Epilepsy lesion localization system and method based on deep learning and electrophysiological signals

Abstract

The present invention discloses an epileptic lesion area positioning system and method based on deep learning and electrophysiological signals, including: a brain image synthesis module for acquiring brain image data, and fusing to determine the patient's brain structure abnormalities and brain metabolism abnormalities, three-dimensional reconstruction of intracranial electrode brain image data, determining the specific brain sulcus and gyrus structure of each electrode site, and generating brain image pathology results; an electroencephalogram signal processing module for acquiring intracranial electroencephalogram signals, and combining multimodal electrophysiological signals and electrical stimulation structure connection signals to classify normal electrodes and epileptogenic area electrodes to generate electrode classification results; a three-dimensional stereo imaging module is respectively connected to a brain image synthesis module and an electroencephalogram signal processing module, and is used to combine the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the electroencephalogram signal processing module, and accurately simulate a three-dimensional model of the brain structure that needs to be removed by the patient. The present invention improves the efficiency and accuracy of detecting and locating epileptic lesions.

Description

Technical Field

[0001] The present invention relates to the technical field of medical equipment, and in particular to a system and method for locating an epileptic lesion area based on deep learning and electrophysiological signals. Background Art

[0002] Epilepsy is one of the neurological diseases that affects 1% of the world's population, and focal epilepsy accounts for 50% of the total number of epilepsy patients. However, the condition of 30% of patients with focal epilepsy cannot be effectively controlled by drugs. Only with the help of surgery can the epileptogenic zone be removed, thereby achieving the purpose of reducing or curbing epileptic seizures. Stereotactic electroencephalography (SEEG), which implants electrodes into the patient's skull, can directly record the neural activity of the sulci and gyri, and thus can more accurately locate the epileptogenic zone. Clinicians make surgical plans through dozens of days of 24-hour uninterrupted SEEG monitoring (resting state, sleep state, and epileptic seizure state). However, this method requires clinicians to interpret a large number of EEGs, which is time-consuming and laborious. In addition, due to the differences in clinical experience of experts, it is difficult to quantify the accuracy of the detection and positioning results, and the stability of the evaluation results is poor. Therefore, how to effectively and accurately detect and locate the epileptogenic zone is a hot topic in current clinical research.

[0003] Epilepsy detection devices in the prior art are rarely able to reliably and accurately detect and locate the epileptogenic zone; therefore, the prior art needs to be improved and developed. Summary of the invention

[0004] The technical problem to be solved by the present invention is that, in response to the above-mentioned defects of the prior art, a system and method for locating epileptic lesions based on deep learning and electrophysiological signals is provided, aiming to solve the problem that the epilepsy detection devices in the prior art can rarely reliably and accurately detect and locate the epileptogenic zone.

[0005] The technical solution adopted by the present invention to solve the problem is as follows:

[0006] In a first aspect, an embodiment of the present invention provides an epileptic lesion location system based on deep learning and electrophysiological signals, comprising:

[0007] A brain image synthesis module is used to obtain brain image data and fuse the brain image data to determine the areas of abnormal brain structure and abnormal brain metabolism of the patient. At the same time, through three-dimensional reconstruction of intracranial electrode brain image data, the specific brain sulcus structure of each electrode site is determined to generate brain image pathology results;

[0008] An EEG signal processing module is used to obtain intracranial EEG signals, and use artificial intelligence algorithms to perform machine interpretation of the intracranial EEG signals, combine multimodal electrophysiological signals and electrical stimulation structure connection signals, classify normal electrodes and epileptogenic zone electrodes, and generate electrode classification results;

[0009] A three-dimensional stereo imaging module is connected to the brain image synthesis module and the electroencephalogram signal processing module respectively, and is used to combine the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the electroencephalogram signal processing module to simulate a three-dimensional model of the brain structure that needs to be removed by the patient.

[0010] The epileptic lesion area localization system based on deep learning and electrophysiological signals, wherein the brain image synthesis module comprises:

[0011] A magnetic resonance image acquisition unit, used for acquiring MRI brain image data through magnetic resonance imaging data processing;

[0012] A positron emission tomography data acquisition unit, used for acquiring PET brain imaging data through positron emission tomography data processing;

[0013] An electronic computer tomography data acquisition unit, used for acquiring CT brain image data through electronic computer tomography data processing;

[0014] A fusion unit, wherein the fusion unit is connected to the magnetic resonance image acquisition unit, the positron emission tomography data acquisition unit, and the electronic computer tomography data acquisition unit, respectively, and is used to fuse the PET brain image data and the MRI brain image data, fuse the imaging results of the PET brain image data and the MRI brain image data, and process the imaging results, adjust the contrast between PET and MRI, compare and locate the brain regions with abnormal brain metabolism and abnormal brain structure of the patient, mark the locations of the brain regions with abnormal brain metabolism and abnormal brain structure of the patient, and send the marked brain regions with abnormal structure or abnormal metabolism to the three-dimensional imaging module.

[0015] The epilepsy lesion area positioning system based on deep learning and electrophysiological signals, wherein the EEG signal processing module includes: an intracranial EEG feature extraction module, a cortical electrical stimulation network construction module, and a graph convolution classification module connected in sequence;

[0016] The intracranial EEG feature extraction module is used to obtain intracranial EEG signals and extract the features of intracranial EEG signals in the sleep state, resting state and attack period through a deep learning algorithm;

[0017] The cortical electrical stimulation network building module is used to create a brain structure connection model between electrodes through signal detection;

[0018] The graph convolution classification module is used to classify intracranial electrodes by using the graph convolution method, combining the attribute features extracted by the intracranial EEG feature extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module, determine normal electrodes and epileptogenic zone electrodes, and generate electrode classification results.

[0019] The epileptic lesion area positioning system based on deep learning and electrophysiological signals, wherein the intracranial EEG feature extraction module comprises:

[0020] A resting state data acquisition unit, used to acquire intracranial EEG signals of a patient in a resting state;

[0021] A sleep state data acquisition unit, used to acquire intracranial EEG signals of the patient during sleep;

[0022] The seizure data acquisition unit is used to obtain the intracranial EEG signals of the patient when the patient has epileptic symptoms;

[0023] A preprocessing unit, the preprocessing unit is connected to the resting state data acquisition unit, the sleeping state data acquisition unit, and the seizure period data acquisition unit, respectively, and is used to preprocess the three modal data of the acquired intracranial EEG signals in the resting state, the intracranial EEG signals in the sleeping period, and the intracranial EEG signals during the onset of epileptic symptoms, perform time-frequency analysis on the three modal data, and extract energy in different frequency bands of the three modal data;

[0024] A deep learning feature extraction module, wherein the deep learning feature extraction module and the preprocessing unit are used to take the energy of different frequency bands of the three modal data extracted by the preprocessing unit as input, and extract potential representations as attribute features of nodes in the graph convolution model through an encoder.

[0025] The epilepsy lesion area localization system based on deep learning and electrophysiological signals, wherein the cortical electrical stimulation network construction module includes:

[0026] A low-frequency electrical stimulation unit, used to generate a low-frequency electrical stimulation signal to stimulate the patient's intracranial electrodes to generate intracranial electrophysiological signals;

[0027] A brain network construction unit is connected to the low-frequency electrical stimulation unit and is used to stimulate the patient's intracranial electrodes to generate intracranial electrophysiological signals and construct a structural connection map between all electrodes in the whole brain.

