Method and system for selecting electrode startup contact for subthalamic nucleus deep brain stimulation

Abstract

A method and system for selecting an electrode startup contact for subthalamic nucleus (STN) deep brain stimulation, which method and system relate to the technical field of electroencephalogram data processing. The method comprises the following steps: firstly, collecting MERs signals of different brain regions of a patient to be tested (S1); respectively extracting time-sequence features and amplitude-sequence features from the MERs signals (S2); fusing the time-sequence features with the amplitude-sequence features, so as to obtain an amplitude-frequency perceptual fusion feature map of the MERs signals (S3); inputting the amplitude-frequency perceptual fusion feature map into an amplitude-frequency perceptual fusion network, and identifying the boundary of an STN and an optimal implantation position of a microelectrode, so as to determine the boundary of the STN and an implantation position of a permanent therapeutic electrode (S4); and analyzing the boundary of the STN and the implantation position of the permanent therapeutic electrode, so as to determine an optimal stimulation contact (S5). Thus, the problem of a postoperative optimal contact selection process requiring repeated tests and being time-consuming and effort-consuming is solved, thereby improving the postoperative program control efficiency, and reducing the dependence on the experience of clinicians.

Description

A method and system for selecting electrode power-on contact points for subthalamic nucleus deep brain stimulation Technical Field

[0001] The embodiments of the present application relate to the technical field of electroencephalogram (EEG) data processing, and in particular to a method and system for selecting electrode power-on contacts for deep brain stimulation of the subthalamic nucleus. Background Art

[0002] Parkinson's disease (PD) is a common progressive neurodegenerative disease that can cause uncontrollable motor symptoms such as tremors, rigidity, slow gait, and lack of coordination, which seriously affect the patient's quality of life. Deep brain stimulation (DBS) therapy has been applied in the field of neuromodulation. In DBS therapy, electrical stimulation is achieved by an implantable deep brain stimulator. The basic working principle of a deep brain stimulator is to implant a stimulating electrode at the target site of the patient's brain function (usually the subthalamic nucleus, STN). The electrode is connected to a pulse generator implanted in the chest by an extension wire implanted under the skin. The pulse generator emits stimulating electrical pulses of specified frequency and intensity, which act on the functional areas deep in the patient's brain to achieve a therapeutic effect.

[0003] DBS therapy mainly includes MERs signal monitoring, treatment electrode placement, and electrical stimulation of the optimal postoperative start-up contact. Currently, the selection of the optimal postoperative contact is usually based on a trial-and-error method, which is based on observing the patient's muscle tension and movement improvements after electrical stimulation of each contact. This process is somewhat blind, requires a lot of time and effort, and must be completed by experienced programmers. In order to achieve precise positioning of the permanent treatment electrode placement and selection of the optimal postoperative contact, neurosurgeons identify the STN boundary and the optimal position of the microelectrode implantation by fusing amplitude and timing features. However, there is currently no joint analysis algorithm for determining the permanent treatment electrode placement position and selecting the optimal postoperative start-up contact.

[0004] Summary of the Invention

[0005] The embodiments of the present application provide a method and system for selecting electrode power-on contacts for deep brain stimulation of the subthalamic nucleus, so as to solve the problem that the optimal contact selection process after surgery requires repeated experiments and is time-consuming and labor-intensive, thereby improving the efficiency of postoperative programming and reducing dependence on the experience of clinical physicians.

[0006] To solve the above technical problems, in the first aspect, an embodiment of the present application provides a method for selecting electrode power-on contacts for deep brain stimulation of the subthalamic nucleus, comprising the following steps: first, collecting MERs signals from different brain regions of the patient to be tested; extracting timing features and amplitude features from the MERs signals respectively; fusing the timing features with the amplitude features to obtain an amplitude-frequency perception fusion feature map of the MERs signal; inputting the amplitude-frequency perception fusion feature map into an amplitude-frequency perception fusion network to identify the boundary of the STN and the optimal position of microelectrode implantation, and determine the boundary of the STN and the placement position of the permanent treatment electrode; analyzing the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact.

[0007] In some exemplary embodiments, time series features and amplitude series features are extracted from MERs signals respectively, including: extracting time series features from MERs signals using a continuous wavelet transform method; and converting MERs signals into GASF visual images using a GASF method, and extracting amplitude series features based on the GASF visual images.

[0008] In some exemplary embodiments, the continuous wavelet transform is expressed as:

[0009] In formula (1) and formula (2), ψ is the wavelet basis function, a is the scaling coefficient that expands or reduces ψ(t) to ψ(t / a), a≠0, and b is the time-shift coefficient of the wavelet position.

[0010] In some exemplary embodiments, the GASF method is used to convert the MERs signal into a GASF visual image, including: normalizing the MERs time series to the interval [-1, 1] to obtain normalized MERs data; the MERs time series is represented as Y = {y1, y2, ..., y n}; convert the normalized MERs data into polar coordinates; based on the polar coordinates, obtain the GASF matrix; based on the GASF matrix, generate the GASF visual image; the polar coordinates are shown in formula (3):

[0011] The GASF matrix is ​​shown in formula (4):

[0012] where the angle φ and radius r are normalized using MERs and the corresponding timestamp t, where N is a constant factor of the regularization span.

[0013] In some exemplary embodiments, the amplitude-frequency perception fusion network uses a CBAM integrated convolutional neural network model as a classifier to identify the amplitude-frequency perception fusion feature map to obtain the STN boundary and the placement position of the permanent treatment electrode.

[0014] In some exemplary embodiments, the amplitude-frequency-aware deep fusion network uses seven stages to perform full-resolution convolution, each stage includes a convolutional layer module and a CBAM module; the last stage includes a fully connected layer and a Softmax function to obtain the output of the amplitude-frequency-aware deep fusion network.

