CN115644892A

CN115644892A - Epileptic focal zone positioning system and method based on deep learning and electrophysiological signals

Info

Publication number: CN115644892A
Application number: CN202211410828.5A
Authority: CN
Inventors: 夏菁; 詹阳
Original assignee: Shenzhen Institute of Advanced Technology of CAS
Current assignee: Shenzhen Institute of Advanced Technology of CAS
Priority date: 2022-11-11
Filing date: 2022-11-11
Publication date: 2023-01-31
Also published as: WO2024098449A1

Abstract

The invention discloses an epileptic focal zone positioning system and method based on deep learning and electrophysiological signals, which comprises the following steps: the brain image synthesis module is used for acquiring brain image data, fusing the brain image data to determine the brain structure abnormality and brain metabolism abnormality areas of the patient, carrying out three-dimensional reconstruction on the brain image data of the intracranial electrodes, determining the specific sulcus structure of each electrode site, and generating a brain image pathological result; the electroencephalogram signal processing module is used for acquiring intracranial electroencephalogram signals, combining the multi-mode electrophysiological signals and the electrical stimulation structure connection signals, and classifying the normal electrodes and the electrodes in the epileptogenic regions to generate electrode classification results; the three-dimensional imaging module is respectively connected with the brain image synthesis module and the electroencephalogram signal processing module and is used for combining the brain image pathological result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module to accurately simulate a three-dimensional model of a brain structure of a patient needing to be cut. The invention improves the efficiency and accuracy of detecting and positioning epileptogenic focus of epilepsy.

Description

Epileptic focal zone positioning system and method based on deep learning and electrophysiological signals

Technical Field

The invention relates to the technical field of medical equipment, in particular to an epileptic focal zone positioning system and method based on deep learning and electrophysiological signals.

Background

Epilepsy is one of the neurological diseases affecting 1% of people worldwide, with focal epilepsy accounting for 50% of the total number of patients with epilepsy. However, the disease condition of 30% of patients with focal epilepsy cannot be effectively controlled by using medicines, and the epilepsy-causing area can be removed only with the help of surgical operation, so that the aim of reducing or suppressing the epileptic seizure is fulfilled. The stereotactic electroencephalogram (stereo-EEG) is used for implanting the electrode into the cranium of a patient, so that the nerve activity of the sulcus can be directly recorded, and the epileptic region caused by the epilepsy can be more accurately positioned. The clinician makes the surgical plan by 24 hours uninterrupted SEEG monitoring (resting state, sleeping state, epileptic seizure state) for tens of days. However, this method requires a large number of electroencephalograms to be interpreted by clinicians, which is time-consuming and labor-consuming, and the differences in clinical experience of experts make it difficult to quantify the accuracy of the positioning results, and the stability of the evaluation results is poor. Therefore, how to effectively and accurately detect and locate epileptogenic epilepsy is a hot spot in clinical research at present.

In the epilepsy detecting device of the prior art, the positioning epilepsy-causing area can be detected reliably and accurately, so that the prior art needs to be improved and developed.

Disclosure of Invention

The present invention is to provide a system and a method for locating an epileptic focal zone based on deep learning and electrophysiological signals, which are intended to solve the problem that in the epilepsy detection apparatus of the prior art, it is very rare to reliably and accurately detect and locate an epileptic focal zone.

The technical scheme adopted by the invention for solving the problems is as follows:

in a first aspect, an embodiment of the present invention provides an epileptic focus positioning system based on deep learning and electrophysiological signals, including:

the brain image synthesis module is used for acquiring brain image data, fusing the brain image data, determining brain structure abnormality and brain metabolism abnormality areas of a patient, and determining a specific sulcus loop structure of each electrode site through three-dimensional reconstruction of intracranial electrode brain image data to generate a brain image pathological result;

the electroencephalogram signal processing module is used for acquiring intracranial electroencephalogram signals, performing machine interpretation on the intracranial electroencephalogram signals by using an artificial intelligence algorithm, combining multi-mode electrophysiological signals and electrical stimulation structure connection signals, classifying normal electrodes and electrodes in an epileptogenic area to generate electrode classification results;

and the three-dimensional imaging module is respectively connected with the brain image synthesis module and the electroencephalogram signal processing module and is used for combining the brain image pathological result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module to simulate a three-dimensional model of a brain structure of a patient needing to be excised.

The epilepsy focus area positioning system based on deep learning and electrophysiological signals, wherein the brain image synthesis module comprises:

the magnetic resonance image acquisition unit is used for acquiring MRI brain image data through magnetic resonance imaging data processing;

a positron emission tomography data acquisition unit for acquiring PET brain image data by positron emission tomography data processing;

an electronic computer tomography data acquisition unit for acquiring CT brain image data by electronic computer tomography data processing;

and the fusion unit is respectively connected with the magnetic resonance image acquisition unit, the positron emission tomography data acquisition unit and the electronic computer tomography data acquisition unit, and is used for fusing the PET brain image data and the MRI brain image data, fusing imaging results of the PET brain image data and the MRI brain image data, processing the imaging results, adjusting the contrast ratio of PET and MRI, comparing and positioning the brain metabolic abnormality of the patient and the brain region position with brain structural abnormality, marking the brain region position with the brain metabolic abnormality of the patient and sending the marked brain region with structural abnormality or metabolic abnormality to the three-dimensional imaging module.

The epilepsy focus area positioning system based on deep learning and electrophysiological signals, wherein the electroencephalogram signal processing module comprises: the intracranial electroencephalogram feature extraction module, the cortical electrical stimulation network construction module and the picture volume integral module are sequentially connected;

the intracranial electroencephalogram feature extraction module is used for acquiring intracranial electroencephalogram signals and extracting features of the intracranial electroencephalogram signals in a sleep state, a resting state and an attack period through a deep learning algorithm;

the cortical electrical stimulation network construction module is used for establishing a brain structure connection model between the electrodes through signal detection;

the graph convolution classification module is used for classifying the intracranial electrodes by utilizing a graph convolution method and combining the attribute characteristics extracted by the intracranial electroencephalogram characteristic extraction module and the structural connection graph constructed by the cortical electrical stimulation network construction module, judging normal electrodes and electrodes in an epileptic region, and generating an electrode classification result.

