Electroencephalogram signal super-resolution generation method and device, equipment and storage medium
By identifying virtual channels and using the DDPM model and Repaint strategy to generate super-resolution ECoG signals, the problem of insufficient ECoG signal resolution was solved, high-density signal generation was achieved, invasiveness and cost were reduced, and the resolution and accuracy of EEG signal analysis were improved.
Patent Information
- Application Number
- CN202511161990.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-19
- Publication Date
- 2025-11-21
AI Technical Summary
The insufficient spatial resolution of ECoG signals makes it difficult to deploy high-density channel electrodes, resulting in high invasiveness, high cost, and increased surgical complexity.
By acquiring low-resolution ECoG signals and preprocessing them, virtual channels in the channel layout of the target super-resolution ECoG signal are identified. Then, the mixed signal is subjected to iterative back-diffusion denoising processing using the unconditional DDPM model and the Repaint strategy to generate a super-resolution ECoG signal.
It can effectively generate high-density ECoG signals with more channels, reduce invasiveness and cost, improve the spatial resolution and accuracy of EEG signal analysis, and provide technical support for the diagnosis and treatment of brain diseases.
Smart Images
Figure CN120983049A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical artificial intelligence technology, and in particular to a method, apparatus, device, and storage medium for generating super-resolution electroencephalogram (EEG) signals. Background Technology
[0002] ECoG signals are primarily obtained by implanting a dense array of electrodes on the surface of the cerebral cortex. Currently, the spatial resolution of ECoG signals is insufficient. To achieve higher spatial resolution, more electrodes must be implanted. High-density channel electrodes are difficult to deploy, highly invasive, resulting in larger wound areas and greater surgical complexity, increasing the surgical risk and postoperative recovery burden for patients. In addition, high-channel ECoG systems are expensive, have complex wiring, and require high equipment stability and synchronization performance.
[0003] Therefore, a method, device, equipment, and storage medium for generating super-resolution EEG signals were developed to solve the above problems. Summary of the Invention
[0004] This invention proposes a method, apparatus, device, and storage medium for generating super-resolution ECoG signals to solve the problems of insufficient spatial resolution of ECoG signals and difficulty in deploying high-density channel electrodes.
[0005] The present invention achieves the above objectives through the following technical solutions: This invention provides a method for super-resolution generation of electroencephalogram (EEG) signals, comprising: Acquire low-resolution ECoG signals; Preprocessing of low-resolution ECoG signals; The target super-resolution ECoG signal channel layout is preset, and virtual channels in the target super-resolution ECoG signal channel layout are identified based on the channel layout of the low-resolution ECoG signal. Gaussian noise is filled into the time series corresponding to the virtual channels in the target super-resolution ECoG signal channel layout to obtain the noise signal; The preprocessed low-resolution ECoG signal is used as the real signal and mixed with the noise signal to obtain a mixed signal; Iterative back-diffusion denoising of the mixed signal is performed based on the unconditional DDPM model and the Repaint strategy to obtain the super-resolution ECoG signal.
[0006] Furthermore, based on the channel layout of the low-resolution ECoG signal, the identification of virtual channels in the channel layout of the target super-resolution ECoG signal includes: Align the coordinate system of the channel layout of the low-resolution ECoG signal to the coordinate system of the channel layout of the target super-resolution ECoG signal; Mark the actual ECoG signal channels in the target super-resolution ECoG signal channel layout based on the alignment results; In the target super-resolution ECoG signal channel layout, the real ECoG signal channels are removed and virtual channels are extracted.
[0007] Furthermore, based on the unconditional DDPM model and the Repaint strategy, iterative back-diffusion denoising is performed on the mixed signal to obtain the super-resolution ECoG signal, including: Initialization: This includes initializing the mixed signal, binary mask, and real signal. A value of 1 in the binary mask represents the channel corresponding to the real signal, and a value of 0 in the binary mask represents the virtual channel. Perform iterative loops within a preset time step: The mixed signal at the current time is input into the pre-trained unconditional DDPM model. The mixed signal is denoised and sampled. The signal of the virtual channel is marked as the unknown region signal based on the binary mask in the denoised sampling result. Noise corresponding to the previous iteration time step is added to the real signal to obtain the known region signal of the previous iteration time step. Based on the Repaint strategy, the signals from the unknown region and the known region are merged to obtain the output of the current iteration time step; After completing the iterative loop, the super-resolution ECoG signal is output.