[0015] In some exemplary embodiments, the CBAM module includes a channel attention module and a spatial attention module; given an intermediate feature map X as input, the entire CBAM process is expressed as:

[0016] Among them, M C is the channel attention module, M S is the spatial attention module, represents element-by-element multiplication, X′ represents the features obtained after the channel attention module, and X″ is the final output of CBAM.

[0017] In some exemplary embodiments, the calculation formula of the channel attention module is as follows: C (X) = σ(MLP(f AvgPool (X))+MLP(f MaxPool (X))) (6)

[0018] Among them, M C is the channel attention module, f AvgPool and f MaxPool are average pooling and maximum pooling respectively; MLP represents multi-layer perceptron, and σ is the sigmoid function;

[0019] The calculation formula of the spatial attention module is as follows: M S (X′)=σ(F cov ([f AvgPool (X); f MaxPool (X)])) (7)

[0020] Among them, M S is the spatial attention module, F cov Represents the convolution operation.

[0021] In some exemplary embodiments, the boundary of the STN and the placement position of the permanent treatment electrode are analyzed to determine the optimal stimulation contact, including: analyzing the boundary of the STN and the placement position of the permanent treatment electrode to obtain the placement strategy of the permanent treatment electrode; determining the optimal stimulation contact based on the placement strategy of the permanent treatment electrode; the placement strategy of the permanent treatment electrode is that the top of the permanent treatment electrode is close to the lower boundary of the STN; the principle of selecting the optimal stimulation contact is to cover the gathering area of ​​the optimal position for implantation of the permanent treatment electrode.

[0022] In the second aspect, the embodiment of the present application also provides an electrode power-on contact selection system for deep brain stimulation of the subthalamic nucleus, including: a MERs signal acquisition module, a feature extraction module, a feature fusion module, an identification module and an analysis module connected in sequence; the MERs signal acquisition module is used to collect MERs signals from different brain regions of the patient to be tested; the feature extraction module is used to extract timing features and amplitude features from the MERs signal respectively; the feature fusion module is used to fuse the timing features with the amplitude features to obtain an amplitude-frequency perception fusion feature map of the MERs signal; the identification module is used to input the amplitude-frequency perception fusion feature map into the amplitude-frequency perception fusion network to identify the boundary of the STN and the optimal position of the microelectrode implantation, and determine the boundary of the STN and the placement position of the permanent treatment electrode; the analysis module is used to analyze the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact.

[0023] The technical solution provided by the embodiments of the present application has at least the following advantages:

[0024] An embodiment of the present application provides a method and system for selecting electrode power-on contacts for deep brain stimulation of the subthalamic nucleus. The method includes the following steps: first, collecting MERs signals from different brain regions of a patient to be tested; extracting timing features and amplitude features from the MERs signals respectively; fusing the timing features with the amplitude features to obtain an amplitude-frequency perception fusion feature map of the MERs signal; inputting the amplitude-frequency perception fusion feature map into an amplitude-frequency perception fusion network to identify the boundary of the STN and the optimal position for microelectrode implantation, thereby determining the boundary of the STN and the placement position of a permanent treatment electrode; and analyzing the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact.

[0025] This application provides a method for selecting optimal contact points for subthalamic nucleus deep brain stimulation (STN-DBS) electrodes based on an amplitude-frequency-aware deep fusion network. This method uses amplitude-frequency-aware fusion features and a CBAM integrated CNN model to identify the STN boundary and the optimal location for microelectrode implantation. This method guides placement of permanent STN-DBS treatment electrodes based on the STN boundary and selects the optimal postoperative contact point based on the optimal microelectrode implantation location distribution. Experimental results show that the proposed method achieves an average accuracy of 96.78%, precision of 97.77%, sensitivity of 95.26%, and specificity of 98.07% for STN identification; and an average accuracy of 98.24%, precision of 98.53%, sensitivity of 96.05%, and specificity of 99.31% for optimal microelectrode implantation location identification. All results outperform existing STN and microelectrode implantation optimal location identification algorithms. The optimal contact points selected by the proposed method are consistent with successful decisions made by neurosurgeons, demonstrating the potential of the proposed method to improve surgical efficiency while reducing reliance on clinician experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] One or more embodiments are exemplarily described by the pictures in the corresponding drawings. These exemplifications do not constitute limitations on the embodiments. Unless otherwise stated, the pictures in the drawings do not constitute proportional limitations.

[0027] FIG1 is a flow chart of a method for selecting electrode power-on contact points for subthalamic nucleus deep brain stimulation according to an embodiment of the present application;

[0028] FIG2 is a schematic diagram of a specific flow chart of a method for selecting electrode power-on contact points for subthalamic nucleus deep brain stimulation according to an embodiment of the present application;

[0029] FIG3 is a schematic diagram of the CNN model structure of the CBAM integration provided in one embodiment of the present application;

[0030] FIG4 is a schematic structural diagram of a system for selecting electrode power-on contacts for subthalamic nucleus deep brain stimulation according to an embodiment of the present application;

[0031] FIG5 is a schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] As can be seen from the background technology, the current selection of the optimal contact point after surgery is usually done by a "trial and error" method, which is judged by observing the patient's muscle tension and movement improvement after electrical stimulation of each contact point. This process is somewhat blind, requires a lot of time and energy, and must be completed by an experienced programmer.

