a fully connected layer.

[0024] It should be noted that the preprocessed multi-source treatment status monitoring dataset is imported into a pre-trained deep convolutional neural network for advanced feature extraction and fusion. This network employs a three-branch architecture, processing physiological signals, device operation data, and their cross-modal fusion. The physiological signal branch contains multiple parallel one-dimensional convolutional channels, each using convolutional kernels of different sizes to process ECG, EEG, and EMG signals respectively. Through convolutional operations, it captures the local temporal patterns and spectral features of various physiological signals, outputting a physiological state feature vector. The device operation branch processes device operation data such as stimulation parameters and impedance values ​​through fully connected layers, mapping this structured data to a high-dimensional feature space to form a high-order feature vector representing the device's operating state. The output features from both branches are then input into the cross-modal attention fusion branch, employing a multi-head attention mechanism. The device operation state features serve as the query vector, while the physiological state features act as both the key and value vectors. By calculating the similarity between the query vector and the key vector, cross-association weights are generated, reflecting the strength of the association between device operation parameters and physiological responses. A temporal attention mechanism is introduced at the network endpoint, using a gated recurrent unit (GRU) to further process the serialized fused features. The GRU selectively remembers and forgets historical information through update and reset gates, capturing long-term dependencies in feature changes during treatment. It then combines calculated cross-correlation weights for multimodal feature fusion, and performs dimensionality transformation and feature compression through fully connected layers, ultimately outputting a multimodal treatment posture feature vector. This preserves the discriminative information of the original signal and encodes the dynamic correlation patterns between physiological states and device operating states.

[0025] Furthermore, in a preferred embodiment of the present invention, the step of performing real-time treatment situation analysis based on the multimodal treatment situation characteristics, analyzing the treatment status and efficacy of the target user in the current treatment course, and generating real-time treatment situation analysis information specifically includes: The multimodal treatment status features are acquired, and the continuous multimodal treatment status features are divided into several feature sequence segments by a sliding time window. Each segment represents a snapshot of the treatment status over a specific time span. The segments are then input into a pre-built dual-channel status analysis network to analyze the treatment status and efficacy of the target user in the current treatment course. The dual-channel situation analysis network includes a treatment status analysis channel and an efficacy analysis channel. The treatment status analysis channel is constructed by a support vector machine, and the width parameter and regularization coefficient of the radial basis kernel function are optimized by grid search cross-validation to obtain the optimal hyperplane parameters. The efficacy analysis channel uses a gradient boosting decision tree regression model as a framework, adopts squared error as the loss function, and iteratively constructs a regression decision tree sequence through a forward step-by-step algorithm. In the treatment status analysis channel, the treatment status snapshot is mapped to a high-dimensional feature space through the radial basis kernel function. The relative distance between the treatment status snapshot and the optimal classification hyperplane is calculated. The relative distance is used to characterize the similarity between the current treatment status snapshot and the abnormal state. The relative distance is converted into a physiological risk probability value using the Sigmoid function. In the efficacy analysis channel, based on the input treatment status snapshot, a forward stepwise algorithm is used to construct multiple regression decision trees as weak learners. Each weak learner fits the negative gradient of the prediction result of the preceding weak learner. The predicted value of the treatment effect within the current time window is obtained by weighted summation of the output results of all weak learners. Finally, real-time treatment status analysis information is generated based on the physiological risk probability value and treatment effect prediction value output by the dual channels.

[0026] It should be noted that the acquisition of multimodal treatment status features involves dividing the continuous feature stream into a series of feature sequence segments through a fixed-duration sliding time window. Each segment constitutes a snapshot of the treatment status, representing the overall treatment situation within a specific time period. These snapshots are then fed in parallel into a pre-constructed dual-channel status analysis network for real-time analysis. This network includes a treatment status analysis channel and an efficacy analysis channel. The treatment status analysis channel is constructed using a support vector machine, and the width parameter and regularization coefficient of the radial basis function kernel are optimized through grid search cross-validation to obtain the optimal hyperplane parameters. The efficacy analysis channel uses a gradient boosting decision tree regression model as its framework. It progressively constructs a sequence of regression decision trees through a forward stepwise algorithm. Each tree uses the residual of the prediction results from the previous model as its learning objective, and continuously improves the prediction accuracy by minimizing the squared error loss function. In the treatment status analysis channel, each treatment status snapshot is nonlinearly mapped to a high-dimensional feature space through the radial basis function kernel. In this space, the functional distance between the snapshot and the optimal classification hyperplane is calculated, quantifying the geometric relationship between the current state and the boundary of the abnormal state, and characterizing the degree of abnormality in the treatment status. The distance value is then converted into a physiological risk probability value within the range of 0 to 1 using the Sigmoid function. In the efficacy analysis channel, multiple regression decision trees are constructed as weak learners based on a forward stepwise algorithm. Each weak learner fits the negative gradient direction of the prediction result of the preceding model using a gradient descent algorithm. Finally, the prediction outputs of all weak learners are weighted and summed to obtain the continuous predicted value of the treatment effect within the current time window, representing the development trend of the treatment effect. Finally, the physiological risk probability values ​​and treatment effect prediction values ​​output in parallel from both channels are time-aligned and fused to generate real-time treatment status analysis information that includes safety status assessment and efficacy trend prediction.