The epilepsy focus area positioning system based on deep learning and electrophysiological signals, wherein the intracranial electroencephalogram feature extraction module comprises:

the resting state data acquisition unit is used for acquiring intracranial electroencephalogram signals of a patient in a resting state;

the sleep state data acquisition unit is used for acquiring intracranial electroencephalogram signals of a patient in a sleep period;

the attack period data acquisition unit is used for acquiring intracranial electroencephalogram signals when epileptic symptoms of a patient attack;

the preprocessing unit is respectively connected with the resting state data acquisition unit, the sleeping state data acquisition unit and the attack period data acquisition unit and is used for preprocessing three modal data of the obtained intracranial electroencephalogram in a resting state, the intracranial electroencephalogram in a sleeping period and the intracranial electroencephalogram in an epileptic symptom attack, carrying out time-frequency analysis on the three modal data and extracting energy of different frequency bands of the three modal data;

and the deep learning feature extraction module and the preprocessing unit are used for taking the energy of the three modal data extracted by the preprocessing unit in different frequency bands as input and extracting potential representations as attribute features of nodes in the graph volume model through an encoder.

The epilepsy focus area positioning system based on deep learning and electrophysiological signals, wherein the cortical electrical stimulation network building module comprises:

the low-frequency electrical stimulation unit is used for generating a low-frequency electrical stimulation signal and stimulating an intracranial electrode of a patient to generate an intracranial electrophysiological signal;

the brain network construction unit is connected with the low-frequency electrical stimulation unit and used for generating intracranial electrophysiological signals for stimulating the intracranial electrodes of the patient and constructing a structural connection map among all the electrodes of the whole brain.

The epileptic focus positioning system based on deep learning and electrophysiological signals, wherein the volume integral module comprises:

the attribute characteristic conversion unit is used for converting the intracranial electrophysiological signals and the low-frequency electrical stimulation signals to obtain node attribute characteristics and edge attribute characteristics;

and the self-adaptive volume integral type unit is connected with the attribute characteristic conversion unit and is used for classifying all the electrodes by adopting a self-adaptive volume convolution method, judging the normal electrodes and the seizure induction area electrodes and generating an electrode classification result so as to distinguish the normal brain area and the seizure induction focus of the patient.

The epileptogenic focus positioning system based on deep learning and electrophysiological signals, wherein the three-dimensional stereo imaging module comprises:

the epilepsy weight assignment unit is used for integrating the brain image pathological result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module, and endowing an epilepsy weight value to each voxel of the whole brain, wherein the epilepsy weight value comprises a brain structure abnormal value, a brain metabolism abnormal value and a volume epilepsy predicted value;

and the three-dimensional imaging unit is used for obtaining the high-grade polymeric nucleus according to the epilepsia weight score of the brain structure abnormal value, the brain metabolism abnormal value and the chart volume epilepsia predicted value, drawing the regional outline of an epileptogenic focus of the epilepsia in the three-dimensional structure and simulating a three-dimensional model of the brain structure of the patient, which needs to be cut off.

A method of epileptic focal zone localization of an epileptic focal zone localization system based on deep learning and electrophysiological signals as claimed in any preceding claim, wherein the method comprises:

controlling to obtain brain image data through a brain image synthesis module, fusing the brain image data, determining the brain structure abnormality and brain metabolism abnormality areas of a patient, simultaneously performing three-dimensional reconstruction on the brain image data of the intracranial electrodes, determining the specific sulcus return structure of each electrode site, and generating a brain image pathological result;

the method comprises the steps that an electroencephalogram signal processing module is controlled to obtain intracranial electroencephalogram signals, machine interpretation is conducted on the intracranial electroencephalogram signals through an artificial intelligence algorithm, and classification of normal electrodes and electrodes of epileptic regions is conducted by combining multi-mode electrophysiological signals and electrical stimulation structure connection signals to generate electrode classification results;

and controlling to simulate and output a three-dimensional stereo model of the brain structure of the patient needing to be cut off by the three-dimensional stereo imaging module according to the brain image pathological result generated by the brain image synthesis module and the electrode classification result generated by the electroencephalogram signal processing module.

The epileptogenic focus positioning method of the epileptogenic focus positioning system based on the deep learning and the electrophysiological signals is characterized in that the control obtains brain image data through a brain image synthesis module, fuses the brain image data, determines the brain structure abnormality and brain metabolism abnormality areas of a patient, simultaneously performs three-dimensional reconstruction on the brain image data of an intracranial electrode, determines the specific sulcus structure of each electrode site, and generates a brain image pathological result, wherein the step of the control comprises the following steps:

controlling a magnetic resonance image acquisition unit to acquire MRI brain image data through magnetic resonance imaging data processing;

controlling a positron emission tomography data acquisition unit to acquire PET brain image data through positron emission tomography data processing;

controlling an electronic computer tomography data acquisition unit to acquire CT brain image data through electronic computer tomography data processing;

the control fusion unit is used for fusing the PET brain image data and the MRI brain image data, fusing imaging results of the PET brain image data and the MRI brain image data, processing the imaging results, adjusting the contrast ratio of PET and MRI, and comparing and positioning brain area positions of brain metabolic abnormality and brain structural abnormality of a patient; marking the positions of the brain areas for positioning the cerebral metabolic abnormality and the cerebral structural abnormality of the patient, and sending the marked cerebral areas with structural abnormality or metabolic abnormality to the three-dimensional imaging module;

the control method comprises the following steps of acquiring an intracranial electroencephalogram signal through an electroencephalogram signal processing module, performing machine interpretation on the intracranial electroencephalogram signal by using an artificial intelligence algorithm, combining a multi-modal electrophysiological signal and an electrical stimulation structure connection signal, classifying a normal electrode and an epilepsia-induced seizure area electrode, and generating an electrode classification result, wherein the steps comprise:

controlling an intracranial electroencephalogram feature extraction module to obtain an intracranial electroencephalogram signal, and extracting features of the intracranial electroencephalogram signal in a sleep state, a resting state and an attack period through a deep learning algorithm;

the control cortex electrical stimulation network construction module is used for creating a brain structure connection model between the electrodes through signal detection;

the control chart convolution classification module classifies the intracranial electrodes by using a chart convolution method and combining the attribute features extracted by the intracranial electroencephalogram feature extraction module and the structural connection map constructed by the cortex electrical stimulation network construction module, judges normal electrodes and electrodes in an epileptic zone, and generates an electrode classification result.

The epileptogenic focus positioning method of the epileptogenic focus positioning system based on the deep learning and the electrophysiological signals, wherein the step of simulating and outputting a three-dimensional stereo model of a brain structure of a patient needing to be excised by the control through a three-dimensional stereo imaging module and a brain image pathological result generated by the brain image synthesis module and an electrode classification result generated by the brain electrical signal processing module comprises the following steps:

controlling an epilepsy weight assignment unit to combine the brain image pathological result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module to give an epilepsy weight value to each voxel of the whole brain, wherein the epilepsy weight value comprises a brain structure abnormal value, a brain metabolism abnormal value and a volume epilepsy predicted value;

and controlling a three-dimensional imaging unit, obtaining a high-grade polymeric nucleus according to the epilepsy weight score of the brain structure abnormal value, the brain metabolism abnormal value and the chart volume epilepsy prediction value, and drawing an area outline of an epilepsy-causing focus in a three-dimensional structure so as to simulate a brain structure three-dimensional model which needs to be cut by a patient.

