An MRI radiomics-based epileptogenic tubercle localization system

Abstract

This invention relates to an epileptogenic nodule localization system based on MRI radiomics, belonging to the field of medical informatics. The system includes an image acquisition module, a delineation module, an extraction module, an analysis module, a modeling module, and an optimization module. It extracts radiomics features from the epileptogenic and non-epileptic nodule lesion regions of original MRI T2-Flair images of TSC patients. The preliminary radiomics features are preprocessed, optimized, analyzed, and screened to obtain differential radiomics features between epileptogenic and non-epileptic nodules. A predictive model is established using machine learning algorithms on the obtained differential radiomics features to calculate the radiomics feature score for each lesion, identifying epileptogenic nodules from cortical epileptogenic nodules. This achieves the goal of non-invasive identification of epileptogenic foci in TSC patients, assisting in clinical diagnosis and personalized surgical planning, and enabling precise treatment and surgical resection of epileptogenic foci in TSC patients.

Description

Technical Field

[0001] This invention belongs to the field of medical information technology and relates to an epileptogenic nodule localization system based on MRI radiomics. Background Technology

[0002] Tuberous sclerosis complex (TSC) is an autosomal dominant genetic syndrome. Epileptogenic nodules in the cortex are a typical feature of TSC patients, who often have multiple epileptogenic nodules in their brains, including both epileptogenic and non-epilepsy nodules. Previous studies have found that epilepsy in TSC patients originates from these epileptogenic nodules, and surgery is the most effective method for controlling seizures in TSC patients. Therefore, accurate preoperative localization of the epileptogenic nodules from all cortical epileptogenic nodules is crucial in determining the surgical approach, outcome, and prognosis for TSC-related epilepsy.

[0003] For the localization of epileptogenic nodes in TSC patients, clinical practice currently often employs a comprehensive surgical assessment method that combines imaging techniques such as MRI, CT, and PET with electrophysiological techniques such as scalp video EEG, subcutaneous EEG, and intracranial EEG.

[0004] Intracranial electrode implantation for EEG monitoring of different cortical epileptogenic nodules is the most reliable method for locating epileptogenic nodules in the epileptogenic cortex. However, intracranial EEG monitoring requires surgical implantation of recording electrodes into different epileptogenic nodules within the brain, which is an invasive procedure for patients, carries surgical risks, and is costly, placing a significant financial burden on some patients. With the increasing potential of radiomics in predicting epileptic foci, the use of non-invasive neuroimaging for early and accurate localization of epileptogenic nodules in TSC patients has crucial clinical value. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide an epileptogenic nodule localization system based on MRI radiomics. Based on MRI radiomics, a model for localizing epileptogenic nodules in TSC patients is established, enabling non-invasive preoperative localization of epileptogenic foci in TSC patients, thereby improving the epileptogenic prognosis and quality of life of TSC patients.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An epileptogenic nodule localization system based on magnetic resonance MRI radiomics, comprising an image acquisition module, a delineation module, an extraction module, an analysis module, a modeling module, and an optimization module;

[0008] The image acquisition module acquires raw MRI T2Flair images of patients with tuberous sclerosis syndrome (TSC) that meet the criteria for radiomics analysis, and preprocesses the acquired raw images.

[0009] The delineation module is used to delineate epileptogenic and non-epileptogenic nodules on MRI T2Flair images;

[0010] The extraction module performs radiomics feature extraction on images of epileptogenic nodules and non-epileptiform nodules.

[0011] The analysis module analyzes and filters the extracted radiomics features to obtain the target radiomics features;

[0012] The modeling module establishes a predictive model for the target image omics features that differ between epileptogenic and non-epileptiogenic nodules.

[0013] The optimization module divides epileptogenic nodules into training and validation sets, trains and validates the prediction model, optimizes the model parameters, and obtains the final model.

[0014] Optionally, the original MRI T2Flair images that conform to radiomics analysis are the original preoperative MRI images of TSC patients who have completed surgery or preoperative comprehensive assessment, specifically including epileptic EEG seizures monitored by video EEG, lesion structural changes clearly identified by MRI and PET imaging, and TSC1 / 2 gene mutations confirmed by genetic testing.