[0028] The epileptic lesion area localization system based on deep learning and electrophysiological signals, wherein the graph convolution classification module includes:

[0029] An attribute feature conversion unit, used for converting intracranial electrophysiological signals and low-frequency electrical stimulation signals to obtain node attribute features and edge attribute features;

[0030] An adaptive graph convolution classification unit is connected to the attribute feature conversion unit and is used to classify all electrodes using an adaptive graph convolution method, determine normal electrodes and epileptogenic zone electrodes, and generate electrode classification results to distinguish between the patient's normal brain area and epileptogenic focus.

[0031] The epileptic lesion area positioning system based on deep learning and electrophysiological signals, wherein the three-dimensional stereo imaging module comprises:

[0032] An epilepsy weight assignment unit, used to integrate the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the EEG signal processing module, and assign an epilepsy weight value to each voxel in the whole brain, wherein the epilepsy weight value includes a brain structure abnormality value, a brain metabolism abnormality value, and a graph convolution epilepsy prediction value;

[0033] The three-dimensional imaging unit is used to obtain high-scoring aggregated nuclei based on the epilepsy weight scores of brain structure abnormalities, brain metabolism abnormalities, and graph convolution epilepsy prediction values, draw the regional contours of epileptic foci in the three-dimensional structure, and simulate the three-dimensional model of the brain structure that needs to be removed in the patient.

[0034] A method for locating an epileptic lesion area of ​​an epileptic lesion area locating system based on deep learning and electrophysiological signals as described in any one of the items, wherein the method comprises:

[0035] Controlling the acquisition of brain image data through the brain image synthesis module, and fusing the brain image data to determine the patient's brain structure abnormalities and brain metabolism abnormalities, and at the same time reconstructing the three-dimensional brain image data of the intracranial electrode to determine the specific brain sulcus and gyrus structure of each electrode site, and generating brain image pathology results;

[0036] Controlling the acquisition of intracranial EEG signals through the EEG signal processing module, and using an artificial intelligence algorithm to perform machine interpretation of the intracranial EEG signals, combining multimodal electrophysiological signals and electrical stimulation structure connection signals, classifying normal electrodes and epileptogenic zone electrodes, and generating electrode classification results;

[0037] The three-dimensional imaging module is controlled to simulate and output a three-dimensional model of the brain structure that needs to be removed of the patient based on the brain image pathology results generated by the brain image synthesis module and the electrode classification results generated by the electroencephalogram signal processing module.

[0038] The epilepsy lesion area positioning method of the epilepsy lesion area positioning system based on deep learning and electrophysiological signals is characterized in that the control acquires brain image data through a brain image synthesis module, and fuses the brain image data to determine the patient's brain structure abnormality and brain metabolism abnormality area, and at the same time reconstructs the intracranial electrode brain image data in three dimensions to determine the specific brain sulcus and gyrus structure of each electrode site, and the step of generating brain image pathology results includes:

[0039] Controlling the magnetic resonance image acquisition unit to acquire MRI brain image data through magnetic resonance imaging data processing;

[0040] Controlling the positron emission tomography data acquisition unit to acquire PET brain imaging data through positron emission tomography data processing;

[0041] Controlling the electronic computer tomography data acquisition unit to acquire CT brain image data through electronic computer tomography data processing;

[0042] Controlling the fusion unit to fuse the PET brain image data and the MRI brain image data, fuse the imaging results of the PET brain image data and the MRI brain image data, and process the imaging results, adjust the contrast between PET and MRI, compare and locate the brain regions with abnormal brain metabolism and abnormal brain structure of the patient; and mark the locations of the brain regions with abnormal brain metabolism and abnormal brain structure of the patient, and send the marked brain regions with abnormal structure or abnormal metabolism to the three-dimensional imaging module;

[0043] The control acquires intracranial EEG signals through an EEG signal processing module, and uses an artificial intelligence algorithm to perform machine interpretation on the intracranial EEG signals, combines multimodal electrophysiological signals and electrical stimulation structure connection signals, and classifies normal electrodes and epileptogenic zone electrodes. The steps of generating electrode classification results include:

[0044] Control the intracranial EEG feature extraction module to obtain intracranial EEG signals, and extract the features of intracranial EEG signals in the sleep state, resting state, and attack period through a deep learning algorithm;

[0045] Controlling the cortical electrical stimulation network building blocks, creating a model of brain structure connections between electrodes through signal detection;

[0046] The control graph convolution classification module uses the graph convolution method, combined with the attribute features extracted by the intracranial EEG feature extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module to classify the intracranial electrodes, determine the normal electrodes and the epileptogenic zone electrodes, and generate the electrode classification results.

[0047] The epilepsy lesion area positioning method of the epilepsy lesion area positioning system based on deep learning and electrophysiological signals, wherein the control uses the three-dimensional imaging module to simulate and output the three-dimensional model of the brain structure that needs to be removed by the patient using the brain image pathology results generated by the brain image synthesis module and the electrode classification results generated by the electroencephalogram signal processing module, including:

[0048] The epilepsy weight assignment unit is controlled to integrate the brain image pathology result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module, and assign an epilepsy weight value to each voxel of the whole brain, wherein the epilepsy weight value includes a brain structure abnormality value, a brain metabolism abnormality value, and a graph convolution epilepsy prediction value;

[0049] The three-dimensional imaging unit is controlled to obtain the high-scoring aggregated nuclei according to the epilepsy weight scores of the brain structure abnormalities, brain metabolism abnormalities, and graph convolution epilepsy prediction values, and the regional contours of the epileptic focus are drawn in the three-dimensional structure to simulate the three-dimensional model of the brain structure that needs to be removed in the patient.

[0050] Beneficial effects of the present invention: The epilepsy lesion area positioning system based on deep learning and electrophysiological signals of the present invention can draw the regional contour of the epilepsy lesion in a three-dimensional structure to accurately locate the epilepsy lesion, thereby improving the efficiency and accuracy of epilepsy lesion detection and positioning; and from the final result map, a clearer EEG result map can be obtained, thereby improving the efficiency of EEG reading.

[0051] The epilepsy lesion area localization system and method based on deep learning and electrophysiological signals in the embodiment of the present invention, combined with artificial intelligence algorithms, greatly saves the workload of doctors in interpreting electroencephalograms. At the same time, combined with artificial judgment of brain structure and brain metabolism, the results of machine judgment are integrated to improve the efficiency and accuracy of epileptogenic focus detection and localization. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 It is a functional principle block diagram of an epileptic lesion area localization system based on deep learning and electrophysiological signals provided in an embodiment of the present invention.

[0054] Figure 2It is a functional principle block diagram of a specific embodiment of an epileptic lesion area localization system based on deep learning and electrophysiological signals provided in an embodiment of the present invention.

[0055] Figure 3 It is a schematic diagram of the EEG signal processing flow of the epilepsy lesion area positioning system based on deep learning and electrophysiological signals provided in an embodiment of the present invention.

[0056] Figure 4 This is a block diagram of the internal functional principles of a brain image synthesis module of an epileptic lesion area localization system based on deep learning and electrophysiological signals provided in an embodiment of the present invention.

[0057] Figure 5 This is a flow chart of PET and MRI image fusion of an epileptic lesion area localization system based on deep learning and electrophysiological signals provided in an embodiment of the present invention.

[0058] Figure 6a It is a schematic diagram of matching a patient CT image with a standardized template in the prior art;

[0059] Figure 6b This is the invented CT and MRI image fusion diagram.