The invention has the beneficial effects that: the epileptogenic focus positioning system based on deep learning and electrophysiological signals can draw the regional contour of the epileptogenic focus in the three-dimensional structure so as to accurately position the epileptogenic focus and improve the detection positioning efficiency and accuracy of the epileptogenic focus; and from the final result picture, can get more clear electroencephalogram result picture, has improved the efficiency that the electroencephalogram reads.

The epileptic focus positioning system and method based on deep learning and electrophysiological signals in the embodiment of the invention are combined with an artificial intelligence algorithm, so that the workload of doctors in interpretation of electroencephalograms is greatly saved. Meanwhile, the brain structure and brain metabolism are judged by combining with the manual work, the result of machine judgment is fused, and the efficiency and the accuracy of detecting and positioning the epileptogenic focus of the epilepsy are improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a functional block diagram of an epileptic focal zone localization system based on deep learning and electrophysiological signals provided by an embodiment of the present invention.

Fig. 2 is a functional block diagram of an epilepsy focal zone positioning system based on deep learning and electrophysiological signals according to an embodiment of the present invention.

Fig. 3 is a schematic diagram of an electroencephalogram signal processing flow of the epileptic focus localization system based on deep learning and electrophysiological signals provided by the embodiment of the present invention.

Fig. 4 is a functional block diagram of the interior of a brain image synthesis module of an epileptic focus positioning system based on deep learning and electrophysiological signals according to an embodiment of the present invention.

Fig. 5 is a PET and MRI image fusion flow chart of the epileptic focal zone positioning system based on deep learning and electrophysiological signals provided by the embodiment of the present invention.

FIG. 6a is a diagram illustrating the matching of a prior art CT image of a patient with a standardized template;

FIG. 6b is the image fusion image of CT and MRI.

Fig. 7 is an internal functional schematic block diagram of an intracranial electroencephalogram feature extraction module of an electroencephalogram signal processing module of an epileptic focus area localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

Fig. 8 is a schematic diagram of an electroencephalogram signal processing flow of an epileptic focus localization system based on deep learning and electrophysiological signals according to an embodiment of the present invention.

Fig. 9 is a schematic diagram of an electroencephalogram signal processing deep learning network and a graph integration algorithm model of an epileptic focus area positioning system based on deep learning and electrophysiological signals according to an embodiment of the present invention.

Fig. 10 is a functional schematic block diagram of the inside of a cortical electrical stimulation network construction module of an electroencephalogram signal processing module of an epileptic focus localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

Fig. 11 is a functional block diagram of the interior of a graph integration class module of an electroencephalogram signal processing module of an epileptic focus localization system based on deep learning and electrophysiological signals in an embodiment of the present invention.

FIG. 12a is a schematic diagram showing the result of the fusion of MRI and CT according to the present invention, and FIG. 12b is a final result of three-dimensional stereo imaging according to the present invention.

Fig. 13 is a functional block diagram of the interior of a three-dimensional imaging module of an epileptic focus positioning system based on deep learning and electrophysiological signals in an embodiment of the present invention.

Fig. 14 is a schematic diagram of an epileptic focus positioning method of the epileptic focus positioning system based on deep learning and electrophysiological signals according to the embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer and clearer, 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 merely illustrative of the invention and are not intended to limit the invention.

It should be noted that, if directional indications (such as up, down, left, right, front, and back … …) are involved in the embodiment of the present invention, the directional indications are only used to explain the relative position relationship between the components, the motion situation, and the like in a specific posture (as shown in the drawing), and if the specific posture is changed, the directional indications are changed accordingly.

In the study of brain networks, researchers have analogized brain networks into attribute maps. The attribute graph is represented by the attribute of each node and the connection relationship between the nodes. The classification task is mostly developed around disease prediction, such as judging whether a patient has a certain disease or not, or predicting a certain disease attack, such as epileptic attack or not. In the clustering task, due to the reasons that unsupervised clustering is difficult, biomedical data is difficult to quantify into features and the like, the content is not researched much, and the clustering is performed by adopting the traditional K-means or spectral clustering and other methods after the features are extracted. Graph convolution essentially achieves a converged feature representation by aggregating neighboring node features. The self-adaptive graph convolution model fully considers the effect of convolution on feature representation, and self-adaptive learning is carried out on the graph volume number, so that a network obtains a good result when finishing a clustering task. Specifically, the electrodes implanted into the patient's cranium are analogized to nodes of the attribute map, and the cortical-to-Cortical Connection (CCEP) between the electrodes is analogized to edges between the nodes of the attribute map. And realizing the polymerization classification of the electrodes by a graph convolution method so as to distinguish a normal brain area and an epileptogenic area.

On the other hand, the brain regions with abnormal structure and metabolism of the patient are determined through manual interpretation of magnetic resonance and positron emission tomography imaging, and the epileptogenic regions of the patient are calibrated by combining the integration results of the graph. The method starts from the pathology and combines with an artificial intelligence algorithm, thereby greatly saving the workload of doctors for interpreting electroencephalograms. Meanwhile, the brain structure and brain metabolism are judged artificially, the result of machine judgment is fused, and the efficiency and the accuracy of detecting and positioning the epileptogenic focus of the epilepsy are ensured.

The existing technology for detecting and positioning an epileptogenic area of epilepsy based on an artificial intelligence algorithm starts from the characteristics and detection of high-frequency signals or determines the classification of brain images simply through the algorithm.