[0033] The selection of the optimal contact point after surgery should cover the entire sensorimotor area as much as possible and avoid affecting the substantia nigra. However, the sensorimotor area of ​​the STN is very small, so it is very difficult for clinicians to locate the boundaries and sensorimotor areas of the STN. The optimal location cluster area for microelectrode implantation identified by MERs signals can usually be considered to be the sensorimotor area. Based on this principle, doctors can be assisted in accurately identifying the optimal contact point for the treatment electrode. In addition, the placement position of the permanent treatment electrode will also affect the selection of the optimal contact point. Therefore, the determination of the placement position of the permanent treatment electrode and the selection of the optimal contact point are indispensable for the success of STN-DBS surgery.

[0034] Clinicians can usually also use MERs signals to identify the best location for microelectrode implantation when selecting the optimal contact point. A related technology proposes a method for classifying MERs signals using GASF features for automatic identification of the best location for microelectrode implantation. The GASF method converts the one-dimensional time series signal recognition problem into an image classification problem using artificial intelligence (AI) methods by converting time series data into two-dimensional GASF images. For some MERs signals with high-frequency and high-amplitude features, the stripe features displayed by the GASF image are not obvious enough. However, for MERs signals with unclear peaks, the time series feature extraction method is not effective. Therefore, the MERs signal recognition algorithm mentioned above that only analyzes the time series features or the GASF amplitude series features will affect the recognition effect of the MERs signal.

[0035] To achieve precise positioning of the permanent treatment electrode placement and selection of the optimal postoperative contact point, neurosurgeons identify the STN boundary and the optimal position for microelectrode implantation by fusing amplitude and time series features. However, there is currently no joint analysis algorithm for determining the permanent electrode placement position and selecting the optimal postoperative start-up contact point. Clinically, the placement position of the permanent treatment electrode is identified by neurosurgeons based on MERs signal characteristics, and then the optimal start-up contact point is determined through a trial-and-error strategy. The MERs signal category is manually calibrated into five levels by experienced clinicians based on peak, frequency, and amplitude characteristics. The optimal position for microelectrode implantation is marked as the top two levels, and the optimal electrode contact point is usually determined based on the clustering area of ​​the optimal microelectrode implantation position.

[0036] To solve the above technical problems, an embodiment of the present application provides a method for selecting electrode power-on contacts for deep brain stimulation of the subthalamic nucleus, comprising the following steps: first, collecting MERs signals from different brain regions of the patient to be tested; extracting timing features and amplitude features from the MERs signals respectively; fusing the timing features with the amplitude features to obtain an amplitude-frequency perception fusion feature map of the MERs signal; inputting the amplitude-frequency perception fusion feature map into an amplitude-frequency perception fusion network to identify the boundary of the STN and the optimal position for microelectrode implantation, and determine the boundary of the STN and the placement position of the permanent treatment electrode; analyzing the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact. An embodiment of the present application provides a method for selecting electrode power-on contacts for deep brain stimulation of the subthalamic nucleus to solve the problem that the optimal contact selection process after surgery requires repeated experiments and is time-consuming and labor-intensive, thereby improving the efficiency of postoperative programming and reducing dependence on the experience of clinicians.

[0037] The following detailed description of the various embodiments of the present application is provided in conjunction with the accompanying drawings. However, those skilled in the art will appreciate that many technical details are provided in the various embodiments of the present application to facilitate a better understanding of the present application. However, even without these technical details and the various variations and modifications based on the following embodiments, the technical solutions claimed in the present application can still be implemented.

[0038] 1 , an embodiment of the present application provides a method for selecting electrode power-on contact points for deep brain stimulation of the subthalamic nucleus, including the following steps:

[0039] Step S1: collecting MERs signals from different brain regions of the patient to be tested.

[0040] Step S2: extracting time series features and amplitude series features from the MERs signal.

[0041] Step S3: Fuse the time series features with the amplitude series features to obtain the amplitude-frequency perception fusion feature map of the MERs signal.

[0042] Step S4: input the amplitude-frequency perception fusion feature map into the amplitude-frequency perception fusion network to identify the boundary of the STN and the optimal position for microelectrode implantation, thereby determining the boundary of the STN and the placement position of the permanent treatment electrode.

[0043] Step S5: Analyze the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact point.

[0044] To address the difficulty of selecting the optimal contact point after surgery, which requires repeated trials and is time-consuming and laborious, this application proposes a method for selecting the optimal contact point for STN-DBS electrodes based on an amplitude-frequency-aware deep fusion network. First, it is proposed to combine the time series characteristics of the MERs signal with the generated GASF features to obtain the amplitude-frequency-aware fusion features. Then, an amplitude-frequency-aware deep fusion network is proposed to classify the graded signals recorded by microelectrodes of different frequencies and amplitudes. This can help clinicians effectively identify the STN boundary and the optimal location for microelectrode implantation, thereby achieving the optimal contact point selection for permanent DBS electrodes.

[0045] Specifically, this application proposes a method for selecting optimal contact points for STN-DBS electrodes based on an amplitude-frequency-aware deep fusion network. This method first identifies the STN boundary to guide permanent treatment electrode placement, then determines the cluster of optimal microelectrode implantation locations, achieving the goal of accurately selecting the optimal postoperative start-up contact point. Specifically, this application utilizes two-dimensional time series images and GASF images to characterize the high-frequency and high-amplitude features of MERs time series signals, respectively. This method transforms the task of hierarchical time series signal recognition into a visual image feature classification task using two-dimensional fusion with AI technology. Furthermore, this application proposes an amplitude-frequency-aware deep fusion network to automatically identify graded MERs signals based on different frequencies and amplitudes, helping clinicians automatically identify STN boundaries and optimal microelectrode implantation locations. Finally, this application also proposes a method for automated decision-making during and after STN-DBS surgery. This method accurately identifies the STN boundary for permanent treatment electrode placement and optimal start-up contact selection. Experimental results show that neurosurgeons and the proposed method select the same optimal start-up contact point. Furthermore, this method outperforms current state-of-the-art methods in identifying optimal STN and microelectrode implantation locations.