[0060] Figure 7 This is a block diagram of the internal functional principles of the intracranial EEG feature extraction module of the EEG signal processing module of the epilepsy lesion area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

[0061] Figure 8 This is a schematic diagram of the EEG signal processing flow of the epileptic lesion area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

[0062] Fig. 9 This is a schematic diagram of a deep learning network for EEG signal processing and a graph convolution classification algorithm model of an epileptic lesion area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

[0063] Fig.10 This is a block diagram of the internal functional principles of the cortical electrical stimulation network construction module of the EEG signal processing module of the epileptic lesion area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

[0064] Fig.11 This is a block diagram of the internal functional principles of the graph convolution classification module of the EEG signal processing module of the epileptic lesion area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

[0065] Fig.12a The figure is a schematic diagram of the result obtained by fusion of MRI and CT in the present invention, Figure 12b This is the final result of the three-dimensional imaging in the embodiment of the present invention.

[0066] Fig.13 This is a block diagram of the internal functional principles of a three-dimensional stereo imaging module of an epileptic lesion area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

[0067] Fig.14 Schematic diagram of an epileptic lesion area locating method of an epileptic lesion area locating system based on deep learning and electrophysiological signals in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0069] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0070] In the study of brain networks, researchers compare brain networks to attribute graphs. Attribute graphs are represented by the attributes of each node and the connection relationship between nodes. It involves many branches such as classification, clustering, link prediction, etc. In the classification task, most of the research is centered on disease prediction, such as determining whether a patient has a certain disease, or predicting the onset of a certain disease, such as whether epilepsy will occur, etc. In clustering tasks, due to the difficulty of unsupervised clustering and the difficulty of quantifying biomedical data into features, there are not many studies on this topic. Most of them use traditional K-means or spectral clustering methods after extracting features for clustering. Graph convolution is essentially to achieve convergent feature representation by aggregating the features of neighboring nodes. The adaptive graph convolution model fully considers the role of convolution in feature representation and achieves good results in completing clustering tasks by adaptively learning the number of graph convolutions. Specifically, the electrodes implanted in the patient's skull are analogized to nodes in the attribute graph, and the cortical-to-cortical connections (CCEP) between electrodes are analogized to edges between nodes in the attribute graph. Through the graph convolution method, the electrodes are aggregated and classified to distinguish normal brain areas from epilepsy-causing areas.

[0071] On the other hand, through manual interpretation of magnetic resonance imaging and positron emission tomography, the patient's brain areas with structural and metabolic abnormalities are determined, and combined with the results of convolution classification, the patient's epileptogenic areas are calibrated. This method starts from the pathology itself and combines artificial intelligence algorithms to greatly save the doctor's workload on EEG interpretation. At the same time, combined with manual judgment of brain structure and brain metabolism, the results of machine judgment are integrated to ensure the efficiency and accuracy of epileptogenic focus detection and positioning.

[0072] Existing technologies for detecting and locating epileptogenic zones based on artificial intelligence algorithms mainly start from the characteristics and detection of high-frequency signals or simply determine brain image classification through algorithms.

[0073] The technology of using artificial intelligence algorithms to detect epileptogenic foci in existing epilepsy detection devices basically starts from two points: first, to identify brain images; second, to capture specific high-frequency signals. However, the existing technology still has the following disadvantages:

[0074] First, the detection technology of the epilepsy detection device in the prior art does not fundamentally reduce the workload of clinicians for the interpretation of brain images. Clinically, when detecting and locating the epileptogenic focus of epilepsy patients, brain imaging data and neuroelectrophysiological information are usually combined for comprehensive judgment. Among them, the bottleneck of detection and positioning lies in the interpretation of neuroelectrophysiological signals. After the implantation of intracranial electrodes, epilepsy patients usually undergo one to two weeks of 24-hour uninterrupted electrophysiological testing, and each patient has 80 to 200 electrode channels implanted in the skull. Therefore, reading such a large amount of EEG data requires a lot of clinical detection and positioning resources. Therefore, the epilepsy detection device of the prior art of the present invention adopts a machine learning method to automatically interpret the EEG, which can save a lot of medical resources. As for the judgment of brain imaging results, in clinical practice, it was originally only used as an auxiliary reference. Therefore, the classification and discrimination of brain imaging data by predecessors using artificial intelligence algorithms has limited effect on truly solving clinical dilemmas;

[0075] Second, the epilepsy detection devices in the prior art cannot obtain the overall symptoms of epilepsy patients by capturing specific high-frequency signals, and ignore the interpretation of abnormal electrophysiological signals of resting and sleeping brain activities. Although high-frequency discharges are typical characteristics of epileptic seizures in most patients, they are not judged by a single component in clinical detection and positioning. The symptoms of seizures in some patients do not involve the release of high-frequency EEG signals, but are carried out in a specific rhythm that is different from normal EEG. On the other hand, since clinicians cannot comprehensively interpret all EEG records, most of the time the epileptic focus is determined by the signal changes before the onset of the attack, and a large amount of useful resting and sleeping data are not included in the evidence for detecting and locating the epileptic focus. Therefore, a technical solution that can integrate various epileptic seizure rhythm signals and include resting and sleeping EEG signals at the same time can greatly enhance the solution of real clinical dilemmas;

[0076] Third, current technology is more inclined to assist neurology in detecting and locating epileptogenic foci, and is not closely integrated with assisting neurosurgery in resection plans. In clinical practice, detection and positioning by neurology and accurate resection of epileptogenic foci by neurosurgery are equally important for patient recovery. In clinical practice, some epilepsy patients have severe deformities in brain structure due to pathological reasons, which increases the difficulty of neurosurgeons during surgery. Inexperienced surgeons often cannot accurately determine the brain structure that needs to be removed. Therefore, a system that integrates the results of neurology detection and positioning and provides virtual reality simulation is of great help to the clinical practice of neurosurgeons, which has never been mentioned in previous technical solutions.

[0077] In view of the problems existing in the prior art, the present invention combines artificial experience with machine interpretation, integrates brain images and brain electrophysiological signals, and proposes an artificial intelligence system to solve the clinical pain points of neurosurgery and neurology. The epilepsy lesion area positioning system and method based on deep learning and electrophysiological signals in the embodiment of the present invention, combined with artificial intelligence algorithms, greatly saves the workload of doctors in EEG interpretation. At the same time, combined with artificial judgment of brain structure and brain metabolism, the results of machine judgment are integrated to improve the efficiency and accuracy of epileptogenic focus detection and positioning.

[0078] Exemplary Devices

[0079] like Figure 1 As shown in , an embodiment of the present invention provides an epileptic lesion area positioning system based on deep learning and electrophysiological signals, and the epileptic lesion area positioning system based on deep learning and electrophysiological signals can be applied to intelligent medical device terminals.

[0080] In the embodiment of the present invention, Figure 1As shown, the epileptic lesion area positioning system based on deep learning and electrophysiological signals includes: a brain image synthesis module, an EEG signal processing module (wherein the EEG signal processing module includes: an intracranial EEG feature extraction module, a cortical electrical stimulation network construction module, and a graph convolution classification module), and a three-dimensional stereo imaging module; the three-dimensional stereo imaging module is respectively connected to the brain image synthesis module and the EEG signal processing module.

[0081] The brain image synthesis module is used to obtain brain image data and fuse the brain image data to determine the patient's brain structure abnormalities and brain metabolism abnormalities. At the same time, through the three-dimensional reconstruction of the intracranial electrode brain image data, the specific brain sulcus structure of each electrode site is determined to generate brain image pathology results. The brain image synthesis module includes: processing of magnetic resonance imaging data (MRI), positron emission tomography data (PET), and computer tomography data (CT).