The technology for detecting the epileptogenic focus of the epilepsy by using the artificial intelligence algorithm in the epilepsy detection device in the prior art basically starts from two points: firstly, distinguishing a brain image; second, a specific high frequency signal is captured. However, the prior art still has the following disadvantages:

first, the prior art epilepsia detection device detection techniques do not substantially reduce the workload of clinicians for interpretation of brain images. Clinically, when detecting and positioning an epileptogenic focus of an epileptic patient, the comprehensive judgment is usually carried out by combining brain image data and neuroelectrophysiological information. The bottleneck in detection and localization lies in the interpretation of the neuroelectrophysiological signals. Epileptic patients typically undergo 24-hour, uninterrupted electrophysiological testing for one to two weeks after implantation of intracranial electrodes, with each patient having an intracranial implant of 80 to 200 unequal electrode channels. Therefore, reading such huge electroencephalogram data consumes a lot of resources for clinical examination positioning. Therefore, the epilepsy detection device in the prior art adopts a machine learning method to automatically interpret the electroencephalogram, so that a large amount of medical resources can be saved. In clinical practice, the judgment of the brain image result is only used as an auxiliary reference originally, so that the classification and judgment of the brain image data by the predecessors by using an artificial intelligence algorithm has a limited effect on really solving the clinical dilemma;

second, the prior art epilepsy detecting device cannot obtain general seizure symptoms of epileptic patients by capturing specific high frequency signals, and ignores interpretation of abnormal electrophysiological signals of resting and sleeping brain activities. Although high frequency discharges are a typical feature of seizures in most patients, in clinical settings, the decision is not made as a single component. Some patient-related symptoms do not involve the delivery of high-frequency brain electrical signals, but rather occur in a specific rhythm that is distinct from the normal brain electrical signal. On the other hand, since clinicians cannot fully interpret all electroencephalogram recordings, most of the time, epileptogenic foci are determined by signal changes in the early stage of onset, and a large amount of useful resting state and sleeping state data are not included in evidence for detecting and locating epileptogenic foci. Therefore, the technical scheme that various epileptic seizure rhythm signals can be integrated, and resting state and sleeping state electroencephalogram signals can be included, so that the method has a great improvement effect on solving the real clinical dilemma;

third, current technology is much more inclined to aid neurology in the detection and localization of epileptogenic foci and is not tightly coupled with neurosurgical-assisted ablation protocols. In clinical practice, detection and positioning in neurology and accurate excision of neurosurgery cause epileptic foci are also important for patients after recovery. In clinical practice, some epileptics have severe malformation of brain structures due to pathological causes, which raises the difficulty in the operation process of neurosurgeons, and surgeons with low experience often cannot accurately judge the brain structures to be excised. Therefore, a set of system integrating the neurology detection positioning results and providing virtual reality simulation is greatly helpful for the clinical practice of neurosurgeons, which is never mentioned in the prior art.

Aiming at the problems in the prior art, the invention combines artificial experience and machine interpretation, synthesizes brain images and electroencephalogram physiological signals, and provides a set of artificial intelligence system for solving clinical pain points in neurosurgery and neurology. Meanwhile, the brain structure and brain metabolism are judged by combining with the manual work, the result of machine judgment is fused, and the efficiency and the accuracy of detecting and positioning the epileptogenic focus of the epilepsy are improved.

Exemplary devices

As shown in fig. 1, an embodiment of the present invention provides an epileptic focus positioning system based on deep learning and electrophysiological signals, which may be applied to an intelligent medical device terminal.

In an embodiment of the present invention, as shown in fig. 1, the epilepsy focus area positioning system based on deep learning and electrophysiological signals includes: the brain image synthesis module, the electroencephalogram signal processing module (wherein, the electroencephalogram signal processing module comprises an intracranial electroencephalogram feature extraction module, a cortex electrical stimulation network construction module, a volume integral module) and a three-dimensional imaging module; the three-dimensional imaging module is respectively connected with the brain image synthesis module and the electroencephalogram signal processing module.

The brain image synthesis module is used for acquiring brain image data, fusing the brain image data, determining the brain structure abnormality and brain metabolism abnormality areas of the patient, and meanwhile determining the specific sulcus structure of each electrode site through three-dimensional reconstruction of the brain image data of the intracranial electrodes to generate a brain image pathological result. The brain image synthesis module comprises: for Magnetic Resonance Imaging data processing (MRI), positron Emission Tomography data Processing (PET), and Computed Tomography data processing (CT).

The electroencephalogram signal processing module is used for acquiring intracranial electroencephalogram signals, performing machine interpretation on the intracranial electroencephalogram signals by using an artificial intelligence algorithm, combining multi-mode electrophysiological signals and electrical stimulation structure connection signals, classifying normal electrodes and electrodes in epileptogenic regions with epilepsy, and generating electrode classification results. The electroencephalogram signal processing module comprises: the intracranial electroencephalogram feature extraction module, the cortical electrical stimulation network construction module and the picture volume integral module are sequentially connected; the intracranial electroencephalogram feature extraction module is used for acquiring intracranial electroencephalogram signals and extracting features of the intracranial electroencephalogram signals in a sleep state, a resting state and an attack period through a deep learning algorithm; the cortical electrical stimulation network construction module is used for establishing a brain structure connection model between the electrodes through signal detection; the graph convolution classification module is used for classifying the intracranial electrodes by utilizing a graph convolution method and combining the attribute characteristics extracted by the intracranial electroencephalogram characteristic extraction module and the structural connection graph constructed by the cortical electrical stimulation network construction module, judging normal electrodes and electrodes in an epileptic region, and generating an electrode classification result.

The three-dimensional imaging module is used for combining the brain image pathological result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module, so that a three-dimensional model of a brain structure of a patient needing to be cut can be accurately simulated, and a surgeon can perform preoperative evaluation and operation simulation.

Namely, the epilepsy focus area positioning system based on deep learning and electrophysiological signals adopts three technical modules: the brain image synthesis module, the electroencephalogram signal processing module and the three-dimensional imaging module are as follows:

the epilepsy focus area positioning system based on deep learning and electrophysiological signals can be a refractory epilepsy patient implanted with an intracranial electrode, and the patient usually performs PET and MRI brain imaging scanning to determine brain structure and brain metabolic information before the electrode is implanted. Through the brain structure and brain metabolism information of the patient and the combination of clinical manifestations of the patient and the results of scalp electroencephalogram monitoring, a neurologist can make a detailed electrode implantation plan for each patient. After the electrode implantation operation, the patient will perform a CT scan of the whole brain to determine the spatial location of the electrodes in the whole brain. Because machines for PET, MRI and CT (electronic computed tomography) scanning are different, there are differences in space between the various brain imaging results, and the system of the present invention performs fusion according to the differences, so as to more accurately locate the brain structural abnormality and brain metabolic abnormality regions of the patient, as described in detail below.

As shown in fig. 3, the brain image synthesizing module converts the spatial positions of PET (positron emission tomography) and CT images from a three-dimensional space by a linear transformation method to fuse the brain imaging results of PET, MRI (magnetic resonance imaging) and CT. As shown in fig. 3, the brain image synthesis module of the present embodiment includes data processing for Magnetic Resonance Imaging (MRI), positron Emission Tomography (PET) and Computed Tomography (CT). The brain image synthesis module is used for fusing brain image data to determine the brain structure abnormality and brain metabolism abnormality areas of the patient; meanwhile, the specific sulcus structure of each electrode site is determined through three-dimensional reconstruction of encephalic electrode brain image data so as to reconstruct the brain structure position of the encephalic electrode.