[0015] The preprocessing of the acquired raw MRI images is as follows: the acquired raw MRI images are divided into training set and test set, and a prediction model is built based on the training set data; raw MRI images with clear images and less interference are selected. Since cortical epileptogenic nodules present high signal on MRI T2Flair images, cortical nodules with high signal on T2Flair images of TSC patients are selected for processing and analysis. This "high" refers to calcification. High signal is a term that can only be expressed in three ways: high, low, and isodense signal.

[0016] Optionally, the delineation module is used to delineate epileptogenic and non-epilepsy nodules on T2Flair images, specifically as follows:

[0017] Based on the comprehensive evaluation results of epileptogenic nodules in TSC patients before, during and after the operation, the localization of epileptogenic and non-epileptic nodules was clarified. The training set of MRI T2Flair images of patients was imported into the 3DSlicer software for processing and target segmentation. Regions of interest (ROIs) of epileptogenic and non-epileptic nodules were delineated and set as epileptogenic nodule ROIs and non-epileptic nodule ROIs, respectively.

[0018] Optionally, the radiomics feature extraction of epileptogenic nodules and non-epileptogenic nodules specifically involves: running the Pyradiomics package on the Python platform to extract the radiomics features of epileptogenic nodule ROIs and non-epileptogenic nodule ROIs respectively.

[0019] Optionally, the analysis and screening of the extracted radiomics features specifically involves: using the univariate feature selection method ANOVA, minimum absolute value convergence and selection operator LASSO regression for screening, and selecting effective radiomics features with non-zero coefficients that can distinguish between epileptogenic nodules and non-epileptic nodules, i.e., target radiomics features.

[0020] Optionally, the establishment of a predictive model for the target radiomics features that differ between epileptogenic and non-epilepsy nodules specifically involves: establishing a radiomics score calculation formula for epileptogenic nodules using Logistic regression and machine algorithms: Rad-score = -0.021875(T2Flair_wavelet-LHL_firstorder_Mean) - 0.005586(T2Flair_wavelet-LHH_glcm_Imc1) - 0.004722(T2Flair_wavelet-LLH_glcm_Imc1) - 0.071362(T2Flair_wavelet-HLL_glcm_MCC) + 0.000532(T2Flair_wavelet-HLL_glcm_ClusterShade)

[0021] -0.032400(T2Flair_wavelet-HLH_glszm_SizeZoneNonUniformity)-0.000319(T2Flair_wavelet-HL H_glrlm_GrayLevelNonUniformityNormalized)-0.069577(T2Flair_square_glszm_SmallAreaEmph asis)-0.002988(T2Flair_wavelet-HHH_glrlm_LowGrayLevelRunEmphasis);

[0022] Wherein, Rad-score is the radiomics score; T2Flair_wavelet-LHL_firstorder_Mean is the mean of the first-order gray-level features of the T2Flair wavelet LHL; T2Flair_wavelet-LHH_glcm_Imc1 is the T2Flair wavelet LHH gray-level co-occurrence matrix Imc1; T2Flair_wavelet-LLH_glcm_Imc1 is the T2Flair wavelet LLH gray-level co-occurrence matrix Imc1; T2Flair_wavelet-HLL_glcm_MCC is the T2Flair wavelet HLL gray-level co-occurrence matrix MCC; T2Flair_wavelet-HLL_glcm_ClusterShade is the T2Flair wavelet HLL gray-level co-occurrence matrix ClusterShade; T2Flair_wavelet-HLH_glszm_Siz eZoneNonUniformity is the T2Flair wavelet HLH grayscale and band matrix SizeZoneNonUniformity; T2Flair_wavelet-HLH_glrlm_GrayLevelNonUniformityNormalized is the T2Flair wavelet HLH grayscale run-length matrix GrayLevelNonUniformityNormalized; T2Flair_square_glszm_SmallAreaEmphasis is the T2Flair square grayscale and band matrix SmallAreaEmphasis; T2Flair_wavelet-HHH_glrlm_LowGrayLevelRunEmphasis is the T2Flair wavelet HHH grayscale run-length matrix LowGrayLevelRunEmphasis;

[0023] A Rad-score greater than 0 in radiomics is considered an epileptogenic nodule; a Rad-score less than 0 in radiomics is considered a non-epilepsogenic nodule.