[0082] The EEG signal processing module is used to obtain intracranial EEG signals, and use artificial intelligence algorithms to perform machine interpretation of the intracranial EEG signals, combine multimodal electrophysiological signals and electrical stimulation structure connection signals, classify normal electrodes and epileptogenic zone electrodes, and generate electrode classification results. The EEG signal processing module includes: an intracranial EEG feature extraction module, a cortical electrical stimulation network construction module, and a graph convolution classification module connected in sequence; wherein, the intracranial EEG feature extraction module is used to obtain intracranial EEG signals, and extract the features of intracranial EEG signals in sleep state, resting state, and attack period through a deep learning algorithm; the cortical electrical stimulation network construction module is used to create a brain structure connection model between electrodes through signal detection; the graph convolution classification module is used to classify intracranial electrodes using a graph convolution method, combined with the attribute features extracted by the intracranial EEG feature extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module, determine normal electrodes and epileptogenic zone electrodes, and generate electrode classification results.

[0083] The three-dimensional stereo imaging module is used to combine the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the electroencephalogram signal processing module, and can accurately simulate a three-dimensional model of the brain structure that needs to be removed by the patient for surgeons to perform preoperative evaluation and surgical simulation.

[0084] That is, the epilepsy lesion area positioning system based on deep learning and electrophysiological signals in the embodiment of the present invention adopts three major technical modules: brain image synthesis module, EEG signal processing module and three-dimensional stereo imaging module, which are as follows:

[0085] The epilepsy lesion area positioning system based on deep learning and electrophysiological signals of the present invention can be used for patients with intractable epilepsy who have intracranial electrodes implanted. Before the electrode implantation surgery, the patient usually undergoes PET and MRI brain imaging scans to determine the brain structure and brain metabolism information. Based on the patient's brain structure and brain metabolism information, combined with the patient's clinical manifestations and scalp EEG monitoring results, the neurologist will develop a detailed electrode implantation plan for each patient. After the electrode implantation surgery, the patient will undergo a full-brain CT scan to determine the spatial position of the electrode in the entire brain. Since the machines for PET, MRI and CT (computer tomography) scanning are different, there are spatial differences between the various brain imaging results. The system of the present invention will be integrated based on this difference to facilitate more accurate positioning of the patient's brain structure abnormalities and brain metabolism abnormalities, as described below.

[0086] like Figure 3 As shown, the brain image synthesis module converts the spatial positions of PET (positron emission tomography) and CT images from three-dimensional space by linear transformation to fuse the brain imaging results of PET, MRI (magnetic resonance imaging) and CT. Figure 3 As shown, the brain image synthesis module of this embodiment includes processing of magnetic resonance imaging data (MRI), positron emission tomography data (PET) and computed tomography data (CT). The brain image synthesis module clarifies the patient's brain structure abnormalities and brain metabolism abnormalities by fusing brain image data; at the same time, through three-dimensional reconstruction of intracranial electrode brain image data, the specific brain sulcus structure of each electrode site is clarified to reconstruct the intracranial electrode brain structure position.

[0087] Specifically, Figure 4 As shown, the brain image synthesis module described in this embodiment includes:

[0088] A magnetic resonance image acquisition (CT) unit, used for acquiring MRI brain image data through magnetic resonance imaging data processing;

[0089] A positron emission tomography (PET) data acquisition unit, used to acquire PET brain imaging data through positron emission tomography data processing;

[0090] An electronic computer tomography (CT) data acquisition unit, used for acquiring CT brain image data through electronic computer tomography data processing;

[0091] A fusion unit, wherein the fusion unit is connected to the magnetic resonance image acquisition unit, the positron emission tomography data acquisition unit, and the electronic computer tomography data acquisition unit, respectively, and is used to fuse the PET brain image data and the MRI brain image data, fuse the imaging results of the PET brain image data and the MRI brain image data, and process the imaging results, adjust the contrast between PET and MRI, compare and locate the brain regions with abnormal brain metabolism and abnormal brain structure of the patient; and mark the locations of the brain regions with abnormal brain metabolism and abnormal brain structure of the patient, and send the marked brain regions with abnormal structure or metabolism to the three-dimensional imaging module.

[0092] like Figure 5 As shown, Figure 5 The present invention is a schematic diagram of the image fusion process of PET brain image data and MRI brain image data. The brain image synthesis module of the embodiment of the present invention fuses the imaging results of PET (positron emission tomography data processing) and MRI (magnetic resonance imaging data processing), processes the imaging results, and adjusts the contrast of PET and MRI, so as to compare and locate the brain regions with abnormal brain metabolism and abnormal brain structure of the patient. The brain regions with abnormal brain metabolism and abnormal brain structure of the patient are marked, and the marked brain regions with abnormal structure or metabolism will be recorded and sent to the three-dimensional imaging module of the present invention, so as to provide brain image data evidence for the final construction of epileptogenic focus.

[0093] As shown in Figure 6, Figure 6b The CT and MRI image fusion diagram invented by Figure 6a The figure is a schematic diagram of the prior art of matching the patient's MRI image with the standardized template; the results show that the traditional method matches the patient's MRI image with the standardized template, and then corrects the CT image with the corresponding parameters. The consequence of this is that the patient's brain will be deformed, and the position of the corresponding CT electrode will also be bent. This approach cannot accurately locate the specific brain structure of the patient's electrode. Figure 6b As shown, the system of the present invention adopts a voxel calculation method based on individual brain structure, reconstructs the structure of the patient's brain sulci and gyri through probability calculation, and locates the electrode position at the individual level in combination with the patient's real CT image to obtain accurate results of localizing the epileptogenic focus. This is crucial for determining the source of the electrode signal in clinical practice.

[0094] Furthermore, the EEG signal processing module of the embodiment of the present invention is specifically described as follows:

[0095] The second main part of the present invention is the EEG signal processing module. Figure 7 As shown, the intracranial EEG feature extraction module of the EEG signal processing module of the embodiment of the present invention includes:

[0096] A resting state data acquisition unit, used to acquire intracranial EEG signals of a patient in a resting state;

[0097] A sleep state data acquisition unit, used to acquire intracranial EEG signals of the patient during sleep;

[0098] The seizure data acquisition unit is used to obtain the intracranial EEG signals of the patient when the patient has epileptic symptoms;

[0099] A preprocessing unit, the preprocessing unit is connected to the resting state data acquisition unit, the sleeping state data acquisition unit, and the seizure period data acquisition unit, respectively, and is used to preprocess the three modal data of the acquired intracranial EEG signals in the resting state, the intracranial EEG signals in the sleeping period, and the intracranial EEG signals during the onset of epileptic symptoms, perform time-frequency analysis on the three modal data, and extract energy in different frequency bands of the three modal data;

[0100] A deep learning feature extraction module, wherein the deep learning feature extraction module and the preprocessing unit are used to take the energy of different frequency bands of the three modal data extracted by the preprocessing unit as input, and extract potential representations as attribute features of nodes in the graph convolution model through an encoder.

[0101] The specific EEG signal processing process is as follows: Figure 8 As shown, the processing of EEG signals mainly incorporates two aspects of data: intracranial electrophysiological signals and low-frequency electrical stimulation signals; respectively, by obtaining the patient's spontaneous intracranial electrophysiological signals and obtaining the intracranial electrophysiological signals induced by low-frequency electrical stimulation of the patient's intracranial electrodes.