[0008] Furthermore, it also includes resampling the output of the current iteration time step several times to obtain the final output of the current iteration time step. Multiple resampling includes re-noising the output of the current iteration time step, then performing denoising again, and repeating this process several times.
[0009] Furthermore, preprocessing includes filtering.
[0010] The present invention also provides a super-resolution electroencephalogram (EEG) signal generation device, comprising: The acquisition module is used to acquire low-resolution ECoG signals; The preprocessing module is used to preprocess low-resolution ECoG signals; The identification module is used to preset the target super-resolution ECoG signal channel layout and identify virtual channels in the target super-resolution ECoG signal channel layout based on the channel layout of the low-resolution ECoG signal. The filling module is used to fill the time series corresponding to the virtual channels in the target super-resolution ECoG signal channel layout with Gaussian noise to obtain a noise signal; The mixing module is used to mix the preprocessed low-resolution ECoG signal as the real signal with the noise signal to obtain a mixed signal; The denoising module is used to perform iterative back-diffusion denoising on the mixed signal based on the unconditional DDPM model and the Repaint strategy to obtain the super-resolution ECoG signal.
[0011] Furthermore, the identification module includes: The alignment module is used to align the coordinate system of the channel layout of the low-resolution ECoG signal to the coordinate system of the channel layout of the target super-resolution ECoG signal. The marking module is used to mark the channels corresponding to the actual signals in the target super-resolution ECoG signal channel layout based on the alignment results; The extraction module is used to remove the real ECoG signal channels from the target super-resolution ECoG signal channel layout and extract the virtual channels.
[0012] Furthermore, the noise reduction module includes: The initialization module is used to initialize the mixed signal, binary mask, and real signal. A value of 1 in the binary mask represents the channel corresponding to the real signal, and a value of 0 in the binary mask represents the virtual channel. The iteration loop module is used to perform iterative loops within a preset time step: The mixed signal at the current time is input into the pre-trained unconditional DDPM model. The mixed signal is denoised and sampled. The signal of the virtual channel is marked as the unknown region signal based on the binary mask in the denoised sampling result. Noise corresponding to the previous iteration time step is added to the real signal to obtain the known region signal of the previous iteration time step. Based on the Repaint strategy, the signals from the unknown region and the known region are merged to obtain the output of the current iteration time step; The output module is used to output a super-resolution ECoG signal after completing the iterative loop.
[0013] The present invention also provides a super-resolution generation device for electroencephalogram (EEG) signals, comprising: Memory, used to store computer programs; A processor is configured to implement the steps of the electroencephalogram (EEG) signal super-resolution generation method as described above when executing the computer program.
[0014] The present invention also provides a storage medium, which is a readable storage medium, and stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of the method for generating super-resolution EEG signals.
[0015] The beneficial effects of this invention are as follows: The present invention proposes a method, device, equipment and storage medium for generating super-resolution ECoG signals, which can effectively generate high-density ECoG signals with more channels and physiological significance using a small number of physical electrodes. This helps to reduce the invasiveness, surgical risks and costs of ECoG recording, while improving the spatial resolution, accuracy and reliability of ECoG signal analysis, and providing better technical support for the diagnosis and treatment of brain diseases. Attached Figure Description
[0016] Figure 1 This is a flowchart of a method for generating super-resolution EEG signals according to this application. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0018] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0019] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0020] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0021] like Figure 1 As shown, a method for generating super-resolution EEG signals includes: This invention provides a method for super-resolution generation of electroencephalogram (EEG) signals, comprising: S1: Acquire low-resolution ECoG signal; For example: 32-channel ECoG data; S2: Preprocess the low-resolution ECoG signal; S3: Preset the target super-resolution ECoG signal channel layout, and identify virtual channels in the target super-resolution ECoG signal channel layout based on the channel layout of the low-resolution ECoG signal. For example, the target super-resolution ECoG signal channel layout is a 64-channel layout; S4: Fill the time series corresponding to the virtual channels in the target super-resolution ECoG signal channel layout with Gaussian noise to obtain the noise signal; S5: The preprocessed low-resolution ECoG signal is used as the real signal and mixed with the noise signal to obtain a mixed signal; S6: Based on the unconditional DDPM model and the Repaint strategy, iterative back-diffusion denoising is performed on the mixed signal to obtain the super-resolution ECoG signal.