[0027] Furthermore, in a preferred embodiment of the present invention, the step of constructing a treatment status prediction model, obtaining the multimodal treatment status characteristics after regulation, and generating a floating regulation boundary in combination with the treatment status prediction model, and dynamically correcting the diagnosis and treatment regulation scheme using the floating regulation boundary to ensure efficacy and treatment safety, specifically includes: A treatment status prediction model was built based on the generative adversarial network framework. Several historical treatment cases were obtained by big data retrieval. The treatment status characteristics of each historical treatment case were extracted and generated into a historical treatment status feature sequence by time-series processing. Based on the historical treatment status feature sequence, the Markov algorithm is used to take the treatment status feature corresponding to each time stamp as a state node, calculate the state transition probability between the state node and the next state node to generate a state transition matrix; the state transition matrix is ​​used as a discriminator constraint, and a training dataset is generated through the historical treatment status feature sequence to train the constructed treatment status prediction model. After the target percutaneous vagus nerve stimulation device is modulated, continuous treatment monitoring is performed to obtain the multimodal treatment status characteristics after modulation. These characteristics are then input into a trained treatment status prediction model to obtain a sequence of predicted physiological risk probabilities and a sequence of predicted treatment effects for future time periods. Based on the physiological risk probability value sequence, the extreme value theory method is used to calculate the distribution characteristics of the maximum points of the physiological risk probability value sequence through the kernel density estimation method, and the high quantile of the maximum point distribution is taken to generate the floating upper limit boundary. Simultaneously, based on the predicted value sequence of treatment effect, the first-order rate of change at each time point is calculated through difference operation, and the sequence is decomposed into trend component, periodic component and residual component using trend decomposition method; the average rate of change is calculated through trend component, and combined with the expected efficacy of the preset treatment plan, a floating lower limit boundary is generated; a floating control boundary is generated based on the floating upper limit boundary and the floating lower limit boundary. In subsequent treatment adjustments, the obtained real-time optimal treatment control strategy is compared with the floating control boundary. The control parameters that exceed the floating control boundary are projected into the boundary range using the least squares projection algorithm to generate a corrected real-time optimal treatment control strategy for treatment management.

[0028] It should be noted that a treatment posture prediction model is constructed based on a generative adversarial network (GAN) framework. A large amount of historical treatment case data is retrieved from a big data platform, from which multimodal treatment posture features of each case are extracted and arranged chronologically to form a historical treatment posture feature sequence library. A Markov algorithm is employed, treating each timestamp's treatment posture feature as a state node. The transition frequencies between state nodes are calculated and normalized to generate a state transition probability matrix representing the evolution of the treatment process. This state transition matrix serves as a priori knowledge constraint for the discriminator in the GAN, forming a training dataset together with the historical treatment posture feature sequence. Through adversarial training, the predictive ability of the generator and the judgment accuracy of the discriminator are continuously optimized, enabling the prediction model to learn the inherent laws of treatment posture development. In the real-time application phase, after the target is regulated by a percutaneous vagus nerve stimulation device, the multimodal treatment posture features after regulation are continuously monitored and input into the fully trained treatment posture prediction model. The model, through forward propagation calculation of the generator network, outputs a sequence of predicted physiological risk probabilities and treatment effects at multiple future time points, forming a multi-step prediction of the treatment development trend. Based on the physiological risk probability value sequence, extreme value theory is used to identify the maximum points in the sequence. The probability distribution characteristics of the maximum points are fitted using kernel density estimation, and the high quantile of this distribution is used as a reference threshold. Combined with the patient's individual safety tolerance range, a dynamic floating upper limit boundary is generated. Simultaneously, the instantaneous rate of change at each time point is calculated using first-order difference operations on the treatment effect prediction value sequence. A time series decomposition method is used to decompose the original sequence into three parts: trend component, periodic component, and residual component. The average rate of change is calculated based on the trend component, and combined with the expected efficacy target of the preset treatment plan, a dynamic floating lower limit boundary is generated. The floating upper and lower limits are combined to form a multidimensional floating control boundary, defining the safe operating range of the treatment parameters. In subsequent treatment control decisions, the obtained real-time optimal treatment control strategy is compared with the multidimensional floating control boundary. A least-squares projection algorithm projects the control parameter vectors exceeding the boundary onto the nearest boundary surface, ensuring that all parameters fall within the safe range while maintaining the original optimization direction. This generates a corrected real-time optimal treatment control strategy that meets both the efficacy optimization target and ensures treatment safety, ultimately achieving precise and safe control of the treatment process.