[0102] In an embodiment of the present invention, the obtained spontaneous electrophysiological signals of the patient include the patient's resting state data, sleep state data and seizure data. Among them, the resting state data refers to the intracranial EEG signal when the patient does not do anything and lies quietly in bed. The signal reflects the spontaneous brain connection relationship of the patient in the resting state. The sleep state data refers to the intracranial EEG signal during the patient's sleep period. Since the epileptogenic focus of epileptic patients will occasionally discharge abnormally during the patient's sleep period, extracting the sleep state data is of guiding significance for distinguishing the epileptogenic focus. The seizure data refers to the intracranial EEG signal when the patient's epileptic symptoms occur. This data is the main basis for clinicians to refer to in the process of diagnosing epilepsy epileptogenic focus.

[0103] See Fig. 9 , Fig. 9 Schematic diagram of the deep learning network for EEG signal processing and graph convolution classification algorithm model of the epileptic lesion area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention. Fig. 9As shown, the present invention pre-processes the data of three modes (intracranial EEG signals in a resting state, intracranial EEG signals during sleep, and intracranial EEG signals during epileptic symptom attacks), performs time-frequency analysis on the data of the three modes, and extracts the energy of different frequency bands of the data of the three modes (2-4Hz, 4-8Hz, 8-12Hz, 12-30Hz, 30-60Hz, 60-150Hz). The extracted energy is used as input to the deep learning feature extraction module, and the features of the original spontaneous power physiological signal are extracted by the encoder of the deep learning feature extraction module. The extracted potential representation (Xn) is used as the attribute feature of the node in the graph convolution model. The deep learning feature extraction module (Autoencoder, AE) is a deep learning neural network model that can copy input features to output. Compared with traditional principal component analysis and other methods, the introduction of the deep learning neural network algorithm compresses the coding dimension, and the potential representation obtained through training can retain most of the effective information of the original data. Then, by reconstructing and upgrading the low-dimensional latent representation so that the reconstructed output features are closer to the input features, it is believed that the low-dimensional latent representation can represent the original high-dimensional features.

[0104] The deep learning feature extraction module AE of the embodiment of the present invention is mainly composed of two parts: encoder and decoder. The autoencoder is similar to a multi-layer perceptron, using an input layer, an output layer and one or more hidden layers to connect; the encoder and decoder are defined as φ and ψ, the input feature is X, the output feature is Y, and the potential feature is expressed as F, and the mapping is as follows:

[0105]

[0106] ψ: F→Y

[0107]

[0108] When there is only one hidden layer, that is, the hidden layer is the potential feature representation, the potential representation is:

[0109]

[0110] Where σ is the activation function, such as sigmoid or relu; W is the weight matrix, b is the bias vector, W and b are iteratively updated through back-propagation, after which the decoder maps h to the output feature Y by reconstruction:

[0111] Y=σ2(W2h+b2)

[0112] The AE is trained to minimize the reconstruction error, i.e., the loss function;

[0113] L(X,Y)=||XY|| 2=||X-σ2(W2(σ1(W1X+b1)))+b2)|| 2 .

[0114] Furthermore, if Fig.10 As shown, the cortical electrical stimulation network construction module of the EEG signal processing module of the embodiment of the present invention includes:

[0115] A low-frequency electrical stimulation unit, used to generate a low-frequency electrical stimulation signal to stimulate the patient's intracranial electrodes to generate intracranial electrophysiological signals;

[0116] A brain network construction unit is connected to the low-frequency electrical stimulation unit and is used to stimulate the patient's intracranial electrodes to generate intracranial electrophysiological signals and construct a structural connection map between all electrodes in the whole brain.

[0117] Specifically, Figure 8 As shown, the cortical electrical stimulation network construction module of the EEG signal processing module of the embodiment of the present invention is used to create a brain structure connection model between electrodes through signal detection, specifically mainly obtaining intracranial electrophysiological signals induced by low-frequency electrical stimulation of the patient's intracranial electrodes, and the low-frequency electrical stimulation signal of the present invention reflects the properties of the structural connection of the brain area where the electrode is located. Direct electrical stimulation of a certain position on the cortex will induce evoked potentials at other positions on the cortex (such as Fig. 9 These cortical evoked potentials (CCEPs) can predict the results of resting-state fMRI and can test the functional and pathological brain network conclusions. CCEPs generally include early (10-50ms) N1 and late (50-500ms) slow wave N2. Fig. 9 The second picture in the second row is the original CCEP amplitude of the subject. The ordinate represents the electrode that emits the stimulus, and the abscissa represents the electrode that receives the stimulus. For example, the first row of the matrix represents the response of the cortical evoked potential CCEP made by all other electrodes to this electrical stimulation when the first pair of electrodes is stimulated. In the specific experiment of the embodiment of the present invention, each pair of electrodes located on the gray matter cortex of each subject is electrically stimulated one by one by the cortical electrical stimulation network construction module of the EEG signal processing module, and a map of the structural connections between all electrodes in the whole brain of a single subject is constructed. By using a threshold, the CCEP (cortical evoked potential) results between electrodes with significant connections are obtained. The CCEP (cortical evoked potential) connection map is used as an edge attribute feature in the graph convolution model.

[0118] In the embodiment of the present invention, Fig.11 As shown, the graph convolution classification module (also called electrode classification module) of the EEG signal processing module includes:

[0119] An attribute feature conversion unit, used for converting intracranial electrophysiological signals and low-frequency electrical stimulation signals to obtain node attribute features and edge attribute features;

[0120] An adaptive graph convolution classification unit is connected to the attribute feature conversion unit and is used to classify all electrodes using an adaptive graph convolution method, determine normal electrodes and epileptogenic zone electrodes, and generate electrode classification results to distinguish between the patient's normal brain area and epileptogenic focus.

[0121] Specifically, Fig. 9 As shown in the third row, an attribute feature conversion unit is added to the graph convolution classification module to convert intracranial electrophysiological signals and low-frequency electrical stimulation signals to obtain node attribute features and edge attribute features, wherein the node attributes are the features of sleep state, resting state, and seizure period data; the edge attribute features are the connection between cortex and cortical electrical stimulation; the adaptive graph convolution classification unit of the present invention uses an adaptive graph convolution method to classify all electrodes to distinguish between normal brain areas and epileptogenic foci of patients. The adaptive graph convolution method is based on the k-order graph convolution method, and uses an adaptive mechanism to determine the k value to ensure that the number of graph convolutions k can obtain a more appropriate value under different graph structures. In this way, under the premise of ensuring the correctness of the results, the purpose of unsupervised learning is further achieved.

[0122] Specifically, in the embodiment of the present invention, the adaptive graph convolution classification unit performs specific processing, such as Fig. 9 As shown in the figure, when the feature embedding is obtained after graph convolution, in order to divide V nodes (i.e., intracranial electrode sites) into m clusters (where clusters represent epileptogenic areas and non-epileptic areas), the k-means method is used to segment the feature embedding, where k is adaptively obtained, and adaptively judging k is a very critical link in graph convolution; in order to select a suitable k, the clustering indicator intra-cluster distance expression is used, intra(C) represents the intra-cluster distance of cluster C, x represents the node feature, i and j represent different nodes respectively, and v represents the set of node features, which is expressed as follows:

[0123]

[0124] The better clustering result should measure the distance within and between clusters when the distance within the cluster is smaller, indicating that the features within the cluster are more similar and the nodes are denser. Based on this, the distance between clusters should be larger to separate nodes from different clusters. The adaptive process of the adaptive graph convolution classification unit in the embodiment of the present invention is based on the change of the distance within the cluster. Each time the graph convolution is performed, the cluster result obtained by embedding the current order graph can be obtained and iterated continuously. If the cluster result obtained by the kth graph convolution is greater than the previous one, then the iteration is stopped. The clustering result is optimal when the last result - the distance within the cluster is the smallest, thereby obtaining the clustering order k-1.

