An MRI-based multi-focal epilepsy localization method and device

Abstract

The present application discloses a method and device for multi-focal epilepsy localization based on MRI. The method for multi-focal epilepsy localization based on MRI includes: obtaining a trained lesion-cortical epilepsy network; according to the trained classification model; obtaining image information to be predicted; extracting image features of the image information to be predicted; inputting the image features into the trained lesion-cortical epilepsy network to obtain a lesion-cortical epilepsy network result; inputting the image features into the trained classification model to obtain a prediction result; and obtaining a final prediction result according to the lesion-cortical epilepsy network result and the prediction result. The method for multi-focal epilepsy localization based on MRI in the present application considers both the epileptic structural network and the functional network, and at the same time combines the multi-lesion characteristics of patients to establish a fusion model, providing a non-invasive, simple and easy-to-implement method for multi-focal epilepsy localization based solely on MRI.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular, to a method for localizing multi-focal epilepsy based on MRI and a device for localizing multi-focal epilepsy based on MRI. Background Art

[0002] Epilepsy is a common and frequently-occurring disease of the nervous system. In addition to frequent epileptic seizures, these patients often suffer from cognitive and mental disorders, which not only seriously affect the physical and mental health of the patients, but also impose a huge economic burden on society and families, becoming one of the prominent social problems.

[0003] Most focal epilepsies have an origin area, but some epileptic patients have multiple pathological foci, such as tuberous sclerosis complex (TSC) (this application takes tuberous sclerosis complex as an example), multiple encephalomalacia, and multiple cavernous hemangiomas of the brain. It is extremely important to find and localize a single epileptogenic focus from multiple pathological foci for the control of epilepsy.

[0004] At present, the localization of epileptic foci mainly relies on clinical symptomatology, the location of pathological foci (MRI), and electroencephalogram (including scalp electroencephalogram and intracranial electrode electroencephalogram, etc.). However, more than 60% of patients cannot be localized by non-invasive methods, 20% of patients can be localized by invasive methods, and 40% of patients cannot localize the epileptogenic focus, which seriously affects the surgical treatment.

[0005] Therefore, it is desirable to have a technical solution to solve or at least mitigate the above deficiencies of the prior art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for localizing multi-focal epilepsy based on MRI to at least solve one of the above technical problems.

[0007] The present invention provides the following solutions:

[0008] According to one aspect of the present invention, there is provided a method for localizing multi-focal epilepsy based on MRI, including:

[0009] Obtain a trained lesion-cortical epilepsy network;

[0010] According to the trained classification model;

[0011] Obtain the image information to be predicted;

[0012] Extract the image features of the image information to be predicted;

[0013] Input the image features into the trained lesion-cortical epilepsy network to obtain the result of the lesion-cortical epilepsy network;

[0014] Input the image features into the trained classification model to obtain a prediction result;

[0015] Obtain a final prediction result according to the lesion-cortical epilepsy network result and the prediction result.

[0016] Optionally, the lesion-cortical epilepsy network is established by the following method:

[0017] Obtain a sensitivity T-Map image of the epileptogenic lesion;

[0018] Obtain a specificity T-Map image of the epileptogenic lesion;

[0019] Overlay the sensitivity T-Map image and the specificity T-Map image, retain the overlapping area, divide the overlapping voxels into a positive connection area and a negative connection area according to the positive and negative T values, calculate the positive functional connection between the positive connection voxels and all cortical voxels, and the negative functional connection between the negative connection voxels and all cortical voxels, and superimpose the positive functional connection and the negative functional connection to obtain the lesion-cortical epilepsy network.

[0020] Optionally, the obtaining of the sensitivity T-Map image of the epileptogenic lesion includes:

[0021] Obtain a group of T1 images, the group of T1 images includes multiple T1 images, one T1 image corresponds to one patient, where the patient includes TSC-related epilepsy patients who have completed resection surgery and patients without seizures 6 months after surgery;

[0022] Obtain a group of T2-Flair images, the group of T2-Flair images includes multiple T2-Flair images, one T2-Flair image corresponds to one patient, where the patient includes TSC-related epilepsy patients who have completed resection surgery and patients without seizures 6 months after surgery, and each T2-Flair image has an epileptogenic lesion ROI on the image;

[0023] Obtain a group of EPI images, the group of EPI images includes multiple EPI images, one EPI image corresponds to one healthy person;

[0024] Register the group of T1 images and the group of T2-Flair images to obtain a first MNI template;

[0025] Normalize the group of EPI images into the first MNI template to obtain a second MNI template;

[0026] Take the epileptogenic focus ROI within the second MNI template as the seed region, calculate the average time series of each seed region in the EPI image, calculate the functional connectivity between the seed region and all cortical voxels using the Pearson correlation coefficient, and use the Fisher Z-transform to ensure that the Pearson correlation coefficient conforms to the normal distribution; among them, each epileptogenic focus ROI will obtain functional connectivity maps with the same number as the EPI images, integrate the above functional connectivity maps into a network diagram using a one-sample T-test, overlay them and calculate the voxels that appear in 50% of the patients, and retain the average brain functional connectivity of the overlapping region to obtain the sensitivity T-Map of the epileptogenic focus.

[0027] Optionally, the obtaining of the specific T-Map of the epileptogenic focus includes:

[0028] Obtain the functional connectivity map of the epileptogenic nodule group and the functional connectivity map of the non-epileptogenic nodule group;

[0029] Use a two-sample T-test to perform an inter-group comparison of the functional connectivity map of the epileptogenic nodule group and the functional connectivity map of the non-epileptogenic nodule group, and perform FDR correction, select the voxels with p < 0.05, and establish a T-Map.