Specifically, as shown in fig. 4, the brain image synthesizing module of the present embodiment includes:

a magnetic resonance image acquisition (CT) unit for acquiring MRI brain image data by magnetic resonance imaging data processing;

a Positron Emission Tomography (PET) data acquisition unit for acquiring PET brain image data by positron emission tomography data processing;

an electronic Computed Tomography (CT) data acquisition unit for acquiring CT brain image data by electronic computed tomography data processing;

the fusion unit is respectively connected with the magnetic resonance image acquisition unit, the positron emission tomography data acquisition unit and the electronic computer tomography data acquisition unit, and is used for fusing the PET brain image data and the MRI brain image data, fusing the imaging results of the PET brain image data and the MRI brain image data, processing the imaging results, adjusting the contrast of PET and MRI, and comparing and positioning the brain metabolic abnormality of the patient with the brain structural abnormality; and marking the positions of the brain areas for positioning the cerebral metabolic abnormality and the cerebral structural abnormality of the patient, and sending the marked cerebral areas with structural abnormality or metabolic abnormality to the three-dimensional imaging module.

As shown in fig. 5, fig. 5 is a schematic view illustrating an image fusion process of PET brain image data and MRI brain image data according to the present invention, the brain image synthesis module according to the embodiment of the present invention fuses PET (positron emission tomography data processing) and MRI (magnetic resonance imaging data processing) imaging results, processes the imaging results, and adjusts contrast between PET and MRI, so as to compare and locate brain metabolic abnormality and brain structural abnormality of a patient. And the brain area positions for positioning the cerebral metabolic abnormality and the cerebral structural abnormality of the patient are marked, and the calibrated cerebral area with the structural abnormality or the metabolic abnormality is recorded and sent to the three-dimensional imaging module, so that cerebral image data evidence is provided for finally constructing an epileptogenic focus of the epilepsy.

FIG. 6b is a CT and MRI image fusion of the present invention, and FIG. 6a is a prior art patient MRI image and standardized template matching; the results show that conventional methods match the patient MRI image to a standardized template and then correct the CT image using the corresponding parameters. This has the consequence that the patient's brain is deformed and the position of the corresponding CT electrode is bent. This approach does not accurately locate the specific brain structure of the patient electrode. As shown in fig. 6b, the system of the present invention uses a voxel calculation method based on the individual brain structure, reconstructs the structure of the sulcus of the patient through probability calculation, and positions the electrode position at the individual level by combining the real CT image of the patient, so as to obtain an accurate positioning result of the epileptogenic focus of epilepsy. This is clinically important to determine the source of the electrode signal.

Further, the electroencephalogram signal processing module according to the embodiment of the present invention is specifically described as follows:

the second main part of the invention is an electroencephalogram signal processing module. As shown in fig. 7, the intracranial electroencephalogram feature extraction module of the electroencephalogram signal processing module according to the embodiment of the present invention includes:

The specific electroencephalogram signal processing flow is shown in fig. 8, and for the processing of the electroencephalogram signal, two aspects of data are mainly merged: intracranial electrophysiological signals and low frequency electrical stimulation signals; the method comprises the steps of obtaining spontaneous intracranial electrophysiological signals of a patient and obtaining intracranial electrophysiological signals induced by electrically stimulating intracranial electrodes of the patient at low frequency respectively.

In the embodiment of the invention, the acquired electrophysiological signals generated by the patient further include patient resting state data, sleeping state data and attack period data. The resting state data refers to intracranial electroencephalogram signals when a patient lies on a bed calmly without doing anything, and the signals reflect spontaneous brain connection relation of the patient in a resting state. The sleep state data refers to intracranial electroencephalogram signals of a patient in a sleep period, and the epileptogenic focus of the epileptic patient accidentally discharges in the sleep period of the patient, so the sleep state data is extracted, and the sleep state data has guiding significance for distinguishing the epileptogenic focus. And the seizure period data refers to intracranial electroencephalogram signals of a patient when epileptic symptoms are seized, and the data is a main reference for a clinician in the process of diagnosing epileptogenic focus.

Referring to fig. 9, fig. 9 is a schematic diagram of an electroencephalogram signal processing deep learning network and a graph integration type algorithm model of an epileptic focus positioning system based on deep learning and electrophysiological signals, and as shown in fig. 9, the invention preprocesses data of three modes (respectively, intracranial electroencephalogram signals in a resting state, intracranial electroencephalogram signals in a sleep period, and intracranial electroencephalogram signals in an epileptic symptom attack), performs time-frequency analysis on the data of the three modes, and extracts energy (2-4 Hz, 4-8Hz, 8-12Hz, 12-30Hz, 30-60Hz, and 60-150 Hz) of the data of the three modes in different frequency bands. The extracted energy is used as input to enter a deep learning characteristic extraction module, and the original characteristics of the self-generating physiological signal are extracted through an encoder of the deep learning characteristic extraction module. The extracted potential tokens (Xn) serve as attribute features of nodes in the graph volume model. The deep learning feature extraction module (AE) is a deep learning neural network model, input features can be copied to output, compared with the traditional methods such as principal component analysis, the deep learning neural network algorithm is introduced to compress coding dimensionality, and potential characteristics obtained through training can retain effective information of most original data. And then reconstructing the low-dimensional potential characterization and raising the dimension of the low-dimensional potential characterization to enable the reconstructed output feature to be close to the input feature, and then considering that the low-dimensional potential characterization can represent the original high-dimensional feature.

The deep learning feature extraction module AE in the embodiment of the invention mainly comprises two parts: an encoder and a decoder. The self-encoder resembles a multi-layer perceptron, using one input layer, one output layer and one or more hidden layer connections; defining the encoder and decoder transitions as phi and psi, the input features as X, the output features as Y and the potential features as F, the following mapping:

ψ：F→Y

when only one hidden layer condition, namely the hidden layer, namely the potential feature, is represented, the potential feature is represented as:

where σ is an activation function, such as sigmoid or relu; w is the weight matrix, b is the bias vector, W and b are updated by back-propagation iterations, after which the decoder maps h to the output feature Y by reconstruction:

Y＝σ ₂ (W ₂ h+b ₂ )

AE is trained to obtain a minimum reconstruction error, i.e. a loss function;

L(X，Y)＝||X-Y|| ² ＝||X-σ ₂ (W ₂ (σ ₁ (W ₁ X+b ₁ )))+b ₂ )|| ² 。

further, as shown in fig. 10, the cortical electrical stimulation network constructing module of the electroencephalogram signal processing module in the embodiment of the present invention includes:

Specifically, as shown in fig. 8, the cortical electrical stimulation network construction module of the electroencephalogram signal processing module in the embodiment of the present invention is used for creating a brain structure connection model between electrodes through signal detection, specifically mainly acquiring intracranial electrophysiological signals induced by low-frequency electrical stimulation of the intracranial electrodes of a patient, where the low-frequency electrical stimulation signals of the present invention reflect the property of structural connection of brain regions where the electrodes are located. Administration of direct electrical stimulation to a site on the cortex induces evoked potentials (CCEPs, second row first diagram shown in fig. 9) elsewhere on the cortex that predict resting state fMRI results and allow for functional and pathological brain network conclusions to be examined. Cortical evoked potentials CCEPs generally include an early (10-50 ms) N1 and a late (50-500 ms) slow wave N2. The second plot in the second row of fig. 9 is the amplitude of the original CCEP of the subject, with the ordinate representing the electrode from which the stimulus was emitted and the abscissa representing the electrode receiving the stimulus. For example, the first row of the matrix represents the response of all other electrodes to the cortical evoked potential CCEP made by this electrical stimulation when the first pair of electrodes is stimulated, and in the specific experiment of the embodiment of the present invention, the cortical electrical stimulation network construction module of the electroencephalogram signal processing module performs pair-by-pair electrical stimulation on each pair of electrodes on the gray matter cortex of each subject, so as to construct a map of structural connections between all electrodes of the whole brain of a single subject. By using a threshold, CCEP (cortical evoked potential) results were obtained connecting significant electrodes to electrodes. The CCEP (cortical evoked potential) linkage map is used as an edge attribute feature in the graph convolution model.

In the embodiment of the present invention, as shown in fig. 11, the volume integral classification module (also called as an electrode classification module) of the electroencephalogram signal processing module includes:

Specifically, as shown in the third row of fig. 9, an attribute feature conversion unit is added in the graph integration module, and is configured to convert the intracranial electrophysiological signals and the low-frequency electrical stimulation signals to obtain node attribute features and edge attribute features, where the node attributes are features of data in a sleep state, a resting state, and an attack period; the edge attribute feature is the connection of the cortex to the electrical stimulation of the cortex; the self-adaptive graph volume integral type unit classifies all the electrodes by adopting a self-adaptive graph convolution method so as to distinguish the normal brain area and the seizure focus of the patient. The self-adaptive graph convolution method is based on a k-order graph convolution method, utilizes a self-adaptive mechanism to determine a k value, and ensures that the graph convolution times k can obtain a more appropriate value under different graph structures, so that the purpose of unsupervised learning is further achieved on the premise of ensuring the correctness of results.

Specifically, in the embodiment of the present invention, when the adaptive graph volume integral type unit performs specific processing, as shown in fig. 9, when feature embedding is obtained after graph convolution, a k-means method is adopted to segment the feature embedding in order to divide V nodes (i.e., intracranial electrode sites) into m clusters (where a cluster represents an epileptogenic region and a non-epileptogenic region), where k is obtained adaptively, and it is adaptively determined that k is a very critical link in graph volume; in order to select a proper k, the intra-cluster distance expression of a clustering index is adopted, intra (C) represents the intra-cluster distance of a cluster C, x represents node characteristics, i and j represent different nodes respectively, and v represents a set of the node characteristics and is represented as follows:

the measurement of the cluster-inside and cluster-to-cluster distances of the better clustering result is that the cluster-inside distances are smaller, which indicates that the features in the clusters are more similar and the nodes are more dense, and based on the measurement, the cluster-to-cluster distances are larger to separate different cluster nodes. According to the self-adaptive graph volume integral type unit self-adaptive process in the embodiment of the invention, according to the change condition of the intra-cluster distance, the intra-cluster result obtained by embedding the current order graph can be obtained every time graph volume is executed, iteration is continuously carried out, if the intra-cluster result obtained by the k-th graph volume is larger than the last time, the iteration is stopped, and the clustering result is optimal when the intra-cluster distance is the minimum from the last time result, so that the clustering order k-1 is obtained.

Convolution of a k-order graph, wherein X represents a potential representation of each electrode extracted by the deep learning feature extraction module, and L represents a Laplace matrix connected between electrodes constructed by cortical-cortical electrical stimulation:

where k is a positive integer representing the number of graph convolutions, and the corresponding graph filter G is:

wherein, U is a potential characteristic corresponding feature vector, and Λ is a feature vector arranged in an ascending order.

The neighbor features of the first-order graph convolution aggregation node are embedded in the features of the first-order aggregation neighbor features during second-order graph convolution, so that the neighbors of the nodes during second-order convolution contain first-order neighbor information of node neighbors, and the second-order neighbor features of the nodes can be aggregated during second-order convolution. Because k-order graph convolution converges k-order neighbor information when updating node characteristics, long-distance data relation is considered when using k-order graph convolution, too few graph convolution times result in insufficient information amount convergence, and too many graph convolution times result in similar node characteristic embedding of the global graph, thereby generating an over-smooth phenomenon. So k max is set to 25.

The K-means in the embodiment of the invention is a more classical partitional clustering algorithm, and is widely applied to clustering large-scale data due to high efficiency of the algorithm. The algorithm principle k-means algorithm takes k as a parameter, and divides n objects into k clusters, so that the clusters have higher similarity and the similarity between the clusters is lower.

Further, in the epilepsy focal zone positioning system based on deep learning and electrophysiological signals according to the embodiment of the present invention, as shown in fig. 13, the three-dimensional stereo imaging module includes:

and the three-dimensional imaging unit is used for obtaining the high-grade polymeric nucleus according to the epilepsia weight score of the brain structure abnormal value, the brain metabolism abnormal value and the chart volume epilepsia predicted value, and drawing the regional outline of an epileptogenic focus of the epilepsia in the three-dimensional structure so as to simulate a three-dimensional model of the brain structure of the patient needing to be cut.

The embodiment of the invention relates to a three-dimensional brain imaging module, which comprises the following processing procedures: the three-dimensional brain imaging module combines the brain image pathological result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module, so that a three-dimensional brain structure model which needs to be cut off by a patient can be accurately simulated, specifically, the epilepsy weight assignment unit comprehensively considers the results obtained by the brain image synthesis module and the electroencephalogram information processing module, and assigns an epilepsy weight to each voxel of the whole brain. The epilepsy weighted value comprises three parts of a brain structure abnormal value, a brain metabolism abnormal value and a chart volume epilepsy predicted value. The three-dimensional stereo imaging unit obtains the high-grade polymeric nuclei according to scores of three aspects of a brain structure abnormal value, a brain metabolism abnormal value and a graph volume epilepsy predicted value, and draws the regional outline (a red region in fig. 12 b) of an epilepsy epileptogenic focus in the three-dimensional stereo structure, wherein the graph in fig. 12a is a schematic diagram of the imaging result of a patient blood vessel and an electrode fused by MRI and CT, and the graph in fig. 12b is a final result diagram obtained by the invention.