[0125] k-order graph convolution, X represents the potential representation of each electrode extracted by the deep learning feature extraction module, and L represents the Laplacian matrix of the connection between electrodes constructed by cortical-to-cortical electrical stimulation:

[0126]

[0127] Where k is a positive integer, indicating the number of graph convolutions, and the corresponding graph filter G is:

[0128]

[0129] Among them, U is the feature vector corresponding to the potential representation, and ∧ is the feature vector arranged in ascending order.

[0130] Among them, the first-order graph convolution aggregates the neighbor features of the node, and the features in the second-order graph convolution are already the feature embeddings of the first-order aggregated neighbor features. Therefore, the neighbors of the node in the second-order convolution contain the first-order neighbor information of the node neighbors, so the second-order convolution can aggregate the second-order neighbor features of the node. Since the k-order graph convolution aggregates the k-order neighbor information when updating the node features, the long-distance data relationship should be considered when using the k-order graph convolution. Too few graph convolutions will not gather enough information, and too many graph convolutions will make the global graph node feature embeddings tend to be similar, resulting in over-smoothing. Therefore, the maximum value of k is set to 25.

[0131] Among them, K-means in the embodiment of the present invention is a relatively classic partitioning clustering algorithm. Due to its high efficiency, it is widely used in clustering large-scale data. Algorithm principle The k-means algorithm uses k as a parameter to divide n objects into k clusters, so that the similarity within the cluster is high, while the similarity between clusters is low.

[0132] Furthermore, the epileptic lesion area localization system based on deep learning and electrophysiological signals described in the embodiment of the present invention is as follows: Fig.13 As shown, the three-dimensional imaging module includes:

[0133] An epilepsy weight assignment unit, used to integrate the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the EEG signal processing module, and assign an epilepsy weight value to each voxel in the whole brain, wherein the epilepsy weight value includes a brain structure abnormality value, a brain metabolism abnormality value, and a graph convolution epilepsy prediction value;

[0134] The three-dimensional imaging unit is used to obtain high-scoring aggregated nuclei according to the epilepsy weight scores of brain structure abnormalities, brain metabolism abnormalities, and graph convolution epilepsy prediction values, and to draw the regional contours of the epileptic focus in the three-dimensional structure to simulate a three-dimensional model of the brain structure that needs to be removed by the patient.

[0135] The processing process of the three-dimensional brain imaging module in the embodiment of the present invention is as follows: the three-dimensional brain imaging module combines the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the EEG signal processing module, and can accurately simulate the three-dimensional model of the brain structure that the patient needs to remove. Specifically, the epilepsy weight assignment unit comprehensively considers the results obtained by the brain image synthesis module and the EEG information processing module, and assigns an epilepsy weight to each voxel in the whole brain. The epilepsy weight value includes three parts: brain structure abnormality value, brain metabolism abnormality value, and graph convolution epilepsy prediction value. The three-dimensional imaging unit obtains high-scoring aggregated nuclei according to the scores of brain structure abnormality value, brain metabolism abnormality value, and graph convolution epilepsy prediction value, and draws the regional outline of the epileptic epileptogenic focus in the three-dimensional structure ( Figure 12b Red area), where Fig.12a FIG. 1 is a schematic diagram of the patient's blood vessels and electrode imaging results obtained by fusion of MRI and CT according to the present invention, and Figure 12b This is the final result of the present invention.

[0136] In a preferred embodiment, the specific processing process of the three-dimensional brain imaging module of the present invention is exemplified as follows: the results obtained by the brain image synthesis module and the EEG information processing module are comprehensively considered, and the whole brain is first divided into about 109 blocks (voxels) of one cubic millimeter in size, and each block is assigned an epilepsy weight index (EI). The epilepsy weight value includes three parts: brain structural abnormality value (structural index, SI), brain metabolic abnormality value (metabolic index, MI), and graph convolution epilepsy prediction value (graph-convolution index, GI):

[0137] EI=W1*SI+W2*MI+W3*GI

[0138] Among them, W1, W2 and W3 are pre-set weight parameters, which can be assigned different weights by clinicians based on experience and clinical manifestations of patients, and pre-set in the system. SI and MI are obtained by the brain image synthesis module. In the brain image synthesis module, according to the marked brain structure abnormality area and brain metabolism abnormality area, the image is mapped to a specific voxel, and the SI / MI of the marked area voxel is 1, and the SI / MI of other areas voxels is 0. Among them, GI is obtained by the EEG information processing module. In the embodiment of the present invention, the EEG information processing module classifies the patient's intracranial electrodes into epileptogenic area electrodes and non-epileptic area electrodes through deep learning combined with a graph convolution algorithm. Each electrode has a corresponding RSA coordinate, and the star area is calibrated with the electrode coordinate position as the center and 1.5 cm as the radius. The voxel GI in the star area near the epileptogenic area electrode is 1, and the voxel GI in other areas is 0.

[0139] The scores of the three aspects are combined to obtain the high-scoring aggregate nuclei, and the regional outline of the epileptogenic focus is drawn in the three-dimensional structure (such as Figure 12b In clinical practice, clinicians can adjust the final resection area according to the patient's overt behavioral manifestations and diagnostic opinions to obtain a three-dimensional model of the final resection area.

[0140] In this way, the system of the present invention can draw the regional outline of the epileptic focus in the three-dimensional structure, and can accurately obtain the patient's epileptic focus, thereby improving the efficiency and accuracy of epileptic focus detection and positioning.

[0141] As can be seen from the above, the epilepsy lesion area positioning system based on deep learning and electrophysiological signals in the embodiment of the present invention can accurately simulate the three-dimensional model of the brain structure that needs to be removed by the patient based on the brain image pathology results generated by the brain image synthesis module and the electrode classification results generated by the EEG signal processing module, so as to provide surgeons with preoperative evaluation and surgical simulation. Combined with artificial intelligence algorithms, it greatly saves the workload of doctors in interpreting EEGs, and at the same time improves the efficiency and accuracy of epileptogenic focus detection and positioning. And the present invention also has the following advantages:

[0142] 1) The epilepsy lesion area positioning system based on deep learning and electrophysiological signals of the present invention can draw the regional contour of the epilepsy lesion in the three-dimensional structure to accurately locate the epilepsy lesion, thereby improving the efficiency and accuracy of epilepsy lesion detection and positioning, and Fig.12a The present invention uses MRI and CT to fuse the patient's blood vessels and electrodes to obtain a schematic diagram of the imaging results and Figure 12b The result graph finally obtained by the present invention can obtain a clearer EEG result graph, thereby improving the efficiency of EEG reading.

[0143] 2) Previous technologies focused on using artificial intelligence algorithms to capture high-frequency signals as the basis for locating epileptogenic areas, and the data set only contained data from the seizure period and pre-seizure period. The present invention is an epilepsy lesion area positioning system based on deep learning and electrophysiological signals, which simultaneously integrates data from the sleep period, resting state, and seizure period. Starting from the full-band EEG signal, combined with the structural connection of cortical to cortical electrical stimulation, the patient's intracranial electrodes are classified. The present invention makes more comprehensive use of the patient's intracranial signals, and the integrated information is more extensive, so the positioning results obtained are more accurate.

[0144] 3) The epilepsy lesion localization system based on deep learning and electrophysiological signals of the present invention integrates the auxiliary data required for neurological diagnosis and neurosurgery in combination with clinical needs, and provides a complete and concrete 3D modeling system and method. It provides an intuitive and specific simulation for the localization of epilepsy lesions, which helps to provide more accurate assistance in the localization of epilepsy lesions.