[0046] The overall flow chart of the method for selecting the optimal contact points of STN-DBS electrodes based on the amplitude-frequency perception deep fusion network proposed in this application is shown in Figure 2. In order to address the difficulties in identifying discontinuous high-frequency, high-amplitude signals and MERs signals with unclear peaks, and the difficulty in selecting the optimal contact points after surgery, which requires repeated trials and is time-consuming and laborious, the entire selection process is summarized into two stages: automatic identification of MERs signals based on the amplitude-frequency perception deep fusion network and selection of the optimal stimulation contact points. The difficulties during and after STN-DBS surgery mainly include STN boundary positioning, identification of the optimal implantation position of the microelectrode, and selection of the optimal contact points for permanent treatment electrodes. This application provides an intelligent decision-making solution for these difficulties during and after STN-DBS surgery. First, the amplitude-frequency perception fusion feature map of the MERs signal is obtained by fusing the time series features extracted by CWT with the GASF features. Then, the CBAM integrated CNN model is used to identify the amplitude-frequency perception fusion feature map, which is used to identify the boundary of the STN and the optimal implantation position of the microelectrode. Once the STN border is determined, the principle for placement of the permanent DBS electrode is to ensure that the electrode tip reaches the inferior border of the STN as much as possible without affecting the substantia nigra, while also avoiding adverse reactions in the patient. Finally, a comprehensive analysis of the placement of the permanent treatment electrode and the distribution of optimal microelectrode implantation locations can determine the optimal stimulation contact point.

[0047] Step S1 is the process of acquiring MERs signals. MERs signals are acquired using an electrophysiological navigation system (NeuroNav 223 / 12, Alpha Omega Engineering, Israel). Neurosurgeons first use MRI images to preliminarily determine the target location and plan the microelectrode implantation trajectory in combination with CT scanning technology. Then, the microelectrodes are implanted into the STN nucleus according to the preset target location and the planned electrode trajectory. Generally, the MERs measurement method is as follows: (1) Signal acquisition is performed by implanting one or two tungsten microelectrodes; (2) The microelectrode (cannula with a needle core) is inserted approximately 10 mm above the preset target; (3) The microelectrode is further moved in steps of approximately 1 mm until it is approximately 5 mm from the preset target, and a MERs signal is measured at each step; (4) The microelectrode is further advanced in steps of approximately 0.5 mm until it is approximately 4.0 to 5.0 mm below the target, and a MERs signal is measured at each step. Microelectrode movement can generate noise interference, and clinicians typically need to wait for a period of time at each microelectrode position for the MERs signal to stabilize. Therefore, the duration of MERs collected at each microelectrode position varies, typically ranging from 1 to 5 seconds. Considering that varying MERs data lengths may affect signal processing results, to eliminate noise interference and maintain MERs signal stability, a 1-second MERs signal segment with minimal noise was selected for each position, as recommended by a professional neurosurgeon. The raw MERs signals were preprocessed by a professional neurosurgeon using Mapfile Converter software (copyright Alpha Omega 2018, version 5.1.11.0). MERs were then cropped and labeled using non-overlapping 1-second sliding windows using MATLAB software (Mathworks, Inc., Natick, MA). All signals were sampled at 24 kHz and amplified at 1000 times. Digital filtering was performed in MATLAB using a 6-pole Butterworth digital bandpass filter from 0.5 kHz to 5 kHz and a 60 Hz notch filter. All raw MERs signals are downsampled to 12kHz using appropriate anti-aliasing filtering, which reduces the number of data sampling points and improves signal recognition efficiency. Microelectrodes typically penetrate the following brain regions, primarily including Rt, IC, ZI, H2, STN, and SN. Because different brain regions have different electrophysiological signals, the differences in the MERs signal characteristics collected from different brain regions can effectively identify the STN boundary and the optimal location for microelectrode implantation. Based on the STN boundary location identified by the collected MERs signals and the implantation path detected by the microelectrode, neurosurgeons can determine the implantation trajectory of the permanent treatment electrode, then remove all microelectrodes and place the final four-contact permanent treatment electrode in the STN.During the postoperative period, clinicians selected the optimal contact points for improving motor symptoms in Parkinson's patients based on their clinical experience. This study was approved by the Human Research Ethics Committee of the local hospital. Written informed consent was obtained from all participants after a thorough explanation of the procedures involved.

[0048] In some embodiments, step S2 extracts timing features and amplitude sequence features from the MERs signal respectively, including: extracting timing features from the MERs signal using a continuous wavelet transform method; and converting the MERs signal into a GASF visual image using a GASF method, and extracting amplitude sequence features based on the GASF visual image.

[0049] Specifically, during STN-DBS surgery, the microelectrode implantation trajectory begins in the prefrontal motor cortex and continues into the substantia nigra. To avoid electrical stimulation of the substantia nigra, precise determination of the treatment electrode placement is crucial during surgery. After the treatment electrode is implanted in the STN, the optimal contact point can be selected based on the distribution of optimal microelectrode implantation locations during surgery. Therefore, the challenge of STN-DBS lies in identifying the STN boundary to determine the treatment electrode placement location, as well as identifying the optimal microelectrode implantation location and selecting the optimal postoperative contact point. Because MERs signals from different brain tissues exhibit distinct characteristics, the STN boundary and optimal microelectrode implantation location can be identified based on these MERs. To identify the optimal microelectrode implantation location, experienced clinicians typically categorize MERs signals into five levels based on peak value, frequency, and amplitude. MERs located at the optimal microelectrode implantation location are classified into the top two levels, while the remaining MERs are located at non-optimal locations. Therefore, identifying the STN boundary and optimal microelectrode implantation location can be formulated as a typical three-class classification problem: MERs from non-STN locations, optimal locations within the STN, and non-optimal locations within the STN.