[0030] Optionally, the epileptogenic focus ROI on the T2-Flair image is obtained by the following method:

[0031] Obtain the first cortical lesion region of interest and the second cortical lesion region of interest of each T2-Flair image;

[0032] Calculate the coincidence degree in the cortical lesion recognition of the first cortical lesion region of interest and the second cortical lesion region of interest of each T2-Flair image;

[0033] Make the following judgments on the coincidence degree in the cortical lesion recognition of each T2-Flair image respectively: Judge whether the coincidence degree is greater than the preset threshold, if so, regard the overlapping region as the epileptogenic focus ROI.

[0034] Optionally, the registration of the T1 image group and the T2-Flair image group to obtain the first MNI template includes:

[0035] Register the T2-Flair image of each patient to the T1 image, then register each epileptogenic focus ROI to the T1 image, and finally standardize it into the MNI template.

[0036] Optionally, the trained classification model is obtained by the following method:

[0037] Obtain a training set, and the training set includes images of epileptic patients and images of non-epileptic patients;

[0038] Calculate the average time series of the lesion area based on the images of non-epileptic patients in the training set. Subsequently, calculate the functional connectivity between the lesion area and the whole-brain voxels, and then perform Fisher z-transform to obtain the functional connectivity matrix;

[0039] The functional connectivity of the sub-patients with epileptogenic lesions as the lesions is designated as positive samples, while the connections of the sub-patients with non-epileptogenic lesions as the lesions are labeled as negative samples;

[0040] Reduce the dimension of the functional connectivity matrix by the principal component analysis method to obtain feature information;

[0041] Obtain a random forest classifier;

[0042] Train the random forest classifier with the feature information to obtain the hyperparameters of the random forest model.

[0043] This application also provides an MRI-based multi-focal epilepsy localization device, and the MRI-based multi-focal epilepsy localization device includes:

[0044] A lesion-cortical epilepsy network acquisition module, which is used to obtain a trained lesion-cortical epilepsy network;

[0045] A classification model acquisition module, which is used to obtain a trained classification model;

[0046] A to-be-predicted image information acquisition module, which is used to obtain to-be-predicted image information;

[0047] An image feature acquisition module, which is used to extract the image features of the to-be-predicted image information;

[0048] A lesion-cortical epilepsy network result acquisition module, which is used to input the image features into the trained lesion-cortical epilepsy network to obtain the lesion-cortical epilepsy network result;

[0049] A prediction result acquisition module, which is used to input the image features into the trained classification model to obtain the prediction result;

[0050] A final prediction result acquisition module, which is used to obtain the final prediction result according to the lesion-cortical epilepsy network result and the prediction result.

[0051] The MRI-based multi-focal epilepsy localization method of this application considers both the epileptic structural network and the functional network, and combines the multi-lesion characteristics of patients to establish a fusion model, providing a non-invasive, simple and easy-to-implement MRI-based multi-focal epilepsy localization method. Description of the Drawings

[0052] Figure 1 is a schematic flowchart of the MRI-based multi-focal epilepsy localization method in an embodiment of this application.

[0053] Figure 2 is a schematic diagram of the MRI-based multi-focal epilepsy localization device in an embodiment of this application.

[0054] Figure 3 is a schematic overall flowchart of the MRI-based multi-focal epilepsy localization method in an embodiment of this application. Detailed Embodiments

[0055] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0056] Figure 1 is a schematic flowchart of the MRI-based multi-focal epilepsy localization method in an embodiment of this application.

[0057] Such as Figure 1 and Figure 3 The MRI-based multi-focal epilepsy localization method shown includes:

[0058] Obtain a trained lesion-cortical epilepsy network;

[0059] According to the trained classification model;

[0060] Obtain the image information to be predicted;

[0061] Extract the image features of the image information to be predicted;

[0062] Input the image features into the trained lesion-cortical epilepsy network to obtain the lesion-cortical epilepsy network result;

[0063] Input the image features into the trained classification model to obtain the prediction result;

[0064] Obtain the final prediction result according to the lesion-cortical epilepsy network result and the prediction result.

[0065] In this embodiment, the lesion-cortical epilepsy network is established by the following method:

[0066] Obtain the sensitivity T-Map image of the epileptogenic lesion;

[0067] Obtain the specificity T-Map image of the epileptogenic lesion;

[0068] Overlay the sensitivity T-Map image and the specificity T-Map image, retain the overlapping area, and divide the overlapping voxels into a positive connection area and a negative connection area according to the positive and negative T values. Calculate the positive functional connection between the positive connection voxels and all cortical voxels, and the negative functional connection between the negative connection voxels and all cortical voxels. Superimpose the positive functional connection and the negative functional connection to obtain the lesion-cortical epilepsy network.