[0029] Figure 2 This is a flowchart of a second method for monitoring the treatment of a percutaneous vagus nerve stimulation device according to an embodiment of the present invention; like Figure 2 As shown, the present invention provides a second method flowchart for a treatment monitoring method for a percutaneous vagus nerve stimulation device, comprising: S202, Obtain real-time treatment status analysis information, import the real-time treatment status analysis information into a state classifier trained based on preset rules to classify the treatment status within the current treatment course, and generate treatment status classification information; S204, the treatment status of the target user in the current treatment course is determined by the treatment status classification information, and the treatment control of the target percutaneous ear vagus nerve stimulation device is carried out with different strategies based on different treatment status determination results. S206, wherein different treatment states include Class I treatment state, Class II treatment state and Class III treatment state, wherein Class I treatment state is normal physiological state and normal treatment effect, Class II treatment state is normal physiological state and abnormal treatment effect, and Class III treatment state is abnormal physiological state and abnormal or normal treatment effect; S208: Several historical control strategies are obtained through historical data retrieval. A control strategy generator is constructed using the whale optimization algorithm and trained using the historical control strategies. When the treatment status classification information is a second-class or third-class treatment status, the optimal control strategy that conforms to the current treatment status is obtained through the control strategy generator to control the target percutaneous ear vagus nerve stimulation device, so as to ensure the treatment safety and efficacy of the target user.

[0030] It should be noted that real-time treatment status analysis information is imported into a machine learning-based state classifier for treatment status assessment. This state classifier uses a classification model trained on a large amount of clinical data and can be built using support vector machines or random forests. It intelligently identifies and classifies the current treatment status according to preset classification rules. The classifier output includes three types of treatment status judgment results: Type I treatment status indicates that physiological indicators are within a safe range and the treatment effect meets the expected standard, belonging to an ideal treatment status; Type II treatment status indicates that physiological indicators are normal but the treatment effect does not meet expectations, suggesting the need to optimize the treatment plan; Type III treatment status indicates that abnormal physiological indicators have occurred, requiring immediate intervention regardless of the treatment effect. Differentiated control strategies are activated based on different treatment status classification results. When a Type I treatment status is determined, the current treatment parameters are maintained for continued observation; when a Type II or Type III treatment status occurs, the control strategy generation mechanism is automatically triggered.

[0031] To generate precise control strategies, a large number of historical control cases were retrieved through historical data retrieval, and successful control strategies were extracted to form a training sample library. A control strategy generator was constructed using the whale optimization algorithm and trained using the training sample library.

[0032] When a Class II or Class III treatment state is detected, the control strategy generator is immediately activated, inputting the current multimodal treatment state characteristics. Through the intelligent search mechanism of the whale optimization algorithm, the most suitable control strategy for the current situation is quickly found in the strategy space. The generated optimal control strategy is implemented through the device control interface, achieving precise control of the percutaneous vagus nerve stimulation device and ensuring optimal treatment efficacy and safety even in complex treatment environments.

[0033] Furthermore, in a preferred embodiment of the present invention, the step of controlling the target percutaneous vagus nerve stimulation device by obtaining the optimal control strategy that conforms to the current treatment situation through the control strategy generator specifically includes: The treatment status type corresponding to the current treatment course is extracted through the treatment status classification information. If it is a third-class treatment status, the physiological risk probability value of the target user is extracted and compared with the preset risk threshold. If it is greater than the preset risk threshold, a stop treatment command is generated and an early warning is issued. If the risk threshold is less than the preset risk threshold, the multimodal treatment status characteristics are obtained, and the optimal treatment delay control strategy is obtained by using the control strategy generator to reduce the treatment control parameters set by the target percutaneous ear vagus nerve stimulation device. If it is a Class II treatment state, then based on the characteristics of the multimodal treatment state, the optimal treatment course change strategy is obtained through the regulation strategy generator to optimize the diagnostic and treatment control parameters set for the target percutaneous ear vagus nerve stimulation device.

[0034] It should be noted that the specific treatment status type corresponding to the current treatment course is analyzed based on the treatment status classification information. When a Class III treatment status is identified, a high-risk response mechanism is immediately activated: First, the real-time physiological risk probability value of the target user is extracted and compared with a preset multi-level safety threshold. If the value exceeds the high-risk threshold, a treatment stop command is immediately generated and a multi-level early warning system is simultaneously triggered, sending an emergency alarm to medical personnel through audio-visual prompts and remote notifications to ensure patient safety. If the risk probability value is less than the high-risk threshold but greater than the safety threshold, the current multimodal treatment status feature vector is automatically obtained, and a regulation strategy generator trained by reinforcement learning is invoked to generate an optimal treatment course delay regulation strategy aimed at risk mitigation. This strategy significantly reduces potential risks while ensuring basic treatment benefits by implementing stepwise downgrade adjustments to key parameters such as stimulation intensity and frequency. When a Class II treatment status is determined, it indicates that the treatment safety is controllable but the efficacy has not met expectations. At this time, an efficacy optimization mechanism is activated: Based on the real-time multimodal treatment status features, a regulation strategy generator trained by deep reinforcement learning is used to generate an optimal treatment course change strategy with efficacy improvement as the core objective. This strategy precisely adjusts deep parameters such as stimulation waveform and timing pattern by analyzing the matching degree between historical therapeutic response data and current characteristic patterns, while introducing safety constraints to ensure that the adjustment process is always within the physiological safety range.