[0145] 4) The epilepsy lesion localization system based on deep learning and electrophysiological signals of the present invention has been proven to have a high accuracy rate in predicting the epileptogenic zone of patients through clinical practice.

[0146] Exemplary Methods

[0147] Based on the above-mentioned embodiment of the epileptic lesion area localization system based on deep learning and electrophysiological signals, Fig.14 As shown, an embodiment of the present invention further provides a method for locating an epileptic lesion area of ​​an epileptic lesion area locating system based on deep learning and electrophysiological signals as described in any one of the above 7 items, and the method comprises the following steps:

[0148] Step S100, controlling the acquisition of brain image data through the brain image synthesis module, and fusing the brain image data to determine the patient's brain structure abnormalities and brain metabolism abnormalities, and at the same time reconstructing the intracranial electrode brain image data in three dimensions to determine the specific brain sulcus and gyrus structure of each electrode site, and generating brain image pathology results, as described in the above system embodiment;

[0149] Step S200, controlling the acquisition of intracranial EEG signals through the EEG signal processing module, and using an artificial intelligence algorithm to perform machine interpretation of the intracranial EEG signals, combining multimodal electrophysiological signals and electrical stimulation structure connection signals, classifying normal electrodes and epileptogenic zone electrodes, and generating electrode classification results, as described in the above system embodiment;

[0150] Step S300, controlling the three-dimensional imaging module to simulate and output a three-dimensional model of the brain structure that needs to be removed of the patient using the brain image pathology results generated by the brain image synthesis module and the electrode classification results generated by the electroencephalogram signal processing module, as described in detail in the above system embodiment.

[0151] Furthermore, the control of step S100 acquires brain image data through a brain image synthesis module, fuses the brain image data, determines the patient's brain structure abnormalities and brain metabolism abnormalities, and simultaneously reconstructs the intracranial electrode brain image data in three dimensions to determine the specific brain sulcus and gyrus structure of each electrode site. The steps of generating brain image pathology results include:

[0152] Controlling the magnetic resonance image acquisition unit to acquire MRI brain image data through magnetic resonance imaging data processing;

[0153] Controlling the positron emission tomography data acquisition unit to process and acquire PET brain imaging data through positron emission tomography data;

[0154] Controlling the electronic computer tomography data acquisition unit to acquire CT brain image data through electronic computer tomography data processing;

[0155] Control the fusion unit to fuse the PET brain image data and the MRI brain image data, fuse the imaging results of the PET brain image data and the MRI brain image data, and process the imaging results, adjust the contrast between PET and MRI, compare and locate the brain areas with abnormal brain metabolism and abnormal brain structure of the patient; and mark the locations of the brain areas with abnormal brain metabolism and abnormal brain structure of the patient, and send the marked brain areas with abnormal structure or metabolism to the three-dimensional imaging module, as described in the above system embodiment.

[0156] The control of step S200 acquires intracranial EEG signals through an EEG signal processing module, and uses an artificial intelligence algorithm to perform machine interpretation on the intracranial EEG signals, combines multimodal electrophysiological signals and electrical stimulation structure connection signals, and classifies normal electrodes and epileptogenic zone electrodes. The steps of generating electrode classification results include:

[0157] Control the intracranial EEG feature extraction module to obtain intracranial EEG signals, and extract the features of intracranial EEG signals in the sleep state, resting state, and attack period through a deep learning algorithm;

[0158] Controlling the cortical electrical stimulation network building blocks, creating a model of brain structure connections between electrodes through signal detection;

[0159] The control graph convolution classification module uses the graph convolution method, combines the attribute features extracted by the intracranial EEG feature extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module to classify the intracranial electrodes, determine the normal electrodes and the epileptogenic zone electrodes, and generate the electrode classification results, as described in the above system embodiment.

[0160] Furthermore, the control of step S300 uses the three-dimensional imaging module to simulate and output a three-dimensional model of the brain structure that needs to be removed by the patient based on the brain image pathology results generated by the brain image synthesis module and the electrode classification results generated by the electroencephalogram signal processing module. The steps include:

[0161] The epilepsy weight assignment unit is controlled to integrate the brain image pathology result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module, and assign an epilepsy weight value to each voxel of the whole brain, wherein the epilepsy weight value includes a brain structure abnormality value, a brain metabolism abnormality value, and a graph convolution epilepsy prediction value;

[0162] Control the three-dimensional imaging unit to obtain high-scoring aggregated nuclei according to the epilepsy weight scores of brain structure abnormalities, brain metabolism abnormalities, and graph convolution epilepsy prediction values, and draw the regional contours of the epileptic focus in the three-dimensional structure to simulate the three-dimensional model of the brain structure that needs to be removed in the patient, as described in the above system embodiment.

[0163] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0164] In summary, the present invention discloses an epileptic lesion area positioning system and method based on deep learning and electrophysiological signals. The epileptic lesion area positioning system based on deep learning and electrophysiological signals of the present invention can draw the regional contour of the epileptic lesion in a three-dimensional structure to accurately locate the epileptic lesion, thereby improving the efficiency and accuracy of epileptic lesion focus detection and positioning; and from the final result map, a clearer EEG result map can be obtained, thereby improving the efficiency of EEG reading.

[0165] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. An epileptic lesion localization system based on deep learning and electrophysiological signals, characterized in that: include: A brain image synthesis module is used to obtain brain image data and fuse the brain image data to determine the areas of abnormal brain structure and abnormal brain metabolism of the patient. At the same time, through three-dimensional reconstruction of intracranial electrode brain image data, the specific brain sulcus structure of each electrode site is determined to generate brain image pathology results; An EEG signal processing module is used to obtain intracranial EEG signals, and use artificial intelligence algorithms to perform machine interpretation of the intracranial EEG signals, combine multimodal electrophysiological signals and electrical stimulation structure connection signals, classify normal electrodes and epileptogenic zone electrodes, and generate electrode classification results; A three-dimensional stereo imaging module is connected to the brain image synthesis module and the electroencephalogram signal processing module respectively, and is used to combine the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the electroencephalogram signal processing module to simulate a three-dimensional model of the brain structure that needs to be removed by the patient.

2. The epileptic lesion area localization system based on deep learning and electrophysiological signals according to claim 1, characterized in that: The brain image synthesis module comprises: A magnetic resonance image acquisition unit, used for acquiring MRI brain image data through magnetic resonance imaging data processing; A positron emission tomography data acquisition unit, used for acquiring PET brain imaging data through positron emission tomography data processing; An electronic computer tomography data acquisition unit, used for acquiring CT brain image data through electronic computer tomography data processing; A fusion unit, wherein the fusion unit is connected to the magnetic resonance image acquisition unit, the positron emission tomography data acquisition unit, and the electronic computer tomography data acquisition unit, respectively, and is used to fuse the PET brain image data and the MRI brain image data, fuse the imaging results of the PET brain image data and the MRI brain image data, and process the imaging results, adjust the contrast between PET and MRI, compare and locate the brain regions with abnormal brain metabolism and abnormal brain structure of the patient, mark the locations of the brain regions with abnormal brain metabolism and abnormal brain structure of the patient, and send the marked brain regions with abnormal structure or abnormal metabolism to the three-dimensional imaging module.

3. The epileptic lesion area localization system based on deep learning and electrophysiological signals according to claim 2, characterized in that: The EEG signal processing module includes: an intracranial EEG feature extraction module, a cortical electrical stimulation network construction module, and a graph convolution classification module which are connected in sequence; The intracranial EEG feature extraction module is used to obtain intracranial EEG signals and extract the features of intracranial EEG signals in the sleep state, resting state and attack period through a deep learning algorithm; The cortical electrical stimulation network building module is used to create a brain structure connection model between electrodes through signal detection; The graph convolution classification module is used to classify intracranial electrodes by using the graph convolution method, combining the attribute features extracted by the intracranial EEG feature extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module, determine normal electrodes and epileptogenic zone electrodes, and generate electrode classification results.