[0069] In this embodiment, the obtaining of the sensitivity T-Map image of the epileptogenic lesion includes:

[0070] Obtain a group of T1 images, where the group of T1 images includes multiple T1 images (T1-weighted images), and one T1 image corresponds to one patient. Among them, the patients include TSC-related epilepsy patients who have completed resection surgery and have no seizures 6 months after surgery;

[0071] Obtain a group of T2-Flair images, where the group of T2-Flair images includes multiple T2-Flair images, and one T2-Flair image (T2-weighted image) corresponds to one patient. Among them, the patients include TSC-related epilepsy patients who have completed resection surgery and have no seizures 6 months after surgery, and each T2-Flair image has an epileptogenic lesion ROI on the image;

[0072] Obtain a group of EPI images, where the group of EPI images (resting-state functional MRI images) includes multiple EPI images, and one EPI image corresponds to one healthy person;

[0073] Register and standardize the group of T1 images and the group of T2-Flair images to obtain the first MNI template;

[0074] Standardize the group of EPI images into the first MNI template to obtain the second MNI template;

[0075] Taking the epileptogenic lesion ROI within the second MNI template as the seed region, calculate the average time series of each seed region in the EPI image, calculate the functional connectivity between the seed region and all cortical voxels using the Pearson correlation coefficient, and use the Fisher Z-transform to ensure that the Pearson correlation coefficient conforms to the normal distribution; among them, each epileptogenic lesion ROI will obtain a functional connectivity map with the same number as the EPI image. Use one-sample T-test to integrate the above functional connectivity maps into a network map, overlay them and calculate the voxels that appear in 50% of the patients, and retain the average brain functional connectivity in the overlapping region to obtain the sensitivity T-Map map of the epileptogenic lesion.

[0076] In this embodiment, the obtaining of the specific T-Map map of the epileptogenic lesion includes:

[0077] Obtain the functional connectivity map of the epileptogenic nodule group and the functional connectivity map of the non-epileptogenic nodule group;

[0078] Use two-sample T-test to perform between-group comparison on the functional connectivity maps of the epileptogenic nodule group and the non-epileptogenic nodule group, and perform FDR correction. Select voxels with p < 0.05 to establish a T-Map map.

[0079] Specifically, first, it is also necessary to obtain the functional connectivity map related to the epileptogenic nodule. This method is the same as the establishment of the functional connectivity map related to the epileptogenic nodule in the construction process of the sensitivity T-Map. When constructing the T-Map map, a non-epileptogenic nodule group is also required. In this embodiment, the non-epileptogenic nodule group is the same. Calculate the functional connectivity between the average time series in the non-epileptogenic nodule mapping seed region and all cortical voxels of healthy subjects in the normal control group. Merge the standardized z-scores (the Z-score is to perform a z-transform on the previously obtained Pearson correlation coefficient, that is, the r value) of the functional connectivity of each nodule with the normal data set to obtain the functional connectivity map related to the non-epileptogenic nodule. Use two-sample T-test to perform between-group comparison on the functional connectivity maps of the epileptogenic nodule group and the non-epileptogenic nodule group, and perform FDR correction. Select voxels with p < 0.05 to establish a T-Map map.

[0080] In this embodiment, the epileptogenic lesion ROI on the T2-Flair image is obtained by the following method:

[0081] Obtain the first cortical lesion region of interest and the second cortical lesion region of interest of each T2-Flair image;

[0082] Calculate the coincidence degree in the cortical lesion recognition of the first cortical lesion region of interest and the second cortical lesion region of interest of each T2-Flair image;

[0083] For each of the T2-Flair images, the degree of overlap in the identification of cortical lesions is determined as follows: Determine whether the degree of overlap is greater than a preset threshold. If so, the overlapping region is regarded as the epileptogenic lesion ROI.

[0084] In this embodiment, registering the T1 image group with the T2-Flair image group to obtain the first MNI template includes:

[0085] Register the T2-Flair image of each patient onto the T1 image, then register each epileptogenic lesion ROI onto the T1 image, and finally standardize it into the MNI template.

[0086] In this embodiment, the trained classification model is obtained by the following method:

[0087] Obtain a training set, where the training set includes images of epileptic patients and non-epileptic patients;

[0088] Calculate the average time series of the lesion area based on the non-epileptic patient images in the training set. Subsequently, calculate the functional connectivity between the lesion area and the whole-brain voxels, and then perform Fisher z-transform to obtain the functional connectivity matrix; To reduce the dimension, the connections with zero values are excluded; For example, it can be understood as a large cube composed of 100 small cubes, and each small cube has a time series. To calculate the functional connectivity (i.e., the correlation coefficient) between the large cube and the seed point, the values of all the small cubes are averaged, and then the functional connectivity calculation (correlation coefficient) is performed with other small cubes in the brain.

[0089] The functional connectivity of the sub-patients with epileptogenic lesions is designated as positive samples, while the connections of the sub-patients with non-epileptogenic lesions are labeled as negative samples;

[0090] Reduce the dimension of the functional connectivity matrix by the principal component analysis method to obtain the feature information;

[0091] Obtain a random forest classifier;

[0092] Train the random forest classifier with the feature information to obtain the hyperparameters of the random forest model.

[0093] In this embodiment, obtaining the final prediction result according to the lesion-cortical epilepsy network result and the prediction result includes:

[0094] Combining the risk scores affected by the overlap and location of epileptic network lesions with the classification probabilities generated by the prediction model reflects the functional characteristics of the lesion area. This fusion method takes into account both structural and functional features, improving the prediction accuracy. Initially, to standardize the risk scores, we selected the 97.5% of their distribution and truncated any value exceeding 1, converting it into a standardized score within the range of 0 to 1. Then, the fusion was performed based on the consistency of the standardized scores and classification probabilities of the two models.

[0095] The present application will be further elaborated in detail by way of example below. It should be understood that this example does not constitute any limitation to the present application.