In a preferred embodiment, the specific processes of the three-dimensional stereo brain imaging module of the present invention are exemplified by: taking the results obtained by the brain image synthesis module and the electroencephalogram information processing module into comprehensive consideration, firstly, dividing the whole brain into about 109 blocks (voxels) with the size of one cubic millimeter, and assigning an epilepsy weight index (EI) to each block, wherein the epilepsy weight comprises three parts of a brain structure abnormal value (SI), a brain metabolism abnormal value (MI) and a map-volume epilepsy predicted value (GI):

EI＝W1*SI+W2*MI+W3*GI

wherein, W1, W2 and W3 are preset weight parameters, which can be given different weights by the clinician according to the experience and the clinical performance of the patient and are preset in the system. The SI and MI are obtained by a brain image synthesis module. The brain image synthesis module maps the marked brain structure abnormal region and the brain metabolism abnormal region to a specific voxel through an image, wherein the voxel SI/MI =1 in the marked region and the voxel SI/MI =0 in other regions. Wherein, GI is obtained by the brain electrical information processing module. According to the embodiment of the invention, an electroencephalogram information processing module classifies the intracranial electrodes of a patient into seizure-causing area electrodes and non-seizure-causing area electrodes through a deep learning combined graph-convolution algorithm. Each electrode has a corresponding RSA coordinate, and the area of the spherical star is calibrated by taking the coordinate position of the electrode as the center of a circle and 1.5cm as the radius. Voxels in the spherical star region near the epileptic zone electrode GI =1, and voxels in the other regions GI =0.

And (4) integrating the scores of the three aspects to obtain high-grade polymeric nuclei, and drawing the regional outline of the epileptogenic focus of the epilepsy in the three-dimensional structure (as shown in a black line marked region 11 in fig. 12 b). In clinical practice, the clinician can adjust the final resection area according to the explicit behavior of the patient and the diagnosis opinions to obtain a three-dimensional model of the final resection area.

Therefore, the regional contour of the epileptogenic focus of the epilepsy is drawn in the three-dimensional structure obtained by the system, the accurate epileptogenic focus of a patient can be obtained, and the efficiency and the accuracy of detecting and positioning the epileptogenic focus of the epilepsy are improved.

As can be seen from the above, according to the epilepsy focus area positioning system based on deep learning and electrophysiological signals, the brain structure three-dimensional model which needs to be cut by a patient can be accurately simulated according to the brain image pathological result generated by the brain image synthesis module and the electrode classification result generated by the electroencephalogram signal processing module, so that a surgeon can make preoperative assessment and surgical simulation, and the workload of interpreting electroencephalograms by the surgeon is greatly saved by combining with an artificial intelligence algorithm, and meanwhile, the efficiency and accuracy of detecting and positioning epilepsy-causing focuses are improved. The invention also has the following advantages:

1) The epileptogenic focus positioning system based on deep learning and electrophysiological signals can draw the regional contour of an epileptogenic focus of epilepsy in a three-dimensional structure to accurately position the epileptogenic focus of epilepsy, so that the detection positioning efficiency and accuracy of the epileptogenic focus of epilepsy are improved, and a clearer electroencephalogram result graph can be obtained from the blood vessel and electrode imaging result schematic diagram of a patient fused by MRI and CT in the invention in figure 12a and the finally obtained result graph in the invention in figure 12b, so that the reading efficiency of electroencephalogram is improved.

2) The prior art focuses on capturing high-frequency signals by using an artificial intelligence algorithm as a basis for positioning an epileptic region, and a data set only comprises data of a seizure period and data of an early seizure period. The epileptic focal zone positioning system based on deep learning and electrophysiological signals integrates data of sleep stage, resting state and attack stage, and classifies intracranial electrodes of patients by starting from electroencephalogram full-band signals and combining cortical-to-cortical electrical stimulation structural connection. The invention has more comprehensive utilization of intracranial signals of patients, wider fused information and more accurate positioning result.

3) The epileptogenic focus positioning system based on deep learning and electrophysiological signals integrates auxiliary data required by neurology diagnosis and neurosurgery in combination with clinical requirements, and provides a set of complete and visualized 3D modeling system and method. The visual and specific simulation is provided for the positioning of the epileptic focus area, and the assistance of providing more accurate positioning help for the epileptic focus area is facilitated.

4) The epilepsy focus area positioning system based on deep learning and electrophysiological signals is used for predicting epilepsy-causing areas of patients through clinical practice, and the accuracy of prediction results is high.

Exemplary method

Based on the above embodiment of the epileptic focal zone localization system based on deep learning and electrophysiological signals, as shown in fig. 14, an embodiment of the present invention further provides an epileptic focal zone localization method of the epileptic focal zone localization system based on deep learning and electrophysiological signals as described in any one of the above 7, where the method includes the following steps:

step S100, controlling a brain image synthesis module to acquire brain image data, fusing the brain image data, determining a brain structure abnormal region and a brain metabolism abnormal region of a patient, performing three-dimensional reconstruction on brain image data of an intracranial electrode, determining a specific sulcus structure of each electrode site, and generating a brain image pathological result, wherein the brain image pathological result is specifically described in the embodiment of the system;

s200, controlling an electroencephalogram signal processing module to acquire an intracranial electroencephalogram signal, performing machine interpretation on the intracranial electroencephalogram signal by using an artificial intelligence algorithm, combining a multi-mode electrophysiological signal and an electrical stimulation structure connection signal, classifying a normal electrode and an epileptogenic zone electrode to generate an electrode classification result, wherein the specific steps are as described in the embodiment of the system;

and S300, controlling a three-dimensional imaging module to simulate and output a three-dimensional stereo model of a brain structure of a patient needing to be cut off according to a brain image pathological result generated by the brain image synthesis module and an electrode classification result generated by the electroencephalogram signal processing module, wherein the three-dimensional stereo model is specifically described in the embodiment of the system.

Further, the step of controlling in step S100 to obtain brain image data through a brain image synthesis module, and fuse the brain image data to determine a brain structural abnormality and a brain metabolic abnormality region of the patient, and at the same time, to three-dimensionally reconstruct brain image data of the intracranial electrode, to determine a specific sulcus loop structure of each electrode site, and to generate a brain image pathological result includes:

the control fusion unit is used for fusing the PET brain image data and the MRI brain image data, fusing imaging results of the PET brain image data and the MRI brain image data, processing the imaging results, adjusting the contrast ratio of PET and MRI, and comparing and positioning brain area positions of brain metabolic abnormality and brain structural abnormality of a patient; and marking the positions of the brain areas for positioning the cerebral metabolic abnormality and the cerebral structural abnormality of the patient, and sending the marked cerebral areas with the structural abnormality or the metabolic abnormality to the three-dimensional imaging module, specifically as described in the above system embodiment.