[0022] In one embodiment, identifying virtual channels in the channel layout of a target super-resolution ECoG signal based on the channel layout of a low-resolution ECoG signal includes: Align the coordinate system of the channel layout of the low-resolution ECoG signal to the coordinate system of the channel layout of the target super-resolution ECoG signal; Mark the actual ECoG signal channels in the target super-resolution ECoG signal channel layout based on the alignment results; In the target super-resolution ECoG signal channel layout, the real ECoG signal channels are removed and virtual channels are extracted.
[0023] The target super-resolution ECoG signal channel layout is a pre-defined, standardized, or desired dense arrangement of electrode contacts, representing the idealized "target" configuration to be achieved. Essentially, the target layout is a ground-based real-space framework that specifies the precise coordinates of all channels (real and virtual) in the final high-resolution data.
[0024] In one embodiment, iterative back-diffusion denoising processing is performed on the mixed signal based on the unconditional DDPM model and the Repaint strategy to obtain a super-resolution ECoG signal, including: Initialization: This includes initializing the mixed signal, binary mask, and real signal. A value of 1 in the binary mask represents the channel corresponding to the real signal, and a value of 0 in the binary mask represents the virtual channel. Perform iterative loops within a preset time step: The mixed signal at the current time is input into the pre-trained unconditional DDPM model. The mixed signal is denoised and sampled. The signal of the virtual channel is marked as the unknown region signal based on the binary mask in the denoised sampling result. Noise corresponding to the previous iteration time step is added to the real signal to obtain the known region signal of the previous iteration time step. Based on the Repaint strategy, the signals from the unknown region and the known region are merged to obtain the output of the current iteration time step; After completing the iterative loop, the super-resolution ECoG signal is output.
[0025] Specifically, let x(0)_known represent the known, preprocessed low-resolution ECoG signal (i.e., the physical electrode signal), and m be a binary mask where the value of the known region (physical electrode position) is 1, and the value of the unknown region (virtual channel position) is 0. The entire back diffusion process starts at time step t=T and iterates until t=1.
[0026] Initialization: Define x(T) as a pure Gaussian noise tensor with the same dimensions as the target high-resolution layout.
[0027] Iterative loop (for each time step t from T to 1): Generating the unknown region: The mixed signal x(t) at the current time step is input into the pre-trained unconditional DDPM model. The model performs one-step denoising sampling on the entire high-resolution signal according to p_theta(x(t-1)|x(t)), resulting in a fully generated, one-step denoised signal x'(t-1). We only take the unknown region part, i.e., (1-m)*x'(t-1) (where * represents element-wise multiplication).
[0028] Sampling the known region: Using the original real signal x(0)_known as the basis, according to the definition of the DDPM forward process q(x(t-1)|x(0)), noise corresponding to time t-1 is added to it to obtain the known region signal x(t-1)_known at time t-1. This step ensures that the signal evolution of the known region strictly follows the mathematical definition of DDPM, rather than being generated out of thin air by the model.
[0029] Merging signals: Merge the results of the first two steps to form the final output x(t-1) of the current iteration step: x(t-1)=m*x(t-1)_known+(1-m)*x'(t-1) This merging operation is the core of the Repaint strategy: force the replacement of the signal in the known region with the result based on the sampling of the real signal, while only retaining the generated content of the model in the unknown region.
[0030] Resampling Step: To further enhance the consistency between the generated region and the known region, x(t-1) obtained in the previous step can be resampled multiple times. That is, x(t-1) is resampled by adding noise in a small step (back to time t), then denoised again (back to time t-1), and this process is repeated several times. This helps to smooth the transition at the boundary between the known and unknown regions, making the final generated signal more harmonious overall.
[0031] Output: When the iteration loop is completed to t=0, the obtained x(0) is the final generated high-density ECoG signal that achieves super-resolution in the channel dimension.