[0035] Furthermore, in a preferred embodiment of the present invention, the step of generating a strategy based on the treatment status type of the current treatment course using a control strategy generator specifically includes: An objective function is generated based on the treatment posture type. The objective function is set to maximize the treatment response index and minimize the physiological risk probability, and is constrained by the comfort index. Several initial control strategies are obtained by similarity retrieval in a preset strategy space in combination with the input multimodal treatment posture features. The preset strategy space is constructed by historical control strategies. Subsequently, each initial regulation strategy is encoded as a whale individual, and the stimulation parameters of the corresponding initial regulation strategy are used as position vectors to form an initial whale population. The regulation parameter features corresponding to each initial strategy are extracted and randomly sampled. Perturbed whale individuals are generated by cross-combination and Gaussian noise addition, and then merged into the initial whale population for individual enrichment to generate a perturbed whale population. The perturbed whale population was clustered using the K-means clustering algorithm. In the first clustering, a whale individual was randomly selected from the perturbed whale population as the cluster center, and K-1 cluster centers were selected according to the maximum distance principle. In subsequent clustering, the cluster centers were recalculated based on the mean value of whale individuals in each current cluster, and each whale individual was assigned to the nearest cluster according to the minimum distance principle. Obtain the clustered whale populations after clustering, solve the fitness problem for each clustered whale population using the set objective function, and select the globally optimal whale individual and the optimal whale individual in each cluster based on the fitness values ​​obtained from the solution. During the position update phase, for each whale individual, the current global best whale individual or the best whale individual corresponding to its own cluster is randomly selected as the leader whale. The position update strategy of the whale individual is determined by random probability and coefficient vector, including the prey encirclement phase, the bubble attack phase, and the prey search phase. After the location update is completed, it is determined whether the preset stopping criterion is met. If the preset stopping criterion is not met, the clustering operation and location update operation are repeated iteratively based on the whale population after the location update until the preset stopping criterion is reached and the final clustered whale population is output. The optimal treatment control strategy is generated by obtaining the final global optimal solution through the final clustering of whale populations, and the treatment control parameters set by the target percutaneous vagus nerve stimulation device are managed.

[0036] It should be noted that this invention provides a method for generating control strategies based on the improved Whale Optimization Algorithm (WOA), the core of which lies in generating optimal treatment control strategies for percutaneous vagus nerve stimulation devices. A target function is constructed based on the treatment posture type, aiming to maximize the treatment response index and minimize the physiological risk probability, with comfort indicators as constraints. Subsequently, similarity searches are performed in a preset strategy space constructed from historical control strategies using multimodal treatment posture features. Several initial control strategies that best match the current posture are selected by calculating Euclidean distance and cosine similarity. Then, each initial control strategy is represented as a whale individual using real-number encoding, and parameters such as stimulation intensity, frequency, and pulse width are used to construct the individual's position vector, thus forming an initial whale population. To enhance population diversity, control parameter features are extracted from the initial strategies and randomly sampled. New perturbed whale individuals are generated through arithmetic crossover operations and the addition of random perturbations conforming to a normal distribution. These perturbed whale individuals are then merged with the initial population to form a perturbed whale population with enhanced diversity. The K-means clustering algorithm is used to intelligently group the perturbed population. In the initial clustering, the minimum-maximum distance principle is used to initialize cluster centers, ensuring sufficient discriminative power between categories. In subsequent iterations, cluster centers are dynamically updated based on the mean of individuals within each cluster, and the nearest neighbor principle is used to further divide individuals. After clustering, a preset objective function is used to calculate the fitness value of each whale individual, thereby selecting the globally optimal individual and the locally optimal individuals within each cluster. During the position update phase, each whale individual chooses to follow either the globally optimal individual or the optimal individual in its own cluster based on a random probability p (p=0.5) and a coefficient vector A, and executes different update strategies accordingly: a linear convergence approach is used to move towards the target position during the prey encirclement phase; a spiral approach is used to approach the optimal solution during the bubble attack phase; and an exploratory search is conducted using a random vector during the prey search phase. Constraint checks are performed after each position update to ensure that the new position satisfies all constraints. The population convergence is continuously monitored during the iteration process, and the final clustered whale population is output when the stopping criterion (such as the maximum number of iterations) is reached. The optimal control strategy is obtained by extracting the individual with the highest fitness from the population and then converting it into specific stimulation parameter adjustment instructions. This enables precise parameter control of the percutaneous vagus nerve stimulation device, thereby maximizing the therapeutic effect while ensuring treatment safety.

[0037] The coefficient vector is calculated as follows: ; in, For random probability, It is a random number between [0, 1].

[0038] in, The iteration count decreases linearly as the number of iterations increases. The specific calculation method is as follows: ; The This represents the maximum number of iterations.