4. The epileptic lesion area localization system based on deep learning and electrophysiological signals according to claim 3, characterized in that: The intracranial EEG feature extraction module comprises: A resting state data acquisition unit, used to acquire intracranial EEG signals of a patient in a resting state; A sleep state data acquisition unit, used to acquire intracranial EEG signals of the patient during sleep; The seizure data acquisition unit is used to obtain the intracranial EEG signals of the patient when the patient has epileptic symptoms; A preprocessing unit, the preprocessing unit is connected to the resting state data acquisition unit, the sleeping state data acquisition unit, and the seizure period data acquisition unit, respectively, and is used to preprocess the three modal data of the acquired intracranial EEG signals in the resting state, the intracranial EEG signals in the sleeping period, and the intracranial EEG signals during the onset of epileptic symptoms, perform time-frequency analysis on the three modal data, and extract energy in different frequency bands of the three modal data; A deep learning feature extraction module, wherein the deep learning feature extraction module and the preprocessing unit are used to take the energy of different frequency bands of the three modal data extracted by the preprocessing unit as input, and extract potential representations as attribute features of nodes in the graph convolution model through an encoder.

5. The epileptic lesion location system based on deep learning and electrophysiological signals according to claim 3, characterized in that: The cortical electrical stimulation network building module comprises: A low-frequency electrical stimulation unit, used to generate a low-frequency electrical stimulation signal to stimulate the patient's intracranial electrodes to generate intracranial electrophysiological signals; A brain network construction unit is connected to the low-frequency electrical stimulation unit and is used to stimulate the patient's intracranial electrodes to generate intracranial electrophysiological signals and construct a structural connection map between all electrodes in the whole brain.

6. The epileptic lesion location system based on deep learning and electrophysiological signals according to claim 3, characterized in that: The graph convolution classification module includes: An attribute feature conversion unit, used for converting intracranial electrophysiological signals and low-frequency electrical stimulation signals to obtain node attribute features and edge attribute features; An adaptive graph convolution classification unit is connected to the attribute feature conversion unit and is used to classify all electrodes using an adaptive graph convolution method, determine normal electrodes and epileptogenic zone electrodes, and generate electrode classification results to distinguish between the patient's normal brain area and epileptogenic focus.

7. The epileptic lesion location system based on deep learning and electrophysiological signals according to claim 1, characterized in that: The three-dimensional imaging module comprises: An epilepsy weight assignment unit, used to integrate the brain image pathology results generated by the brain image synthesis module with the electrode classification results generated by the EEG signal processing module, and assign an epilepsy weight value to each voxel in the whole brain, wherein the epilepsy weight value includes a brain structure abnormality value, a brain metabolism abnormality value, and a graph convolution epilepsy prediction value; The three-dimensional imaging unit is used to obtain high-scoring aggregated nuclei based on the epilepsy weight scores of brain structure abnormalities, brain metabolism abnormalities, and graph convolution epilepsy prediction values, draw the regional contours of epileptic foci in the three-dimensional structure, and simulate the three-dimensional model of the brain structure that needs to be removed in the patient.

8. A method for locating an epileptic lesion area of ​​an epileptic lesion area locating system based on deep learning and electrophysiological signals according to any one of claims 1 to 7, characterized in that: The method comprises: Controlling the acquisition of brain image data through the brain image synthesis module, and fusing the brain image data to determine the patient's brain structure abnormalities and brain metabolism abnormalities, and at the same time reconstructing the three-dimensional brain image data of the intracranial electrode to determine the specific brain sulcus and gyrus structure of each electrode site, and generating brain image pathology results; Controlling the acquisition of intracranial EEG signals through the EEG signal processing module, and using an artificial intelligence algorithm to perform machine interpretation of the intracranial EEG signals, combining multimodal electrophysiological signals and electrical stimulation structure connection signals, classifying normal electrodes and epileptogenic zone electrodes, and generating electrode classification results; The three-dimensional imaging module is controlled to simulate and output a three-dimensional model of the brain structure that needs to be removed of the patient based on the brain image pathology results generated by the brain image synthesis module and the electrode classification results generated by the electroencephalogram signal processing module.

9. The epilepsy lesion area positioning method of the epilepsy lesion area positioning system based on deep learning and electrophysiological signals according to claim 8, characterized in that: The control acquires brain image data through a brain image synthesis module, fuses the brain image data, determines the patient's brain structure abnormality and brain metabolism abnormality area, and simultaneously reconstructs the intracranial electrode brain image data in three dimensions to determine the specific brain sulcus and gyrus structure of each electrode site, and generates the brain image pathology results in the following steps: Controlling the magnetic resonance image acquisition unit to acquire MRI brain image data through magnetic resonance imaging data processing; Controlling the positron emission tomography data acquisition unit to acquire PET brain imaging data through positron emission tomography data processing; Controlling the electronic computer tomography data acquisition unit to acquire CT brain image data through electronic computer tomography data processing; Controlling the fusion unit to fuse the PET brain image data and the MRI brain image data, fuse the imaging results of the PET brain image data and the MRI brain image data, and process the imaging results, adjust the contrast between PET and MRI, compare and locate the brain regions with abnormal brain metabolism and abnormal brain structure of the patient; and mark the locations of the brain regions with abnormal brain metabolism and abnormal brain structure of the patient, and send the marked brain regions with abnormal structure or abnormal metabolism to the three-dimensional imaging module; The control acquires intracranial EEG signals through an EEG signal processing module, and uses an artificial intelligence algorithm to perform machine interpretation on the intracranial EEG signals, combines multimodal electrophysiological signals and electrical stimulation structure connection signals, and classifies normal electrodes and epileptogenic zone electrodes. The steps of generating electrode classification results include: Control the intracranial EEG feature extraction module to obtain intracranial EEG signals, and extract the features of intracranial EEG signals in the sleep state, resting state, and attack period through a deep learning algorithm; Controlling the cortical electrical stimulation network building blocks, creating a model of brain structure connections between electrodes through signal detection; The control graph convolution classification module uses the graph convolution method, combined with the attribute features extracted by the intracranial EEG feature extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module to classify the intracranial electrodes, determine the normal electrodes and the epileptogenic zone electrodes, and generate the electrode classification results.

10. The epilepsy lesion area positioning method of the epilepsy lesion area positioning system based on deep learning and electrophysiological signals according to claim 8, characterized in that: The step of controlling the three-dimensional imaging module to simulate and output a three-dimensional model of the brain structure that needs to be removed by the patient by combining the brain image pathology results generated by the brain image synthesis module and the electrode classification results generated by the electroencephalogram signal processing module comprises: The epilepsy weight assignment unit is controlled to integrate the brain image pathology result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module, and assign an epilepsy weight value to each voxel of the whole brain, wherein the epilepsy weight value includes a brain structure abnormality value, a brain metabolism abnormality value, and a graph convolution epilepsy prediction value; The three-dimensional imaging unit is controlled to obtain the high-scoring aggregated nuclei according to the epilepsy weight scores of the brain structure abnormalities, brain metabolism abnormalities, and graph convolution epilepsy prediction values, and the regional contours of the epileptic focus are drawn in the three-dimensional structure to simulate the three-dimensional model of the brain structure that needs to be removed in the patient.

Images