[0032] In one embodiment, the method further includes resampling the output of the current iteration time step several times to obtain the final output of the current iteration time step. The multiple resampling includes re-adding noise to the output of the current iteration time step, performing noise reduction again, and repeating the process several times.
[0033] In one embodiment, the preprocessing includes filtering.
[0034] The present invention also provides a super-resolution electroencephalogram (EEG) signal generation device, comprising: The acquisition module is used to acquire low-resolution ECoG signals; The preprocessing module is used to preprocess low-resolution ECoG signals; The identification module is used to preset the target super-resolution ECoG signal channel layout and identify virtual channels in the target super-resolution ECoG signal channel layout based on the channel layout of the low-resolution ECoG signal. The filling module is used to fill the time series corresponding to the virtual channels in the target super-resolution ECoG signal channel layout with Gaussian noise to obtain a noise signal; The mixing module is used to mix the preprocessed low-resolution ECoG signal as the real signal with the noise signal to obtain a mixed signal; The denoising module is used to perform iterative back-diffusion denoising on the mixed signal based on the unconditional DDPM model and the Repaint strategy to obtain the super-resolution ECoG signal.
[0035] In one embodiment, the identification module includes: The alignment module is used to align the coordinate system of the channel layout of the low-resolution ECoG signal to the coordinate system of the channel layout of the target super-resolution ECoG signal. The marking module is used to mark the channels corresponding to the actual signals in the target super-resolution ECoG signal channel layout based on the alignment results; The extraction module is used to remove the real ECoG signal channels from the target super-resolution ECoG signal channel layout and extract the virtual channels.
[0036] In one embodiment, the noise reduction module includes: The initialization module is used to initialize the mixed signal, binary mask, and real signal. A value of 1 in the binary mask represents the channel corresponding to the real signal, and a value of 0 in the binary mask represents the virtual channel. The iteration loop module is used to perform iterative loops within a preset time step: The mixed signal at the current time is input into the pre-trained unconditional DDPM model. The mixed signal is denoised and sampled. The signal of the virtual channel is marked as the unknown region signal based on the binary mask in the denoised sampling result. Noise corresponding to the previous iteration time step is added to the real signal to obtain the known region signal of the previous iteration time step. In one embodiment, the unknown region signal and the known region signal are merged based on the Repaint strategy to obtain the output of the current iteration time step; The output module is used to output a super-resolution ECoG signal after completing the iterative loop.
[0037] The present invention also provides a super-resolution generation device for electroencephalogram (EEG) signals, comprising: Memory, used to store computer programs; A processor is configured to implement the steps of the electroencephalogram (EEG) signal super-resolution generation method as described above when executing the computer program.
[0038] The present invention also provides a storage medium, which is a readable storage medium, and stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of the method for generating super-resolution EEG signals.
[0039] The present invention proposes a method, device, equipment and storage medium for generating super-resolution ECoG signals, which can effectively generate high-density ECoG signals with more channels and physiological significance using a small number of physical electrodes. This helps to reduce the invasiveness, surgical risks and costs of ECoG recording, while improving the spatial resolution, accuracy and reliability of ECoG signal analysis, and providing better technical support for the diagnosis and treatment of brain diseases.
[0040] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for super-resolution generation of electroencephalogram (EEG) signals, characterized in that, include: Acquire low-resolution ECoG signals; Preprocessing of low-resolution ECoG signals; The target super-resolution ECoG signal channel layout is preset, and virtual channels in the target super-resolution ECoG signal channel layout are identified based on the channel layout of the low-resolution ECoG signal. Gaussian noise is filled into the time series corresponding to the virtual channels in the target super-resolution ECoG signal channel layout to obtain the noise signal; The preprocessed low-resolution ECoG signal is used as the real signal and mixed with the noise signal to obtain a mixed signal; Iterative back-diffusion denoising of the mixed signal is performed based on the unconditional DDPM model and the Repaint strategy to obtain the super-resolution ECoG signal.
2. The method for super-resolution generation of electroencephalogram (EEG) signals according to claim 1, characterized in that, Identifying virtual channels in the channel layout of a target super-resolution ECoG signal based on the channel layout of a low-resolution ECoG signal, including: Align the coordinate system of the channel layout of the low-resolution ECoG signal to the coordinate system of the channel layout of the target super-resolution ECoG signal; Mark the actual ECoG signal channels in the target super-resolution ECoG signal channel layout based on the alignment results; In the target super-resolution ECoG signal channel layout, the real ECoG signal channels are removed and virtual channels are extracted.