[0039] Furthermore, the treatment monitoring method for a percutaneous vagus nerve stimulation device provided by the present invention further includes the following steps: After completing the current treatment course, extract multi-source treatment status monitoring data, real-time treatment status analysis information, treatment control plan execution records, and final treatment result evaluation data of the target user throughout the entire treatment process from the data storage unit to construct a full-process treatment monitoring dataset. The treatment process monitoring dataset is time-aligned and missing values ​​are processed. Then, three-dimensional key feature indicators are extracted, including time dimension features, spatial dimension features and effect dimension features, to generate a treatment process feature matrix. Based on the feature matrix of the treatment process, the t-SNE dimensionality reduction algorithm is used to map the high-dimensional treatment process features to a three-dimensional space. The key state nodes and state transition patterns in the treatment process are identified by the density clustering algorithm, and a treatment efficacy graph is constructed with state nodes as vertices and state transition probabilities as edges. In the therapeutic efficacy map, the node size represents the duration of stay in that state, the edge thickness represents the frequency of transitions between states, and the color intensity represents the average therapeutic efficacy level of each state. The treatment efficacy map is used to analyze the treatment adaptation path of the target user. The dynamic time warping algorithm is used to calculate the similarity between the current user's treatment path and the ideal treatment path. The density peak clustering method is used to identify the user's specific treatment adaptation pattern and generate a treatment state transition sequence. The PrefixSpan algorithm is used to analyze the frequent treatment stage transition patterns from the treatment state transition sequence. A depth-first search strategy is used to mine all frequent subsequences that meet the minimum support threshold. The frequency of frequent subsequences and their occurrence are statistically analyzed to determine the user's state transition preferences and stability between different treatment stages. The optimal treatment adaptation path for the target user is obtained and personalized diagnosis and treatment auxiliary reports are generated and pushed in combination with the treatment efficacy map.

[0040] It should be noted that due to individual differences, a treatment plan based on the same standard may not be suitable for the target user. Furthermore, because the target user's response varies during actual treatment, the corresponding preset control parameters may not meet treatment expectations and safety. Therefore, after completing the current treatment course, multi-source heterogeneous data generated by the target user throughout the entire treatment process is extracted from the data storage unit. This includes multimodal treatment status monitoring data, real-time treatment status analysis information, treatment control plan execution records, and final treatment result evaluation data, constructing a complete treatment process monitoring dataset. This dataset undergoes time-series alignment processing, employing a dynamic time warping algorithm to eliminate time-series biases caused by different sampling rates. Data quality is improved through a K-nearest neighbor-based missing value imputation method. Subsequently, key feature indicators are extracted from the time, spatial, and effect dimensions, including time-series features, spatial distribution features, and efficacy evolution features, forming a high-dimensional treatment process feature matrix. Based on the treatment process feature matrix, the t-SNE nonlinear dimensionality reduction algorithm is used to map the high-dimensional features to a three-dimensional visualization space. By leveraging the similarity relationships between samples in the high-dimensional and low-dimensional spaces, the topological structure features of the treatment process are effectively preserved. Subsequently, the density-based DBSCAN clustering algorithm identifies key state nodes in the treatment process, including different treatment stages such as the steady-state treatment period, parameter adjustment period, and abnormal response period. The transition probabilities between state nodes are calculated, constructing a treatment efficacy graph with state nodes as vertices and state transition probabilities as edges. In the graph visualization, node size uses a logarithmic scale to represent the relative duration of stay in that state, while edge thickness uses a linear scale to represent the frequency of transitions between states. A heatmap color mapping is used to represent the average treatment efficacy level of each state, forming an intuitive three-dimensional visualization graph.

[0041] Furthermore, the treatment efficacy graph analysis is used to analyze the treatment adaptation path of the target user. A dynamic time warping algorithm is employed to calculate the similarity distance between the current user's treatment state sequence and the standard ideal treatment path. By constructing a cumulative cost matrix, the optimal nonlinear alignment path between the two sequences is found. Density peak clustering is used to identify the user's specific treatment adaptation patterns, including typical patterns such as rapid response, slow adaptation, and fluctuating response, and corresponding treatment state transition sequences are generated. Next, the PrefixSpan sequence pattern mining algorithm is used to analyze frequently occurring treatment stage transition patterns from the state transition sequences. A depth-first search strategy is used to mine all frequent subsequences that meet the minimum support threshold. By statistically analyzing the frequent subsequences and their frequency, the user's state transition preferences (represented by physiological state response) and their stability (represented by efficacy feedback) between different treatment stages are determined, thereby deriving the user's optimal treatment adaptation path. Finally, combined with the treatment efficacy graph analysis results, a personalized diagnostic and treatment assistance report is generated and pushed to medical professionals and users to help medical personnel better formulate corresponding treatment plans in the next treatment process.