[0024] Optionally, the training and validation of the prediction model specifically involves: inputting the radiomics features of the test set into the prediction model, evaluating the model using ROC curves, testing the accuracy of the prediction model, and optimizing the prediction model parameters based on the test results to continuously improve the accuracy of the trained model in predicting epileptogenic nodules; for newly admitted TSC patients, using the prediction model to predict the patient's epileptogenic nodules based on their MRI T2Flair images; combining clinical comprehensive preoperative assessment methods to evaluate the accuracy of the epileptogenic nodules predicted by the prediction model, and dynamically adjusting the radiomics features and coefficients of each radiomics feature included in the radiomics score to achieve a prediction model accuracy of 90%.

[0025] The beneficial effects of this invention are as follows: By extracting radiomics features from the epileptogenic and non-epileptic nodule lesion areas of the original MRI T2Flair images of TSC patients, the preliminary radiomics features are preprocessed, optimized, analyzed, and screened to obtain differential radiomics features between epileptogenic and non-epileptic nodules. A predictive model is established using machine learning algorithms on the obtained differential radiomics features to calculate the radiomics feature score for each lesion, identifying epileptogenic nodules from cortical epileptogenic nodules, achieving the goal of non-invasive identification of epileptogenic foci in TSC patients, assisting in clinical diagnosis and personalized surgical planning, and enabling precise treatment and surgical resection of epileptogenic foci in TSC patients. The clinical application and promotion of this invention is expected to reduce the invasive damage caused by implanting intracranial electrodes to locate epileptogenic nodules and greatly reduce the economic burden on patients, possessing significant clinical value and promising prospects.

[0026] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0027] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0028] Figure 1 This is a system diagram of the present invention;

[0029] Figure 2 For LASSO regression model;

[0030] Figure 3 For MSE plot;

[0031] Figure 4 This is the ROC curve. Detailed Implementation

[0032] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0033] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0034] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0035] The system of this invention is divided into the following six modules: image acquisition module, delineation module, extraction module, analysis module, modeling module, and optimization module. The system structure diagram is shown below. Figure 1 As shown.

[0036] 1. Image acquisition module: Acquires raw MRI T2Flair images and clinical information of TSC patients that meet the criteria for radiomics analysis, and preprocesses the acquired images.

[0037] 2. Delineation Module: Delineates epileptogenic and non-epileptic nodules on T2Flair images, and sets them as epileptogenic nodule ROIs and non-epileptic nodule ROIs, respectively.

[0038] 3. Extraction Module: Extracts radiomics features from images of epileptogenic and non-epilepsy nodules.

[0039] 4. Analysis Module: Analyzes and filters the extracted radiomics features to obtain the target radiomics features.

[0040] 5. Modeling Module: Establishes predictive models for the radiomic features of target nodules that differ between epileptogenic and non-epilepsy nodules.

[0041] 6. Optimization Module: The epileptogenic nodules are divided into training and validation sets to train and validate the prediction model, optimize various parameters of the model, and obtain an effective final model.

[0042] The connection and data flow of each module are described below:

[0043] 1. Image Acquisition Module—Acquiring T2-Flair MRI Images of TSC Patients: Collecting raw preoperative MRI images of TSC patients who have undergone surgery or completed preoperative comprehensive evaluation. All acquired data are divided into training and testing sets. Predictive modeling is based on the training set data. Raw TSC patient images with better image quality are selected. Since cortical epileptogenic nodules typically present high signal intensity on T2-Flair images, raw T2-Flair images of TSC patients are selected for further processing and analysis.

[0044] 2. Delineation Module—Delineating epileptogenic and non-epileptic nodules on T2Flair images: Based on the comprehensive evaluation results of epileptogenic nodules in the cortex of TSC patients before, during, and after surgery, the localization of epileptogenic and non-epileptic nodules is clarified. The training set of patient MRI T2Flair images is imported into 3DSlicer software for processing and target area segmentation, and the regions of interest (ROIs) of epileptogenic and non-epileptic nodules are delineated, which are defined as epileptogenic nodule ROIs and non-epileptic nodule ROIs, respectively.

[0045] 3. Extraction Module—Radiomics Feature Extraction of Epileptogenic and Non-Epileptic Nodule Images: Using the Pyradiomics package on the Python platform, radiomics features of epileptogenic nodule ROIs and non-epilepsy nodule ROIs are extracted respectively.