3. The method for super-resolution generation of electroencephalogram (EEG) signals according to claim 1, characterized in that, Iterative back-diffusion denoising of the mixed signal is performed based on the unconditional DDPM model and the Repaint strategy to obtain the super-resolution ECoG signal, including: Initialization: This includes initializing the mixed signal, binary mask, and real signal. A value of 1 in the binary mask represents the channel corresponding to the real signal, and a value of 0 in the binary mask represents the virtual channel. Perform iterative loops within a preset time step: The mixed signal at the current time is input into the pre-trained unconditional DDPM model. The mixed signal is denoised and sampled. The signal of the virtual channel is marked as the unknown region signal based on the binary mask in the denoised sampling result. Noise corresponding to the previous iteration time step is added to the real signal to obtain the known region signal of the previous iteration time step. Based on the Repaint strategy, the signals from the unknown region and the known region are merged to obtain the output of the current iteration time step; After completing the iterative loop, the super-resolution ECoG signal is output.
4. The method for super-resolution generation of electroencephalogram (EEG) signals according to claim 3, characterized in that, It also includes resampling the output of the current iteration time step several times to obtain the final output of the current iteration time step. Multiple resampling includes re-adding noise to the output of the current iteration time step, then performing noise reduction again, and repeating this process several times.
5. The method for super-resolution generation of electroencephalogram (EEG) signals according to claim 1, characterized in that, Preprocessing includes filtering.
6. A device for generating super-resolution electroencephalogram (EEG) signals, characterized in that, include: The acquisition module is used to acquire low-resolution ECoG signals; The preprocessing module is used to preprocess low-resolution ECoG signals; The identification module is used to preset the target super-resolution ECoG signal channel layout and identify virtual channels in the target super-resolution ECoG signal channel layout based on the channel layout of the low-resolution ECoG signal. The filling module is used to fill the time series corresponding to the virtual channels in the target super-resolution ECoG signal channel layout with Gaussian noise to obtain a noise signal; The mixing module is used to mix the preprocessed low-resolution ECoG signal as the real signal with the noise signal to obtain a mixed signal; The denoising module is used to perform iterative back-diffusion denoising on the mixed signal based on the unconditional DDPM model and the Repaint strategy to obtain the super-resolution ECoG signal.
7. The electroencephalogram (EEG) signal super-resolution generation device according to claim 6, characterized in that, The recognition module includes: The alignment module is used to align the coordinate system of the channel layout of the low-resolution ECoG signal to the coordinate system of the channel layout of the target super-resolution ECoG signal. The marking module is used to mark the channels corresponding to the actual signals in the target super-resolution ECoG signal channel layout based on the alignment results; The extraction module is used to remove the real ECoG signal channels from the target super-resolution ECoG signal channel layout and extract the virtual channels.
8. The method for super-resolution generation of electroencephalogram (EEG) signals according to claim 6, characterized in that, The noise reduction module includes: The initialization module is used to initialize the mixed signal, binary mask, and real signal. A value of 1 in the binary mask represents the channel corresponding to the real signal, and a value of 0 in the binary mask represents the virtual channel. The iteration loop module is used to perform iterative loops within a preset time step: The mixed signal at the current time is input into the pre-trained unconditional DDPM model. The mixed signal is denoised and sampled. The signal of the virtual channel is marked as the unknown region signal based on the binary mask in the denoised sampling result. Noise corresponding to the previous iteration time step is added to the real signal to obtain the known region signal of the previous iteration time step. Based on the Repaint strategy, the signals from the unknown region and the known region are merged to obtain the output of the current iteration time step; The output module is used to output a super-resolution ECoG signal after completing the iterative loop.
9. A device for generating super-resolution electroencephalogram (EEG) signals, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the electroencephalogram (EEG) signal super-resolution generation method as described in any one of claims 1 to 5.
10. A storage medium, characterized in that, The storage medium is a readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the electroencephalogram signal super-resolution generation method as described in any one of claims 1 to 5.