[0042] Figure 3 A treatment monitoring system 3 for a percutaneous vagus nerve stimulation device is provided in one embodiment of the present invention. The system includes: a memory 301, a processor 302, and a communication interface 303. The memory 301 contains a treatment monitoring method program for the percutaneous vagus nerve stimulation device. When the processor 302 executes the treatment monitoring method program for the percutaneous vagus nerve stimulation device, it performs the following steps: Within a treatment course, multi-source treatment status monitoring is performed on the target user to obtain multi-source treatment status monitoring data, and deep feature extraction and neural state decoding are performed to construct multimodal treatment status features; Based on the multimodal treatment situation characteristics, real-time treatment situation analysis is performed to analyze the treatment status and efficacy of the target user in the current treatment course and generate real-time treatment situation analysis information. Based on the real-time treatment status analysis information, dynamic status classification is performed to assess the treatment status of the target user in the current treatment course. When there is an abnormal treatment status, a treatment control plan is generated to manage the current treatment course. A treatment status prediction model is constructed to obtain the multimodal treatment status characteristics after regulation. The floating regulation boundary is generated by combining the treatment status prediction model with the floating regulation boundary. The diagnosis and treatment regulation plan is dynamically corrected by the floating regulation boundary to ensure efficacy and treatment safety.

[0043] A third aspect of the present invention provides a computer-readable storage medium comprising a treatment monitoring method program for a percutaneous ear vagus nerve stimulation device, wherein when the treatment monitoring method program for a percutaneous ear vagus nerve stimulation device is executed by a processor, the program implements the steps of the treatment monitoring method for a percutaneous ear vagus nerve stimulation device as described in any of the preceding claims.

[0044] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0045] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0046] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0047] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0048] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, 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 methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0049] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring the treatment of a percutaneous vagus nerve stimulation device, characterized in that, include: Within a treatment course, multi-source treatment status monitoring is performed on the target user to obtain multi-source treatment status monitoring data, and deep feature extraction and neural state decoding are performed to construct multimodal treatment status features; Based on the multimodal treatment situation characteristics, real-time treatment situation analysis is performed to analyze the treatment status and efficacy of the target user in the current treatment course and generate real-time treatment situation analysis information. Based on the real-time treatment status analysis information, dynamic status classification is performed to assess the treatment status of the target user in the current treatment course. When there is an abnormal treatment status, a treatment control plan is generated to manage the current treatment course. A treatment status prediction model is constructed to obtain the multimodal treatment status characteristics after regulation. The floating regulation boundary is generated by combining the treatment status prediction model with the floating regulation boundary. The diagnosis and treatment regulation plan is dynamically corrected by the floating regulation boundary to ensure efficacy and treatment safety.

2. The treatment monitoring method for a percutaneous vagus nerve stimulation device according to claim 1, characterized in that, Within a treatment course, multi-source treatment status monitoring is performed on the target user to obtain multi-source treatment status monitoring data. Deep feature extraction and neural state decoding are then performed to construct multimodal treatment status features. Specifically, this includes: During a treatment course, the treatment status of the percutaneous vagus nerve stimulation device used in the treatment and the treatment user is monitored by a set multi-source sensor array to obtain multi-source treatment status monitoring data, including physiological status data and device operation data. Physiological state data is imported into an adaptive filter, and power frequency interference and baseline drift are dynamically eliminated by the nearest small mean square algorithm. Motion artifact detection algorithm is used to identify and remove signal distortion segments to obtain purified physiological state data. Wavelet transform was performed on the purified physiological state data, and Daubechies wavelet basis was selected for multi-scale decomposition. High-frequency noise was filtered out by soft thresholding while retaining effective physiological signal components to generate preprocessed physiological state data. The device's operating data is compared with a preset operating benchmark to calculate the output error. The electrode impedance is calculated using the voltage difference measured by the constant current source circuit. When the impedance value exceeds the preset safety range, an impedance abnormality flag is generated. Combined with the pre-processed physiological state data, a multi-source treatment status monitoring dataset is constructed.

3. The treatment monitoring method for a percutaneous vagus nerve stimulation device according to claim 1, characterized in that, The step of acquiring multi-source treatment situation monitoring data for the target user, performing deep feature extraction and neural state decoding to construct multimodal treatment situation features, also includes: The multi-source treatment situation monitoring dataset is imported into a pre-trained deep convolutional neural network, and a three-branch architecture is used to construct multimodal treatment situation features. The three-branch architecture consists of a physiological signal branch, a device operation branch, and a cross-modal attention fusion branch. The physiological signal branch extracts physiological state features corresponding to physiological state data through multiple parallel one-dimensional convolution channels; the device operation branch maps device operation data into higher-order representations through fully connected layers to obtain device operation state features. Physiological state features and device operation features are input into the cross-modal attention fusion branch. Based on the multi-head attention mechanism, the device operation features are used as the query, and the physiological state features are used as the key and value. The cross-correlation weights between the physiological state features and the device operation features are dynamically calculated. A temporal attention mechanism is introduced at the network end, a gated recurrent unit is used to capture long-range dependencies, multimodal feature fusion is performed using the calculated cross-association weights, and finally multimodal treatment status features are output through a fully connected layer.