[0096] Data source acquisition:

[0097] Based on existing TSC-related epilepsy patients

[0098] Inclusion criteria: Age is not limited, male or female; MRI shows that the number of intracranial cortical nodules > 3; Clinically conforms to the 2021 edition of the international diagnostic criteria for TSC, with or without TSC gene mutations; The diagnosis of epilepsy is clear, and more than 3 anti-epileptic drugs have been taken; mTOR inhibitors or ketogenic diet therapy have been applied or not; After strict preoperative examination and evaluation, it meets the surgical indications identified in the "Chinese Expert Consensus on Surgical Treatment of TSC-related Epilepsy"; Underwent single epileptogenic nodule resection and completed 6-month follow-up; Family members agreed and signed the consent form.

[0099] Exclusion criteria: Postoperative MRI confirmed that the epileptogenic nodule determined preoperatively was not completely resected; Obvious subependymal giant cell astrocytoma; Patients who developed severe complications after surgery or had a history of brain trauma, craniotomy, or vagus nerve stimulation therapy within half a year, etc., which affected the prognosis judgment; Patients whose family members refused to participate in the study or did not complete the 6-month follow-up after surgery.

[0100] (2) Preoperative evaluation and data collection of TSC-related epilepsy patients

[0101] MRI: Collect the MRI data of the patients. The scanning parameters of the 3D-T1 sequence are as follows: repetition time = 2300 ms, echo time = 2.98 ms, flip angle 9°, FOV 256×256 mm 2 , voxel 0.5×0.5×1 mm 3 , without interval scanning; The scanning parameters of the fluid-attenuated inversion recovery sequence (Flair) are: TR = 11000 ms, TE = 140 ms, TI = 2850 ms, voxel 1×1×1 mm 3 , without interval scanning.

[0102] Clinical data: Collect the patient's gender, operative age, onset age, disease course, seizure frequency, seizure duration, number of antiepileptic drugs, and number of epileptogenic nodules.

[0103] Scalp EEG and seizure symptomatology collection: EEG is a non-invasive examination method for measuring brain electrical activity. By placing electrodes on the scalp, the electrical activity of brain neurons is recorded to help determine the type of epilepsy, localize the epileptogenic focus, the conduction of epileptic discharges, etc., and guide the medical and surgical treatment of epilepsy. EEG is an essential and indispensable examination in the pre-surgical evaluation of epilepsy. In-depth analysis of EEG is of great significance for improving the clinical diagnosis rate.

[0104] 18F-FDG PET-CT scan: PET-CT examination has important significance in the pre-surgical evaluation of TSC-related epilepsy. By detecting regions of abnormal glucose metabolism in the brain, especially the hypometabolic regions during the interictal period, it reflects the epileptic seizure onset zone and abnormal metabolic activities in other regions of the brain during epileptic seizures. It helps to understand the impact of epileptic foci on surrounding brain regions and evaluate the severity and extent of epilepsy.

[0105] SEEG: Select TSC-related epilepsy patients who have undergone comprehensive evaluations including clinical history collection, neurological examination, neuropsychological testing, high-resolution MRI, and scalp EEG, and have determined the epileptogenic nodule hypothesis (including possible primary and secondary epileptogenic nodules, etc.).

[0106] (3) Surgical treatment and follow-up of TSC-related epilepsy patients

[0107] Epileptogenic nodule resection: According to the epileptogenic nodules determined by SEEG, determine the surgical plan and perform surgical treatment under general anesthesia. The resection scope includes the epileptogenic nodules and adjacent cerebral cortex.

[0108] Postoperative follow-up and criteria for determining epileptogenic nodules: Follow up the seizure situation 6 months after surgery and classify it according to the ILAE six-level classification method. According to whether there are seizures, it is divided into the group without postoperative seizures (ILAE-I level) and the group with postoperative seizures (ILAE II-VI level). Among them, in the group without postoperative seizures, the resected nodules are considered to be accurately determined as epileptogenic nodules, and other nodules except epileptogenic nodules in these patients are considered non-epileptogenic nodules. For patients with seizures 6 months after surgery, it is considered that the determination of epileptogenic nodules is incorrect.

[0109] (4) Retrospectively collect the database of normal subject individual data

[0110] Inclusion criteria: Age is not limited, male or female; there is no abnormal signal in the 3T-MRI structural image; there is no history of seizures or neurological diseases; rs-fMRI scan has been completed and the image quality is good for subsequent analysis.

[0111] Exclusion criteria: There are motion artifacts in the MRI data; the 3T-MRI structural image data lacks 3D-MPRAGE.

[0112] (5) rs-fMRI scanning parameters and data preprocessing for normal subjects

[0113] Scanning parameters: A dataset was acquired using a gradient echo-planar imaging (GRE-EPI) sequence with a 3.0T MR750 GE scanner: TR = 2000 ms, TE = 30 ms, FA = 90°, FOV = 240×240 mm 2 , matrix = 64×64, voxel = 3.75×3.75×4 mm 3 . All subjects were informed to stay awake, close their eyes naturally, and avoid head movement during the scan.

[0114] Data preprocessing: Resting-state data preprocessing was performed using the Matlab toolbox DPABI (http: / / rfmri.org / DPABI), including the following steps: removing the first 10 time points and head motion correction; rigid registration of the T1 image to the EPI mean image; normalizing the EPI image to the Montreal Neurological Institute (MNI) standard space using the T1 image and subsequently resampling to 3×3×3 mm 3 ; removing noise, including whole-brain signal, head motion, and linear trends; temporal filtering (0.01–0.08 Hz).

[0115] In this part, each patient was divided into several "sub-patients" with only a single cortical lesion, that is, each patient could be divided into as many sub-patients as there are lesions.