[0050] Because MERs signals exhibit complex characteristics such as randomness, multi-components, and non-stationarity, converting one-dimensional MERs signals into two-dimensional images can overcome the challenges of analyzing their complex one-dimensional features. The two most commonly used feature extraction methods for converting MERs time series images are time series feature extraction and amplitude series feature extraction. The continuous wavelet transform, as a time-frequency domain transform, has advantages in analyzing non-stationary signals and can effectively extract time series features.

[0051] In some embodiments, given a time series signal, the continuous wavelet transform (CWT) is expressed as:

[0052] In formula (1) and formula (2), ψ is the wavelet basis function, a is the scaling coefficient that expands or reduces ψ(t) to ψ(t / a), a≠0, and b is the time-shift coefficient of the wavelet position.

[0053] In some embodiments, the GASF method is used to convert the MERs signal into a GASF visual image, including: normalizing the MERs time series to the interval [-1, 1] to obtain normalized MERs data; the MERs time series is represented as Y = {y1, y2, ..., y n}; Convert the normalized MERs data into polar coordinates; Based on the polar coordinates, obtain the GASF matrix; Based on the GASF matrix, generate a GASF visual image.

[0054] Complex Morlet wavelet is a widely used wavelet basis function. It has the best time-frequency resolution and the best focusing performance for time series feature extraction in CWT. Therefore, this application selects complex Morlet wavelet (cmor3-3) as the wavelet basis function. Since the longer the scale sequence, the higher the frequency resolution, and a scale sequence that is too long may cause a loss of time resolution, the scale sequence length is set to 256. Since the sampling frequency of the signal is 12kHz, the sampling time is, the total time series length is 1s, and the time-shift coefficient sequence can be obtained along the time axis through the sampling time. The wavelet coefficients are calculated by changing the time-shift coefficients to obtain the final time-frequency feature map. Using the typical time series feature extraction method CWT, the MERs signals at the optimal position of STN, the non-optimal position of STN and non-STN can be generated into a 3D time-frequency image.

[0055] In order to obtain the amplitude sequence characteristics of the MERs signal, the GASF method is used to convert the MERs signal into a GASF visual image. Given the MERs time series, Y is normalized to the interval [-1, 1]. The normalized MERs data is then converted to polar coordinates, as shown in formula (3):

[0056] By examining the triangular sum, the time correlation within different time intervals can be easily identified, and the GASF matrix is ​​shown in formula (4):

[0057] where the angle φ and radius r are normalized using MERs and the corresponding timestamp t, where N is a constant factor of the regularization span.

[0058] The GASF matrix is ​​then used to generate a GASF image of size 100 × 100 × 3. The GASF images generated by the three types of MERs signals from the optimal position of STN, the non-optimal position of STN, and non-STN have different texture and stripe characteristics.

[0059] In some embodiments, the amplitude-frequency perception fusion network uses a CBAM integrated convolutional neural network model as a classifier to identify the amplitude-frequency perception fusion feature map to obtain the STN boundary and the placement position of the permanent treatment electrode.

[0060] To accurately identify the STN boundary and the optimal location for microelectrode implantation, the time-series feature map obtained by CWT is fused with the amplitude-series feature map obtained by the GASF method to obtain an amplitude-frequency-aware fusion feature map with a dimension of 100 × 100 × 6. Because the attention mechanism can focus on extracting feature information that is beneficial for classification from the amplitude-frequency-aware fusion features of the MERs signal and suppress irrelevant neural activity features, this application proposes a CBAM-integrated CNN model as a classifier for identifying the amplitude-frequency-aware fusion feature map, which can accurately identify the STN boundary and the optimal location for microelectrode implantation.

[0061] In some embodiments, the amplitude-frequency-aware deep fusion network uses seven stages to perform full-resolution convolution, each stage includes a convolutional layer module and a CBAM module; the last stage includes a fully connected layer and a Softmax function to obtain the output of the amplitude-frequency-aware deep fusion network.

[0062] Specifically, to extract richer information in classification tasks while avoiding information loss caused by pooling, the amplitude-frequency-aware deep fusion network uses seven stages of full-resolution convolution. Furthermore, considering that the input image size is set to 100×100×3, the receptive field of the neurons is sufficient to cover the input after four stages of convolution. Therefore, pooling layers are used in the final three stages to obtain high-level features for classification tasks. As shown in Figure 3, the amplitude-frequency-aware deep fusion network consists of seven stages, each of which contains a convolutional layer module and a CBAM module. The final stage is a fully connected layer and a softmax function to obtain the output of the entire network. Maxpooling layers follow the fourth, fifth, and sixth stages, respectively. Each convolutional layer is followed by batch normalization and ReLU.

[0063] As shown in FIG3 , in some embodiments, the CBAM module includes a channel attention module and a spatial attention module. Given an intermediate feature map X as input, the entire CBAM process is expressed as:

[0064] Among them, M Cis the channel attention module, M S is the spatial attention module, represents element-by-element multiplication, X′ represents the features obtained after the channel attention module, and X″ is the final output of CBAM.

[0065] For the channel attention module, the input feature map undergoes both average pooling and maximum pooling, and then passes through a multi-layer perceptron. The multi-layer perceptron contains a shared MLP, and then the output features of the MLP are element-wise summed and then output through a sigmoid activation function. The calculation formula of the channel attention module is as follows: C (X) = σ(MLP(f AvgPool (X))+MLP(f MaxPool (X))) (6)

[0066] Among them, M C is the channel attention module, f AvgPool and f MaxPool are average pooling and maximum pooling respectively; MLP represents multi-layer perceptron, and σ is the sigmoid function.