4. The treatment monitoring method for a percutaneous vagus nerve stimulation device according to claim 1, characterized in that, The step of performing real-time treatment situation analysis based on the multimodal treatment situation characteristics, analyzing the target user's treatment status and efficacy in the current treatment course, and generating real-time treatment situation analysis information specifically includes: The multimodal treatment status features are acquired, and the continuous multimodal treatment status features are divided into several feature sequence segments by a sliding time window. Each segment represents a snapshot of the treatment status over a specific time span. The segments are then input into a pre-built dual-channel status analysis network to analyze the treatment status and efficacy of the target user in the current treatment course. The dual-channel situation analysis network includes a treatment status analysis channel and an efficacy analysis channel. The treatment status analysis channel is constructed by a support vector machine, and the width parameter and regularization coefficient of the radial basis kernel function are optimized by grid search cross-validation to obtain the optimal hyperplane parameters. The efficacy analysis channel uses a gradient boosting decision tree regression model as a framework, adopts squared error as the loss function, and iteratively constructs a regression decision tree sequence through a forward step-by-step algorithm. In the treatment status analysis channel, the treatment status snapshot is mapped to a high-dimensional feature space through the radial basis kernel function. The relative distance between the treatment status snapshot and the optimal classification hyperplane is calculated. The relative distance is used to characterize the similarity between the current treatment status snapshot and the abnormal state. The relative distance is converted into a physiological risk probability value using the Sigmoid function. In the efficacy analysis channel, based on the input treatment status snapshot, a forward stepwise algorithm is used to construct multiple regression decision trees as weak learners. Each weak learner fits the negative gradient of the prediction result of the preceding weak learner. The predicted value of the treatment effect within the current time window is obtained by weighted summation of the output results of all weak learners. Finally, real-time treatment status analysis information is generated based on the physiological risk probability value and treatment effect prediction value output by the dual channels.

5. A treatment monitoring method for a percutaneous vagus nerve stimulation device according to claim 1, characterized in that, The process involves dynamic status classification based on the real-time treatment status analysis information to assess the target user's treatment status in the current treatment course. When abnormal treatment conditions exist, a treatment control plan is generated to manage and control the current treatment course. Specifically, this includes: Acquire real-time treatment status analysis information, import the real-time treatment status analysis information into a state classifier trained based on preset rules to classify the treatment status within the current treatment course, and generate treatment status classification information; The treatment status of the target user in the current treatment course is determined by the treatment status classification information, and different strategies and degrees of treatment control are applied to the target percutaneous ear vagus nerve stimulation device based on different treatment status determination results. Among them, the different treatment states include Class I treatment state, Class II treatment state and Class III treatment state. Class I treatment state is normal physiological state and normal treatment effect. Class II treatment state is normal physiological state and abnormal treatment effect. Class III treatment state is abnormal physiological state and abnormal or normal treatment effect. Several historical control strategies are obtained by retrieving historical data. A control strategy generator is constructed using the whale optimization algorithm and trained using the historical control strategies. When the treatment status classification information is a type II or type III treatment status, the optimal control strategy that matches the current treatment status is obtained through the control strategy generator to control the target percutaneous ear vagus nerve stimulation device, so as to ensure the treatment safety and efficacy of the target user.

6. A treatment monitoring method for a percutaneous vagus nerve stimulation device according to claim 5, characterized in that, The process of controlling the target percutaneous vagus nerve stimulation device by obtaining the optimal control strategy that conforms to the current treatment situation through the control strategy generator specifically includes: The treatment status type corresponding to the current treatment course is extracted through the treatment status classification information. If it is a third-class treatment status, the physiological risk probability value of the target user is extracted and compared with the preset risk threshold. If it is greater than the preset risk threshold, a stop treatment command is generated and an early warning is issued. If the risk threshold is less than the preset risk threshold, the multimodal treatment status characteristics are obtained, and the optimal treatment delay control strategy is obtained by using the control strategy generator to reduce the treatment control parameters set by the target percutaneous ear vagus nerve stimulation device. If it is a Class II treatment state, then based on the characteristics of the multimodal treatment state, the optimal treatment course change strategy is obtained through the regulation strategy generator to optimize the diagnostic and treatment control parameters set for the target percutaneous ear vagus nerve stimulation device.