[0116] Construction of the lesion-cortex related epileptic network based on rs-fMRI of normal subjects and MRI of patients: Establishment of the risk score model

[0117] Discovery set and validation set: Among the enrolled TSC (Tuberous sclerosis complex)-related epilepsy patients who completed resection surgery, 50 cases were randomly selected as the validation set according to the random number method, and patients without seizures 6 months after surgery were selected as the discovery set from the remaining patients.

[0118] Extraction of the region of interest (ROI) of cortical lesions: Two clinicians used ITK-SNAP software (V4.2.0, http: / / www.itksnap.org) to draw the ROI of cortical lesions on the T2-Flair weighted images of each patient in the discovery set and the validation set. The overlap degree in the identification of cortical lesions was calculated based on the nilearn library in Python (https: / / nilearn.github.io / stable / index.html). That is, the percentage of non-overlap in the identification of lesions by two neurologists for the same patient. If the percentage is less than 5%, the overlapping area is regarded as the cortical lesion and used for subsequent analysis. When the identification of cortical lesions by the two clinicians is inconsistent (>5%), the data is discarded or the selection of the ROI of cortical lesions is determined by a third senior clinician. Then, using FSL software (https: / / fsl.fmrib.ox.ac.uk / ), the T2-Flair image of each patient was registered to the T1 image, and then each labeled lesion was registered to the T1 image, and finally normalized to the MNI template.

[0119] Voxel-based mapping of cortical lesions and seizure symptoms: To verify that a single brain cortical voxel in patients with TSC-related epilepsy is not related to seizures, a voxel-based mapping of cortical lesions and seizure symptoms was established using a general linear model. The form of the general linear model is Y = βX + ε. Y represents the lesion area of each voxel, 1 indicates that the voxel is within the lesion area, and 0 indicates that the voxel is outside the lesion area. X represents the correlation between cortical lesions and epilepsy, 1 indicates the epileptogenic lesion, and 0 indicates the non-epileptogenic lesion. The model parameter β is estimated, and ε is the estimated residual. Finally, the p-value of the T-test for β (corrected by FDR based on each brain region, p < 0.05) represents the sensitivity of the brain region to seizures.

[0120] Construction of the sensitivity T-Map: In the discovery set cases, the epileptogenic lesion ROI after registration to the MNI template was used as the seed region. The average time series of each seed region was calculated in 400 healthy subjects in the normal control group. The functional connectivity between the corresponding region and all cortical voxels was calculated using the Pearson correlation coefficient, and the Fisher Z transformation was used to ensure that the Pearson correlation coefficient conforms to the normal distribution. Finally, 400 groups of functional connectivity maps were obtained for each epileptogenic lesion. The above functional connectivity maps were integrated into a network diagram using a one-sample T-test, and they were superimposed and the voxels that appeared in 50% of the patients were calculated. The average brain functional connectivity of the overlapping area was retained to obtain the sensitivity T-Map of the epileptogenic lesion.

[0121] Specific T-Map construction: Similar to the construction of sensitive T-Map, the functional connectivity map related to the epileptogenic focus was obtained in the discovery set of cases. Similarly, for the non-epileptogenic focus group, the mean time series within the mapped seed region of the non-epileptogenic focus was calculated for functional connectivity with all cortical voxels of healthy subjects in the normal control group. The z-scores of the functional connectivity of each focus with the normal dataset were merged to obtain the functional connectivity map related to the non-epileptogenic focus. A two-sample t-test was used to compare the functional connectivity maps between the epileptogenic focus group and the non-epileptogenic focus group, and FDR correction was performed. Voxels with p < 0.05 were selected to establish the T-Map.

[0122] Epileptic focus-cortex epilepsy network construction: The specific T-Map and the sensitive T-Map were overlapped, and the overlapping area was retained. According to the positive and negative T values, the overlapping voxels were divided into positive connection regions and negative connection regions. Based on a large number of healthy subjects, the positive functional connectivity between the positive connection voxels and all cortical voxels, and the negative functional connectivity between the negative connection voxels and all cortical voxels were calculated respectively, and the two were superimposed to obtain the epileptic focus-cortex epilepsy network. In the standard space, according to the distribution of the epileptic focus-cortex epilepsy network, the positive epilepsy network weight and the negative epilepsy network weight of the overlapping part of the focus and this network were calculated respectively and added to obtain the damage intensity, that is, the risk score. The higher the score, the greater the possibility that the cortical focus is an epileptogenic focus. The risk score can be simply understood as the intensity of the damaged voxels overlapping with the epilepsy network.

[0123] Establishment of a prediction model based on the functional network

[0124] Based on epileptic and non-epileptic patients in the training dataset, a binary classification model was developed using machine learning techniques to predict the classification probability of each lesion. Initially, the mean time series of the lesion region was calculated based on the fMRI data of the rs fMRI dataset. Subsequently, the functional connectivity between the lesion region and all brain voxels was calculated and then Fisher z-transformed. To reduce the dimension, connections with zero values were excluded.

[0125] The functional connectivity of sub-patients with epileptogenic foci was designated as positive samples, while the connections of sub-patients with non-epileptogenic foci were labeled as negative samples. Principal component analysis (PCA) was used to further reduce the dimension. PCA uses orthogonal transformation to convert related variables into a set of linearly uncorrelated variables, called principal components. This transformation combines multiple original indicators into a new set of independent comprehensive indicators, retaining as much information of the original variables as possible. PCA helps to reduce the computational complexity, eliminate noise and irrelevant features, and improve the usability and interpretability of the dataset.