The step S200 of obtaining an intracranial electroencephalogram signal through the electroencephalogram signal processing module, performing machine interpretation on the intracranial electroencephalogram signal by using an artificial intelligence algorithm, combining a multi-modal electrophysiological signal and an electrical stimulation structure connection signal, classifying a normal electrode and an epileptic-induced region electrode, and generating an electrode classification result comprises the following steps:

the control cortical electrical stimulation network construction module is used for establishing a brain structure connection model between the electrodes through signal detection;

the control chart convolution classification module classifies the intracranial electrodes by using a chart convolution method and combining the attribute features extracted by the intracranial electroencephalogram feature extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module, judges the normal electrodes and the electrodes in the seizure area, and generates an electrode classification result, which is specifically described in the system embodiment.

Further, the step S300 of simulating and outputting a three-dimensional stereo model of the brain structure of the patient to be excised by the three-dimensional stereo imaging module according to the brain image pathology result generated by the brain image synthesis module and the electrode classification result generated by the electroencephalogram signal processing module includes:

controlling an epilepsy weight assignment unit to combine the brain image pathological result generated by the brain image synthesis module with the electrode classification result generated by the electroencephalogram signal processing module, and endowing an epilepsy weight value to each voxel of the whole brain, wherein the epilepsy weight value comprises a brain structure abnormal value, a brain metabolism abnormal value and a volume epilepsy predicted value;

and controlling a three-dimensional imaging unit, obtaining a high-grade polymeric nucleus according to the epilepsia weight score of the brain structure abnormal value, the brain metabolism abnormal value and the chart volume epilepsia predicted value, and drawing the regional outline of an epileptogenic focus of the epilepsia in the three-dimensional structure so as to simulate a three-dimensional brain structure model which needs to be cut by a patient, specifically as described in the embodiment of the system.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by hardware instructions of a computer program, which may be stored in a non-volatile computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may 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. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus (Rambus) direct RAM (RDRAM), direct Rambus Dynamic RAM (DRDRAM), and Rambus Dynamic RAM (RDRAM), among others.

In conclusion, the deep learning and electrophysiological signal-based epileptic focus positioning system and method can draw the regional outline of an epileptic focus in a three-dimensional structure so as to accurately position the epileptic focus and improve the efficiency and accuracy of detection and positioning of the epileptic focus; and from the final result picture, can get more clear electroencephalogram result picture, has improved the efficiency that the electroencephalogram reads.

It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.

Claims

1. An epileptic focal zone localization system based on deep learning and electrophysiological signals, comprising:

2. The deep learning and electrophysiological signal based epileptic focal zone localization system of claim 1, wherein the brain image synthesis module comprises:

3. The deep learning and electrophysiological signal based epileptic focal zone localization system of claim 2, wherein the electroencephalogram signal processing module comprises: the intracranial electroencephalogram feature extraction module, the cortical electrical stimulation network construction module and the picture volume integral module are sequentially connected;

the intracranial electroencephalogram feature extraction module is used for acquiring intracranial electroencephalogram signals and extracting features of the intracranial electroencephalogram signals in sleep states, rest states and attack periods through a deep learning algorithm;

4. The deep learning and electrophysiological signal based epileptic focal zone localization system of claim 3, wherein the intracranial electroencephalogram feature extraction module comprises:

5. The deep learning and electrophysiological signal based epileptic focal zone localization system of claim 3, wherein the cortical electrical stimulation network building module comprises:

6. The deep learning and electrophysiological signal based epileptic focal zone localization system of claim 3, wherein the graph convolution classification module comprises:

the attribute feature conversion unit is used for converting the intracranial electrophysiological signals and the low-frequency electrical stimulation signals to obtain node attribute features and edge attribute features;

7. The deep learning and electrophysiological signal based epileptic focal zone localization system of claim 1, wherein the three-dimensional stereo imaging module comprises:

8. A method for epileptic focal zone localization according to the system for epileptic focal zone localization based on deep learning and electrophysiological signals of any of claims 1-7, the method comprising:

acquiring brain image data through a brain image synthesis module, fusing the brain image data, determining brain structure abnormality and brain metabolism abnormality areas of a patient, performing three-dimensional reconstruction on the brain image data of the intracranial electrode, determining a specific sulcus loop structure of each electrode site, and generating a brain image pathological result;

9. The method for locating epileptic focal zone of an epileptic focal zone locating system based on deep learning and electrophysiological signals as claimed in claim 8, wherein the step of controlling to obtain brain image data through a brain image synthesis module, and to fuse the brain image data to determine the brain structural abnormality and brain metabolic abnormality regions of the patient, and to perform three-dimensional reconstruction of brain image data of intracranial electrodes to determine the specific sulcus structure of each electrode site, and to generate the pathological result of the brain image comprises:

the control method comprises the following steps of obtaining intracranial electroencephalogram signals through an electroencephalogram signal processing module, performing machine interpretation on the intracranial electroencephalogram signals by using an artificial intelligence algorithm, combining multi-mode electrophysiological signals and electrical stimulation structure connection signals, classifying normal electrodes and electrodes in epileptogenic regions of epilepsy, and generating electrode classification results, wherein the steps comprise:

the control chart convolution classification module classifies the intracranial electrodes by using a chart convolution method and combining the attribute characteristics extracted by the intracranial electroencephalogram characteristic extraction module and the structural connection map constructed by the cortical electrical stimulation network construction module, judges the normal electrodes and the electrodes in the seizure area and generates an electrode classification result.

10. The method for locating epileptic focal zone of the system for locating epileptic focal zone based on deep learning and electrophysiological signals as claimed in claim 8, wherein the step of controlling the brain image pathological result generated by the brain image synthesizing module and the electrode classification result generated by the electroencephalogram signal processing module to simulate and output a three-dimensional stereo model of brain structure to be excised by the patient via a three-dimensional stereo imaging module comprises:

and controlling a three-dimensional imaging unit, obtaining a high-grade polymeric nucleus according to the epilepsia weight score of the brain structure abnormal value, the brain metabolism abnormal value and the chart volume epilepsia predicted value, and drawing the regional outline of an epileptogenic focus of the epilepsia in the three-dimensional structure so as to simulate a three-dimensional model of the brain structure of the patient, which needs to be cut off.