7. A treatment monitoring method for a percutaneous vagus nerve stimulation device according to claim 6, characterized in that, The process of generating a strategy using a control strategy generator based on the treatment status type of the current treatment course specifically includes: An objective function is generated based on the treatment posture type. The objective function is set to maximize the treatment response index and minimize the physiological risk probability, and is constrained by the comfort index. Several initial control strategies are obtained by similarity retrieval in a preset strategy space in combination with the input multimodal treatment posture features. The preset strategy space is constructed by historical control strategies. Subsequently, each initial regulation strategy is encoded as a whale individual, and the stimulation parameters of the corresponding initial regulation strategy are used as position vectors to form an initial whale population. The regulation parameter features corresponding to each initial strategy are extracted and randomly sampled. Perturbed whale individuals are generated by cross-combination and Gaussian noise addition, and then merged into the initial whale population for individual enrichment to generate a perturbed whale population. The perturbed whale population was clustered using the K-means clustering algorithm. In the first clustering, a whale individual was randomly selected from the perturbed whale population as the cluster center, and K-1 cluster centers were selected according to the maximum distance principle. In subsequent clustering, the cluster centers were recalculated based on the mean value of whale individuals in each current cluster, and each whale individual was assigned to the nearest cluster according to the minimum distance principle. Obtain the clustered whale populations after clustering, solve the fitness problem for each clustered whale population using the set objective function, and select the globally optimal whale individual and the optimal whale individual in each cluster based on the fitness values ​​obtained from the solution. During the position update phase, for each whale individual, the current global best whale individual or the best whale individual corresponding to its own cluster is randomly selected as the leader whale. The position update strategy of the whale individual is determined by random probability and coefficient vector, including the prey encirclement phase, the bubble attack phase, and the prey search phase. After the location update is completed, it is determined whether the preset stopping criterion is met. If the preset stopping criterion is not met, the clustering operation and location update operation are repeated iteratively based on the whale population after the location update until the preset stopping criterion is reached and the final clustered whale population is output. The optimal treatment control strategy is generated by obtaining the final global optimal solution through the final clustering of whale populations, and the treatment control parameters set by the target percutaneous vagus nerve stimulation device are managed.

8. A treatment monitoring method for a percutaneous vagus nerve stimulation device according to claim 1, characterized in that, The construction of the treatment status prediction model, obtaining the multimodal treatment status characteristics after regulation, and generating a floating regulation boundary based on the treatment status prediction model, and using the floating regulation boundary to dynamically correct the diagnosis and treatment regulation plan to ensure efficacy and treatment safety, specifically includes: A treatment status prediction model was built based on the generative adversarial network framework. Several historical treatment cases were obtained by big data retrieval. The treatment status characteristics of each historical treatment case were extracted and generated into a historical treatment status feature sequence by time-series processing. Based on the historical treatment status feature sequence, the Markov algorithm is used to take the treatment status feature corresponding to each time stamp as a state node, calculate the state transition probability between the state node and the next state node to generate a state transition matrix; the state transition matrix is ​​used as a discriminator constraint, and a training dataset is generated through the historical treatment status feature sequence to train the constructed treatment status prediction model. After the target percutaneous vagus nerve stimulation device is modulated, continuous treatment monitoring is performed to obtain the multimodal treatment status characteristics after modulation. These characteristics are then input into a trained treatment status prediction model to obtain a sequence of predicted physiological risk probabilities and a sequence of predicted treatment effects for future time periods. Based on the physiological risk probability value sequence, the extreme value theory method is used to calculate the distribution characteristics of the maximum points of the physiological risk probability value sequence through the kernel density estimation method, and the high quantile of the maximum point distribution is taken to generate the floating upper limit boundary. Simultaneously, based on the predicted value sequence of treatment effect, the first-order rate of change at each time point is calculated through difference operation, and the sequence is decomposed into trend component, periodic component and residual component using trend decomposition method; the average rate of change is calculated through trend component, and combined with the expected efficacy of the preset treatment plan, a floating lower limit boundary is generated; a floating control boundary is generated based on the floating upper limit boundary and the floating lower limit boundary. In subsequent treatment adjustments, the obtained real-time optimal treatment control strategy is compared with the floating control boundary. The control parameters that exceed the floating control boundary are projected into the boundary range using the least squares projection algorithm to generate a corrected real-time optimal treatment control strategy for treatment management.

9. A treatment monitoring system for a percutaneous vagus nerve stimulation device, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory contains a treatment monitoring method program for a percutaneous vagus nerve stimulation device. When the processor executes the treatment monitoring method program for the percutaneous vagus nerve stimulation device, it performs the following steps: Within a treatment course, multi-source treatment status monitoring is performed on the target user to obtain multi-source treatment status monitoring data, and deep feature extraction and neural state decoding are performed to construct multimodal treatment status features; Based on the multimodal treatment situation characteristics, real-time treatment situation analysis is performed to analyze the treatment status and efficacy of the target user in the current treatment course and generate real-time treatment situation analysis information. Based on the real-time treatment status analysis information, dynamic status classification is performed to assess the treatment status of the target user in the current treatment course. When there is an abnormal treatment status, a treatment control plan is generated to manage the current treatment course. A treatment status prediction model is constructed to obtain the multimodal treatment status characteristics after regulation. The floating regulation boundary is generated by combining the treatment status prediction model with the floating regulation boundary. The diagnosis and treatment regulation plan is dynamically corrected by the floating regulation boundary to ensure efficacy and treatment safety.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a treatment monitoring method program for a percutaneous ear vagus nerve stimulation device, which, when executed by a processor, implements the steps of the treatment monitoring method for a percutaneous ear vagus nerve stimulation device as described in any one of claims 1 to 8.

Images