[0126] The features obtained after PCA dimensionality reduction are used as the input of the random forest classifier to predict the classification probability of each lesion. The random forest is a classifier composed of multiple decision trees and is selected for its high efficiency, robustness to high-dimensional data, avoidance of overfitting, and ability to effectively handle imbalanced datasets. Subsequently, for each individual lesion, the model generated a probability score indicating the probability of the epileptogenic focus based on the functional brain network.

[0127] Establishment of the cortical epileptogenic focus prediction fusion model

[0128] To establish a more accurate epileptogenic focus probability model, we combined the risk scores affected by the overlap and location of epilepsy network lesions with the classification probabilities generated by the prediction model, reflecting the functional characteristics of the lesion area. This fusion method takes into account both structural and functional features, improving the prediction accuracy. Initially, to standardize the risk scores, we selected the 97.5% of its distribution and truncated any value exceeding 1, converting it into a standardized score within the range of 0 to 1. Then, the fusion was performed based on the consistency of the standardized scores and classification probabilities of the two models. If both the standardized score and the classification probability are below the threshold, the minimum value is selected as the final probability.

[0129] Verification of the cortical epileptogenic focus prediction fusion model

[0130] For the cortical lesion ROIs of 50 TSC-related epilepsy patients in the validation set, according to the established distribution of the lesion-cortical epilepsy network, the cortical lesion risk scores were calculated. Furthermore, the relevant clinical, PET-CT, EEG and other characteristics of the patients were input into the selected machine learning model to obtain the epileptogenicity scores of each lesion. The results obtained from the lesion-cortical epilepsy network and the machine learning model were input into the fusion model to evaluate the epileptogenicity of the cortical lesions and determine the epileptogenic foci. The actual outcomes of 50 patients in the validation set who were seizure-free after surgery and 15 patients who still had seizures after surgery were used as the gold standard for the determination accuracy of epileptogenic foci. The actually resected cortical lesions of patients without seizures after surgery were the true epileptogenic foci, while the un-resected lesions were the true non-epileptogenic foci; for patients with seizures after surgery, the resected lesions were the true non-epileptogenic foci, and the epileptogenicity of the un-resected lesions could not be determined.

[0131] (1) We processed each multi-lesion related epilepsy patient as multiple single-lesion "sub-patients".

[0132] (2) We located the epileptogenic zone (epileptogenic focus) relying only on MRI, without relying on clinical symptoms and electroencephalogram. Clinically, the determination of epileptogenic foci requires electroencephalogram during seizures, which is usually difficult to obtain. This method can evaluate the epileptogenic foci only through the information of structural images, enabling early localization of epileptogenic foci and early intervention for epilepsy.

[0133] (3) We established a risk score for the epileptogenic focus based on the lesion-related network, used the brain network and machine learning to establish a prediction model, and finally formed a fusion model.

[0134] (4) This fusion model takes into account both the epileptic structural network and the functional network, and combines the multi-lesion characteristics of the patients.

[0135] (5) The application of this model is not restricted by units, MRI equipment types, etc., and has a wide range of applications.

[0136] This method is based on the MRI data of existing TSC-related epilepsy patients and the rs-fMRI data of 400 existing healthy subjects. According to the postoperative epilepsy control situation, epileptogenic lesions and non-epileptogenic lesions are extracted and distinguished, and methods such as lesion network mapping technology, machine learning, and ensemble learning are used to construct a lesion-cortical epilepsy network, a machine learning model, and an epileptogenic lesion prediction fusion model; currently, the epileptogenic lesion is determined by the epilepsy control outcome after the patient's resection surgery, and the accuracy of the fusion model in predicting the epileptogenic lesion is verified.

[0137] The purpose of this study is to establish a fusion model, establish a non-invasive, simple, easy-to-implement, and accurate method for determining epileptogenic lesions based on conventional MRI, and it can be used for multi-focal epilepsy such as multiple encephalomalacia and multiple cavernous hemangiomas. Reduce the surgical treatment gap for TSC-related epilepsy and other multi-focal epilepsy, reduce the usage rate of invasive examinations and invasive treatments, and improve the clinical efficacy and quality of life of patients.

[0138] This application also provides an MRI-based multi-focal epilepsy localization device, and the MRI-based multi-focal epilepsy localization device includes:

[0139] A lesion-cortical epilepsy network acquisition module, which is used to acquire a trained lesion-cortical epilepsy network;

[0140] A classification model acquisition module, which is used to obtain a trained classification model;

[0141] A module for acquiring information of the image to be predicted, which is used to acquire information of the image to be predicted;

[0142] An image feature acquisition module, which is used to extract the image features of the information of the image to be predicted;

[0143] A module for obtaining the result of the lesion-cortical epilepsy network, which is used to input the image features into the trained lesion-cortical epilepsy network to obtain the result of the lesion-cortical epilepsy network;

[0144] A prediction result acquisition module, which is configured to input the image features into the trained classification model to obtain a prediction result;

[0145] A final prediction result acquisition module, which is configured to obtain a final prediction result according to the lesion-cortical epilepsy network result and the prediction result

[0146] The above explanation of the method also applies to the explanation of the device.

[0147] Figure 2 It is a block diagram of the structure of an electronic device provided by one or more embodiments of the present invention.

[0148] As Figure 2 shown, the present application also discloses an electronic device (i.e., the general controller in the present application), including: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus; a computer program is stored in the memory. When the computer program is executed by the processor, the processor is caused to execute the steps of the multi-focal epilepsy localization method based on MRI.