[0046] 4. Analysis Module—Optimization and Statistical Analysis of Extracted Data: For all extracted radiomics features, the univariate feature selection method (ANOVA) and the least absolute value convergence and selection operator regression (LASSO regression) are used to screen and select effective radiomics features (coefficients not zero) that can distinguish between epileptogenic nodules and non-epilepsy nodules, i.e., target radiomics features.

[0047] 5. Modeling Module—Establishing Predictive Models for Target Radiomic Features of Epileptogenic and Non-Epileptic Nodules: Based on the target radiomic features of epileptogenic and non-epilepsy nodules, an epileptogenic nodule prediction model is established through Logistic regression and machine algorithms.

[0048] 6. Optimization Module—Training, Testing, and Optimizing the Predictive Model: The predictive model is built based on the radiomics features of the training set. The radiomics features of the test set are then incorporated into the predictive model, and the model is evaluated using ROC curves to test its accuracy. Based on the test results, the parameters of the predictive model are optimized to achieve a more stable training state. Newly admitted TSC patients are included, and their epileptogenic nodules are predicted using the predictive model based on their MRI T2-FLAIR images. Combined with comprehensive preoperative clinical assessment methods, the accuracy of the predicted epileptogenic nodules is evaluated, and the various parameters of the predictive model are continuously optimized to improve the prediction accuracy and achieve the clinical application effect of auxiliary diagnosis.

[0049] Example:

[0050] 1. Obtain MRI T2Flair images of TSC patients: Collect raw preoperative MRI images of 32 TSC patients who had completed surgery or preoperative comprehensive assessment (a total of 49 epileptogenic nodules and 114 non-epileptogenic nodules were obtained).

[0051] 2. Preprocessing of acquired imaging images: Original imaging data of TSC patients with good image quality were selected. Since epileptogenic nodules in the cortex usually present high signal on T2-Flair images, original T2-Flair images of TSC patients were selected for further processing and analysis. Ultimately, 44 epileptogenic nodules and 108 non-epileptic nodules were selected for further research. All data were divided into a training set (33 epileptogenic nodules, 81 non-epileptic nodules) and a test set (11 epileptogenic nodules, 27 non-epileptic nodules) at a 3:1 ratio. Predictive modeling was performed based on the training set data.

[0052] 3. Delineate epileptogenic and non-epileptic nodules on T2Flair images: Based on the comprehensive evaluation results of epileptogenic nodules in the cortex of TSC patients before, during, and after surgery, clarify the localization of epileptogenic and non-epileptic nodules. Use 3DSlicer software to import the training set of patients' MRI T2Flair images for processing and target segmentation, and delineate the regions of interest (ROIs) of epileptogenic and non-epileptic nodules, which are respectively set as the ROI of epileptogenic nodules and the ROI of non-epileptic nodules.

[0053] 4. Radiomics feature extraction of epileptogenic and non-epileptic nodule images: The Pyradiomics package was run on the Python platform to extract radiomics features of epileptogenic nodule ROI and non-epileptic nodule ROI respectively.

[0054] 5. Optimization and statistical analysis of the extracted data: For all extracted radiomics features, LASSO regression was used for screening to identify 85 radiomics features most relevant to epileptogenic nodules (19 first-order grayscale features, 26 shape features, 40 texture features (Gray-Level Co-occurrence Matrix—GLCM, Gray-Level Run-Length Matrix—GLRLM), and 8 wavelet transform features), i.e., target radiomics features, such as... Figure 2 and Figure 3 As shown.

[0055] 6. Establish a predictive model for radiomic features that differ between epileptogenic and non-epilepsy nodules: Based on the target radiomic signs of epileptogenic and non-epilepsy nodules, a radiomic score calculation formula for epileptogenic nodules is established using Logistic regression and machine learning algorithms: Rad-score = -0.021875(T2Flair_wavelet-LHL_firstorder_Mean)-0.005586(T2Flair_wavelet-LHH_glcm_Imc1)-0.004722(T2Flair_wavelet-LLH_glcm_Imc1)-0.071362

[0056] (T2Flair_wavelet-HLL_glcm_MCC)+0.000532

[0057] (T2Flair_wavelet-HLL_glcm_ClusterShade)-0.032400(T2Flair_wavelet-HLH_glszm_SizeZoneNo nUniformity)-0.000319(T2Flair_wavelet-HLH_glrlm_GrayLevelNonUniformityNormalized)-0.069577(T2Flair_square_glszm_SmallAreaEmphasis)0.002988(T2Flair_wavelet-HHH_glrlm_LowGray LevelRunEmphasis).