[0067] The output feature map of the channel attention module is then used as the input feature map of the spatial attention module. First, the input features are processed by maximum pooling and average pooling, and compressed sampling is performed on the channel dimension. Then, a convolution layer is passed to obtain a two-dimensional spatial attention feature map. The calculation formula of the spatial attention module is as follows: S (X′)=σ(F cov ([f AvgPool (X); f MaxPool (X)])) (7)

[0068] Among them, M S is the spatial attention module, F cov Represents a convolution operation with a convolution kernel size of 7×7.

[0069] The CBAM module can effectively learn to emphasize or suppress features and effectively refine intermediate features. In the amplitude-frequency perception deep fusion network proposed in this application, the number of channels of the convolutional layers in stages 1 to 7 are 16, 32, 64, 128, 128, 128, 128, and 128, respectively. The convolution kernel size is 3×3, and the kernel size of the maximum pooling layer is 2×2. The model network is initialized using the "random_uniform" method, and the Adam algorithm is used for network optimization. The learning rate and batch size are set to 1e-2 and 32, respectively. During the training phase, the number of iterations is set to 50.

[0070] According to the above model analysis, the MERs signal recognition algorithm process based on the amplitude-frequency perception deep fusion network is shown in Table 1:

[0071] Table 1 Implementation process of the amplitude-frequency perception fusion network algorithm

[0072] The specific algorithm is described as follows:

[0073] As shown in Table 1, D CWT ={a1,a2,…,a 100} and D GASF ={b1,b2,…,b 100} represent the sample sets composed of the feature maps generated by the CWT and GASF methods, respectively, where a1 and b1 represent the CWT and GASF feature map sets of the first patient, respectively, and so on. First, D CWT and D GASF The sample set is spliced ​​to obtain the fusion feature map sample set X. The sample set X represents the set of two fusion feature maps generated by the MERs data of 100 people. Second, the 100-person sample set is split into 80 people for training and 20 people for testing, and the training set X can be obtained. tr ={c1,c2,…,c 80} and test set X test ={c 81 ,c 82 ,…,c 100}. Third, X tr 8 people in a group, divided into 10 blocks X tr ={x1,x2,…,x 10}, where represents a set of feature maps consisting of 8 people in a group. Fourth, a ten-fold cross method is used to use a piece of data in the training set as the validation set X val , then the rest of the data in the training set is updated as a new training set, and the new training set and validation set are input into the CBAM enhanced network model to train the model parameters. Fifth, in the testing process, the test set is input into the trained model to obtain the predicted label value y pre , and then according to the test set label y test By calculating the confusion matrix between the predicted results and the actual results, we can obtain the classification accuracy, precision and other evaluation results of the algorithm, which are used to evaluate the model performance.

[0074] In some embodiments, in step S5, the boundary of the STN and the placement position of the permanent treatment electrode are analyzed to determine the optimal stimulation contact, including: analyzing the boundary of the STN and the placement position of the permanent treatment electrode to obtain the placement strategy of the permanent treatment electrode; determining the optimal stimulation contact based on the placement strategy of the permanent treatment electrode; the placement strategy of the permanent treatment electrode is that the top of the permanent treatment electrode is close to the lower boundary of the STN; the principle of selecting the optimal stimulation contact is to cover the gathering area of ​​the optimal position for implantation of the permanent treatment electrode.

[0075] The selection of optimal stimulation contacts includes the placement strategy of permanent treatment electrodes and the selection strategy of optimal activation contacts after surgery.

[0076] Regarding the placement strategy of permanent treatment electrodes, their placement location significantly impacts the selection of the optimal contact point. This is because if the electrodes are placed too deeply, exceeding the lower border of the STN and contacting the substantia nigra, adverse consequences for the patient can occur, including severe complications. If the electrodes are not fully implanted within the STN, coverage of the target area for postoperative activation programming is insufficient, and the patient's motor status cannot be fully improved. Before permanent treatment electrode placement, microelectrodes must be used for multiple probing to locate the STN boundary and the optimal microelectrode placement location. Section 3 describes the main process for identifying the STN boundary and the optimal electrode placement location based on MERs signals. Because permanent treatment electrode placement is subject to multiple constraints, developing a reasonable permanent treatment electrode placement strategy based on the STN boundary, the optimal microelectrode placement location, and other constraints is challenging.

[0077] Permanent treatment electrode placement is subject to several constraints, including STN size, permanent treatment electrode size, and the area covered by the contact stimulation, as described below:

[0078] (1) STN size. Based on the MERs signal characteristics, the STN boundary range is from -1 to 4.5 points, so the actual STN size is 5.5 mm. In addition, the lower boundary of the STN also has obvious differences. Therefore, the placement of permanent treatment electrodes must strictly refer to the actual identified STN area.

[0079] (2) The size of the permanent treatment electrode. The permanent treatment electrode is approximately 7.5 mm long and consists of four 1.5 mm long contacts spaced 0.5 mm apart. Each contact can be electrically stimulated to program the target area, and the programmable area is also limited.

[0080] (3) Coverage area of ​​contact electrical stimulation. The optimal electrode contact and electrical stimulation should cover the STN as much as possible, especially the sensorimotor area, while avoiding stimulation of the SN to prevent side effects of postoperative programming on the patient.

[0081] Therefore, based on a comprehensive analysis of the three limiting factors of STN size, permanent treatment electrode size, and contact electrical stimulation coverage area, this application summarizes the optimal position for DBS permanent treatment electrode placement as the top of the electrode being close to the lower border of STN.