[0149] The present application also provides a computer-readable storage medium, which stores a computer program executable by an electronic device. When the computer program runs on the electronic device, it can implement the steps of the multi-focal epilepsy localization method based on MRI.

[0150] The communication bus mentioned in the above electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity, only a thick line is used in the figure to represent it, but it does not mean that there is only one bus or one type of bus.

[0151] The electronic device includes a hardware layer, an operating system layer running on top of the hardware layer, and an application layer running on the operating system. The hardware layer includes hardware such as a central processing unit (CPU, Central Processing Unit), a memory management unit (MMU, Memory Management Unit), and memory. The operating system can be any one or more computer operating systems that implement the control of the electronic device through processes. For example, Linux operating system, Unix operating system, Android operating system, iOS operating system, or Windows operating system, etc. And in the embodiments of the present invention, the electronic device can be a handheld device such as a smart phone or a tablet computer, or an electronic device such as a desktop computer or a portable computer. The embodiments of the present invention do not particularly limit this.

[0152] The execution subject of the electronic device control in the embodiments of the present invention can be the electronic device, or a functional module in the electronic device that can call and execute a program. The electronic device can obtain the firmware corresponding to the storage medium, and the firmware corresponding to the storage medium is provided by the supplier. The firmware corresponding to different storage media can be the same or different, and this is not limited here. After the electronic device obtains the firmware corresponding to the storage medium, it can write the firmware corresponding to the storage medium into the storage medium. Specifically, it burns the firmware corresponding to the storage medium into the storage medium. The process of burning the firmware into the storage medium can be implemented by existing technologies and will not be elaborated in the embodiments of the present invention.

[0153] The electronic device can also obtain the reset command corresponding to the storage medium, and the reset command corresponding to the storage medium is provided by the supplier. The reset commands corresponding to different storage media can be the same or different, and this is not limited here.

[0154] At this time, the storage medium of the electronic device is the storage medium written with the corresponding firmware. The electronic device can respond to the reset command corresponding to the storage medium in the storage medium written with the corresponding firmware. Thus, the electronic device resets the storage medium written with the corresponding firmware according to the reset command corresponding to the storage medium. The process of resetting the storage medium according to the reset command can be implemented by existing technologies and will not be elaborated in the embodiments of the present invention.

[0155] For the convenience of description, when describing the above device, it is divided into various units and modules according to functions for description. Of course, when implementing the present application, the functions of each unit and module can be implemented in the same or multiple software and / or hardware.

[0156] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the field to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless specifically defined.

[0157] For method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the described order of actions, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0158] From the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multifocal epilepsy localization method based on MRI, characterized in that: The MRI-based multifocal epilepsy localization method comprises: Obtaining a trained lesion-cortical epilepsy network; According to the trained classification model; Obtaining image information to be predicted; Extracting image features of the image information to be predicted; Inputting the image features into the trained lesion-cortical epilepsy network to obtain lesion-cortical epilepsy network results; Inputting the image features into the trained classification model to obtain prediction results; Obtaining a final prediction result according to the lesion-cortical epilepsy network result and the prediction result; The lesion-cortical epilepsy network is established by the following method: Obtain the sensitivity T-Map of the epileptogenic lesion; Obtain specific T-Map of epileptogenic lesions; The sensitivity T-Map and the specificity T-Map were overlapped, and the overlapped areas were retained. The overlapped voxels were divided into positive connection areas and negative connection areas according to the positive and negative T values. The positive functional connections between the positively connected voxels and all the cortical voxels, and the negative functional connections between the negatively connected voxels and all the cortical voxels were calculated respectively. The positive functional connections and the negative functional connections were superimposed to obtain the lesion-cortical epilepsy network. The method of obtaining the sensitivity T-Map of the epileptogenic lesion comprises: Acquire a T1 image group, wherein the T1 image group includes a plurality of T1 images, and one T1 image corresponds to one patient, wherein the patient includes a TSC-related epilepsy patient who has completed resective surgery and a patient who has been seizure-free for 6 months after surgery; Acquire a T2-Flair image group, wherein the T2-Flair image group includes a plurality of T2-Flair images, and one T2-Flair image corresponds to one patient, wherein the patient includes a TSC-related epilepsy patient who has completed resective surgery and a patient who has been seizure-free for 6 months after surgery, and each T2-Flair image includes an epileptogenic focus ROI; Acquire an EPI image group, wherein the EPI image group includes a plurality of EPI images, and one EPI image corresponds to one healthy person; Registering the T1 image group with the T2-Flair image group to obtain a first MNI template; Normalizing the EPI image group to the first MNI template, thereby obtaining a second MNI template; The epileptogenic lesion ROI in the second MNI template was used as the seed region, and the average time series of each seed region in the EPI image was calculated. The functional connection between the seed region and all cortical voxels was calculated using the Pearson correlation coefficient, and the Fisher Z transformation was used to ensure that the Pearson correlation coefficient conformed to the normal distribution; each epileptogenic lesion ROI would obtain the same number of functional connection maps as the EPI image, and the above functional connection maps were integrated into a network map using a single-sample T test, which was superimposed and the voxels that appeared in 50% of the patients were calculated, and the average brain functional connection of the overlapping area was retained to obtain the sensitivity T-Map of the epileptogenic lesion; The method of obtaining a specific T-Map of an epileptogenic focus comprises: Obtaining the functional connectivity map of the epileptogenic nodule group and the functional connectivity map of the non-epileptic nodule group; The two-sample T test was used to compare the functional connectivity maps of the epileptogenic nodule group and the non-epileptic nodule group, and FDR correction was performed. Voxels with p < 0.05 were selected to establish the T-Map.

2. The MRI-based multifocal epilepsy localization method according to claim 1, characterized in that: The epileptogenic lesion ROI on the T2-Flair image is obtained by the following method: Acquire the first cortical lesion region of interest and the second cortical lesion region of interest of each T2-Flair image; Calculating the overlap in cortical lesion identification of the first cortical lesion region of interest and the second cortical lesion region of interest of each T2-Flair image; The overlap degree in the cortical lesion identification of each T2-Flair image is judged as follows: whether the overlap degree is greater than a preset threshold is judged, and if so, the overlapping area is regarded as the epileptogenic lesion ROI.

3. The MRI-based multifocal epilepsy localization method according to claim 2, characterized in that: The registering the T1 image group with the T2-Flair image group to obtain a first MNI template includes: The T2-Flair image of each patient was registered to the T1 image, and then each epileptogenic lesion ROI was registered to the T1 image and finally standardized to the MNI template.

4. The MRI-based multifocal epilepsy localization method according to claim 1, characterized in that: The trained classification model is obtained by the following method: Acquire a training set, wherein the training set includes images of epileptic patients and images of non-epileptic patients; The average time series of the lesion area was calculated based on the images of non-epileptic patients in the training set. Subsequently, the functional connectivity between the lesion area and the whole brain voxels was calculated, and then Fisher z transformation was performed to obtain the functional connectivity matrix. The functional connections of the sub-patients whose lesions were epileptogenic lesions were designated as positive samples, while those of the sub-patients whose lesions were non-epileptogenic lesions were labeled as negative samples; The functional connection matrix is ​​reduced in dimension by principal component analysis to obtain characteristic information; Get a random forest classifier; The random forest classifier is trained through feature information to obtain hyperparameters of the random forest model.

5. A multifocal epilepsy localization device based on MRI, characterized in that: The MRI-based multifocal epilepsy localization device comprises: A lesion-cortical epilepsy network acquisition module, wherein the lesion-cortical epilepsy network acquisition module is used to acquire a trained lesion-cortical epilepsy network; A classification model acquisition module, wherein the classification model acquisition module is used to obtain a classification model according to a trained classification model; A module for acquiring information of an image to be predicted, wherein the module is used to acquire information of an image to be predicted; An image feature acquisition module, the image feature acquisition module is used to extract image features of the image information to be predicted; A lesion-cortical epilepsy network result acquisition module, wherein the lesion-cortical epilepsy network result acquisition module is used to input the image features into the trained lesion-cortical epilepsy network, thereby acquiring a lesion-cortical epilepsy network result; A prediction result acquisition module, wherein the prediction result acquisition module is used to input the image features into the trained classification model to obtain a prediction result; A final prediction result acquisition module, the final prediction result acquisition module is used to acquire a final prediction result according to the lesion-cortical epilepsy network result and the prediction result; The lesion-cortical epilepsy network is established by the following method: Obtain the sensitivity T-Map of the epileptogenic lesion; Obtain specific T-Map of epileptogenic lesions; The sensitivity T-Map and the specificity T-Map were overlapped, and the overlapped areas were retained. The overlapped voxels were divided into positive connection areas and negative connection areas according to the positive and negative T values. The positive functional connections between the positively connected voxels and all the cortical voxels, and the negative functional connections between the negatively connected voxels and all the cortical voxels were calculated respectively. The positive functional connections and the negative functional connections were superimposed to obtain the lesion-cortical epilepsy network. The method of obtaining the sensitivity T-Map of the epileptogenic lesion comprises: Acquire a T1 image group, wherein the T1 image group includes a plurality of T1 images, and one T1 image corresponds to one patient, wherein the patient includes a TSC-related epilepsy patient who has completed resective surgery and a patient who has been seizure-free for 6 months after surgery; Acquire a T2-Flair image group, wherein the T2-Flair image group includes a plurality of T2-Flair images, and one T2-Flair image corresponds to one patient, wherein the patient includes a TSC-related epilepsy patient who has completed resective surgery and a patient who has been seizure-free for 6 months after surgery, and each T2-Flair image includes an epileptogenic focus ROI; Acquire an EPI image group, wherein the EPI image group includes a plurality of EPI images, and one EPI image corresponds to one healthy person; Registering the T1 image group with the T2-Flair image group to obtain a first MNI template; Normalizing the EPI image group to the first MNI template, thereby obtaining a second MNI template; The epileptogenic lesion ROI in the second MNI template was used as the seed region, and the average time series of each seed region in the EPI image was calculated. The functional connection between the seed region and all cortical voxels was calculated using the Pearson correlation coefficient, and the Fisher Z transformation was used to ensure that the Pearson correlation coefficient conformed to the normal distribution; each epileptogenic lesion ROI would obtain the same number of functional connection maps as the EPI image, and the above functional connection maps were integrated into a network map using a single-sample T test, which was superimposed and the voxels that appeared in 50% of the patients were calculated, and the average brain functional connection of the overlapping area was retained to obtain the sensitivity T-Map of the epileptogenic lesion; The method of obtaining a specific T-Map of an epileptogenic focus comprises: Obtaining the functional connectivity map of the epileptogenic nodule group and the functional connectivity map of the non-epileptic nodule group; The two-sample T test was used to compare the functional connectivity maps of the epileptogenic nodule group and the non-epileptic nodule group, and FDR correction was performed. Voxels with p < 0.05 were selected to establish the T-Map.

Images