[0082] Regarding the strategy for selecting the optimal postoperative activation contact, after permanent treatment electrodes are placed in the STN and a pulse generator is implanted in the chest, the STN-DBS procedure is complete. After STN-DBS surgery, patients undergo activation programming, which involves administering electrical stimulation to the optimal contact of the treatment electrode to improve motor function. The optimal activation contact should stimulate the sensorimotor area as closely as possible to achieve optimal therapeutic efficacy. Since the optimal microelectrode implantation location cluster can be considered the sensorimotor area of ​​the STN, the optimal stimulation contact can be determined based on the distribution of the optimal microelectrode implantation location. Care should also be taken to avoid electrical stimulation affecting the SN. The MERs signal can effectively identify the optimal microelectrode implantation location. The principle for selecting the optimal stimulation contact is to cover the optimal microelectrode implantation location cluster. The optimal microelectrode implantation location cluster is primarily located within the electrode depth range of 1 to 2.5 mm. The electrode contact covered in this range is labeled 2. Therefore, the optimal postoperative stimulation contact selection is based on the optimal microelectrode implantation location cluster.

[0083] The method for selecting the electrode power-on contact for deep brain stimulation of the subthalamic nucleus provided by this application was experimentally verified, and the results confirmed that the recognition accuracy of the optimal position of the microelectrode implantation of this application was 97.47%, which can meet the clinical accuracy requirements. In addition, this application can accurately identify the optimal stimulation contacts of 18 patients from 20 test patients, so the accuracy of the optimal stimulation contacts is Accuracy = 18 / (20*100% =) 90%. In clinical practice, clinicians usually select the optimal and suboptimal electrical stimulation contacts as the final candidate contacts. In combination with clinical practice, taking the suboptimal contacts into consideration in the evaluation of the final results, this application can effectively identify the optimal contacts of 20 test patients with an accuracy of 100%, meeting the requirements of precision surgery. Therefore, the selection of this application is consistent with the selection of clinicians and can provide useful guidance for neurosurgeons.

[0084] 4 , an embodiment of the present application further provides a system for selecting electrode power-on contact points for deep brain stimulation of the subthalamic nucleus, comprising: a MERs signal acquisition module 101, a feature extraction module 102, a feature fusion module 103, an identification module 104, and an analysis module 105 connected in sequence; the MERs signal acquisition module 101 is used to collect MERs signals from different brain regions of the patient to be tested; the feature extraction module 102 is used to extract timing features and amplitude features from the MERs signal respectively; the feature fusion module 103 is used to fuse the timing features with the amplitude features to obtain an amplitude-frequency perception fusion feature map of the MERs signal; the identification module 104 is used to input the amplitude-frequency perception fusion feature map into an amplitude-frequency perception fusion network to identify the boundary of the STN and the optimal position of microelectrode implantation, and determine the boundary of the STN and the placement position of the permanent treatment electrode; the analysis module 105 is used to analyze the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact point.

[0085] Referring to Figure 5, another embodiment of the present application provides an electronic device, including: at least one processor 110; and a memory 111 communicatively connected to the at least one processor; wherein the memory 111 stores instructions that can be executed by the at least one processor 110, and the instructions are executed by the at least one processor 110 to enable the at least one processor 110 to execute any of the above method embodiments.

[0086] The memory 111 and the processor 110 are connected using a bus. The bus may include any number of interconnected buses and bridges. The bus connects various circuits of one or more processors 110 and the memory 111. The bus may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits. These are all well known in the art and are therefore not described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver may be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor 110 is transmitted over a wireless medium via an antenna. Furthermore, the antenna receives data and transmits the data to the processor 110.

[0087] The processor 110 is responsible for managing the bus and general processing, and may also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 111 may be used to store data used by the processor 110 when performing operations.

[0088] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program, which implements the above method embodiment when executed by a processor.

[0089] Based on the above technical solution, the embodiment of the present application provides a method and system for selecting electrode power-on contacts for deep brain stimulation of the subthalamic nucleus, which includes the following steps: first, collecting MERs signals from different brain regions of the patient to be tested; extracting timing features and amplitude features from the MERs signals respectively; fusing the timing features with the amplitude features to obtain an amplitude-frequency perception fusion feature map of the MERs signal; inputting the amplitude-frequency perception fusion feature map into an amplitude-frequency perception fusion network to identify the boundary of the STN and the optimal position of microelectrode implantation, and determine the boundary of the STN and the placement position of the permanent treatment electrode; analyzing the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact.

[0090] This application provides a method for selecting optimal contact points for subthalamic nucleus deep brain stimulation (STN-DBS) electrodes based on an amplitude-frequency-aware deep fusion network. This method uses amplitude-frequency-aware fusion features and a CBAM integrated CNN model to identify the STN boundary and the optimal location for microelectrode implantation. This method guides placement of permanent STN-DBS treatment electrodes based on the STN boundary and selects the optimal postoperative contact point based on the optimal microelectrode implantation location distribution. Experimental results show that the proposed method achieves an average accuracy of 96.78%, precision of 97.77%, sensitivity of 95.26%, and specificity of 98.07% for STN identification; and an average accuracy of 98.24%, precision of 98.53%, sensitivity of 96.05%, and specificity of 99.31% for optimal microelectrode implantation location identification. All results outperform existing STN and microelectrode implantation optimal location identification algorithms. The optimal contact points selected by the proposed method are consistent with successful decisions made by neurosurgeons, demonstrating the potential of the proposed method to improve surgical efficiency while reducing reliance on clinician experience.

[0091] That is, those skilled in the art will understand that all or part of the steps in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a program, which is stored in a storage medium and includes a number of instructions for causing a device (which may be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps in the above-mentioned methods of each embodiment of the present application. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., various media that can store program code.

[0092] Those skilled in the art will appreciate that the above-described embodiments are specific examples for implementing the present application, and that in actual applications, various changes in form and detail may be made thereto without departing from the spirit and scope of the present application. Any person skilled in the art may make changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be subject to the scope defined in the claims.

Claims

1. A method for selecting electrode power-on contact points for deep brain stimulation of the subthalamic nucleus, characterized in that: include: Collect MERs signals from different brain regions of the patient to be tested; Extracting time series features and amplitude series features from the MERs signal respectively; The time series feature is fused with the amplitude series feature to obtain an amplitude-frequency perception fusion feature map of the MERs signal; Inputting the amplitude-frequency sensing fusion feature map into the amplitude-frequency sensing fusion network, identifying the boundary of STN and the optimal position for microelectrode implantation, and determining the boundary of STN and the placement position of the permanent treatment electrode; The boundaries of the STN and the placement of the permanent treatment electrode are analyzed to determine the optimal stimulation contact point.

2. The method for selecting electrode power-on contacts for subthalamic nucleus deep brain stimulation according to claim 1, characterized in that: Extracting time series features and amplitude series features from the MERs signal respectively includes: A continuous wavelet transform method is used to extract time series features from the MERs signal; and a GASF method is used to convert the MERs signal into a GASF visual image, and based on the GASF visual image, amplitude series features are extracted.

3. The method for selecting electrode power-on contacts for subthalamic nucleus deep brain stimulation according to claim 2, characterized in that: The continuous wavelet transform is expressed as: In formula (1) and formula (2), ψ is the wavelet basis function, a is the scaling coefficient that expands or reduces ψ(t) to ψ(t / a), a≠0, and b is the time shift coefficient of the wavelet position.

4. The method for selecting electrode power-on contacts for subthalamic nucleus deep brain stimulation according to claim 2, characterized in that: The MERs signal is converted into a GASF visual image using a GASF method, including: Normalize the MERs time series to the interval [-1,1] to obtain the normalized MERs data; MERs time series is represented as Y = {y1, y2, ..., y n }; Converting the normalized MERs data into polar coordinates; Based on the polar coordinates, a GASF matrix is ​​obtained; Based on the GASF matrix, generating a GASF visual image; The polar coordinates are shown in formula (3): The GASF matrix is ​​shown in formula (4): where the angle φ and radius r are normalized using the MERs The values ​​of and the corresponding timestamp t are represented by N, where N is a constant factor of the regularization span.

5. The method for selecting electrode power-on contact points for subthalamic nucleus deep brain stimulation according to claim 1, characterized in that: The amplitude-frequency perception fusion network adopts the CBAM integrated convolutional neural network model as a classifier to identify the amplitude-frequency perception fusion feature map to obtain the STN boundary and the placement position of the permanent treatment electrode.

6. The method for selecting electrode power-on contact points for subthalamic nucleus deep brain stimulation according to claim 5, characterized in that: The amplitude-frequency-aware deep fusion network uses seven stages to perform full-resolution convolution, each stage includes a convolution layer module and a CBAM module; the last stage includes a fully connected layer and a Softmax function to obtain the output of the amplitude-frequency-aware deep fusion network.

7. The method for selecting electrode power-on contacts for subthalamic nucleus deep brain stimulation according to claim 6, characterized in that: The CBAM module includes a channel attention module and a spatial attention module; Given an intermediate feature map X as input, the entire CBAM process is expressed as: Among them, M c is the channel attention module, M S is the spatial attention module, represents element-by-element multiplication, X′ represents the features obtained after the channel attention module, and X″ is the final output of CBAM.

8. The method for selecting electrode power-on contact points for subthalamic nucleus deep brain stimulation according to claim 7, characterized in that: The calculation formula of the channel attention module is as follows: C (X) = σ(MLP(f AvgPool (X))+MLP(f MaxPool (X))) (6) Among them, M C is the channel attention module, f AvgPool and f MaxPool They are average pooling and maximum pooling respectively; MLP stands for multi-layer perceptron, and σ is the sigmoid function; The calculation formula of the spatial attention module is as follows: M S (X′)=σ(F cov ([f AvgPool (X);f MaxPool (X)])) (7) Among them, M S is the spatial attention module, F cov Represents a convolution operation.

9. The method for selecting electrode power-on contact points for subthalamic nucleus deep brain stimulation according to claim 1, characterized in that: Analyzing the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact point includes: Analyzing the boundary of the STN and the placement position of the permanent treatment electrode to obtain a permanent treatment electrode placement strategy; determining an optimal stimulation contact point based on the permanent treatment electrode placement strategy; The placement strategy of the permanent treatment electrode is that the top of the permanent treatment electrode is close to the lower boundary of the STN; The optimal stimulation contacts are selected based on the principle of covering a concentrated area of ​​optimal location for permanent therapeutic electrode implantation.

10. A system for selecting electrode start-up contacts for deep brain stimulation of the subthalamic nucleus, characterized in that: include: A MERs signal acquisition module, a feature extraction module, a feature fusion module, a recognition module and an analysis module connected in sequence; The MERs signal acquisition module is used to acquire MERs signals from different brain regions of the patient to be tested; The feature extraction module is used to extract time series features and amplitude series features from the MERs signal respectively; The feature fusion module is used to fuse the time series feature with the amplitude series feature to obtain an amplitude-frequency perception fusion feature map of the MERs signal; The recognition module is used to input the amplitude-frequency sensing fusion feature map into the amplitude-frequency sensing fusion network, identify the boundary of STN and the optimal position for microelectrode implantation, and determine the boundary of STN and the placement position of the permanent treatment electrode; The analysis module is used to analyze the boundary of the STN and the placement position of the permanent treatment electrode to determine the optimal stimulation contact point.

Images