[0058] A Rad-score greater than 0 is considered an epileptogenic nodule; a Rad-score less than 0 is considered a non-epilepsogenic nodule.

[0059] 7. Training and testing the prediction model: The prediction model was established based on the radiomics features of the training set obtained in the early stage, with an AUC of 0.79; the radiomics features of the test set were brought into the prediction model to test the accuracy of the prediction model in predicting epileptogenic nodules of the test set, with an AUC of 0.78, indicating that the prediction model is relatively stable.

[0060] 8. Optimize model parameters to obtain an effective final model: Five newly admitted TSC patients (7 epileptogenic nodules and 13 non-epileptogenic nodules) were included. Based on their MRI T2-FLAIR images, a predictive model was used to predict the epileptogenic nodules. The predictive model parameters were optimized, and the optimized model AUC was 0.86, effectively playing an auxiliary diagnostic role in the localization of epileptogenic nodules in clinical TSC patients. Figure 4 As shown.

[0061] The above description is only a preferred embodiment of the present invention. For relevant clinical medical workers and researchers in the field, improvements and modifications can be made based on the technical principles of the present invention, and these improvements and modifications should be considered to be within the protection scope of the present invention.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A system for locating epileptogenic nodules based on magnetic resonance MRI radiomics, characterized in that: The system includes an image acquisition module, a delineation module, an extraction module, an analysis module, a modeling module, and an optimization module; The image acquisition module acquires raw MRI T2Flair images of patients with tuberous sclerosis syndrome (TSC) that meet the criteria for radiomics analysis, and preprocesses the acquired raw images. The delineation module is used to delineate epileptogenic and non-epileptogenic nodules on MRI T2Flair images; The extraction module performs radiomics feature extraction on images of epileptogenic nodules and non-epileptiform nodules. The analysis module analyzes and filters the extracted radiomics features to obtain the target radiomics features; The modeling module establishes a predictive model for the target image omics features that differ between epileptogenic and non-epileptiogenic nodules. The optimization module divides epileptogenic nodules into a training set and a validation set, trains and validates the prediction model, optimizes the model parameters, and obtains the final model. The specific steps for analyzing and screening the extracted radiomics features are as follows: using the univariate feature selection method ANOVA, minimum absolute value convergence and selection operator LASSO regression to screen and select effective radiomics features with non-zero coefficients that can distinguish between epileptogenic nodules and non-epileptic nodules, i.e. target radiomics features. The specific steps for establishing a predictive model for the target radiomics features that differentiate between epileptogenic and non-epilepsy nodules are as follows: Using Logistic regression and machine learning algorithms, a radiomics score calculation formula for epileptogenic nodules is established: Rad-score = -0.021875(T2Flair_wavelet-LHL_firstorder_Mean)-0.005586(T2Flair_wavelet-LHH_glcm_Imc1) -0.004722(T2Flair_wavelet-LLH_glcm_Imc1)-0.071362(T2Flair_wavelet-HLL_glcm_MCC) + 0.000532(T2Flair_wavelet-HLL_glcm_ClusterShade) -0.032400(T2Flair_wavelet-HLH_glszm_SizeZoneNonUniformity)-0.000319(T2Flair_wavelet-HLH_glrlm_GrayLevelNonUniformityNormalized)-0.069577(T2Flair_square_glszm_SmallAreaEmphasis) -0.002988(T2Flair_wavelet-HHH_glrlm_LowGrayLevelRunEmphasis); Wherein, Rad-score is the image omics score; T2Flair_wavelet-LHL_firstorder_Mean is the mean of the first-order gray-level features of the T2Flair wavelet LHL; T2Flair_wavelet-LHH_glcm_Imc1 is the T2Flair wavelet LHH gray-level co-occurrence matrix Imc1; T2Flair_wavelet-LLH_glcm_Imc1 is the T2Flair wavelet LLH gray-level co-occurrence matrix Imc1; T2Flair_wavelet-HLL_glcm_MCC is the T2Flair wavelet HLL gray-level co-occurrence matrix MCC; T2Flair_wavelet-HLL_glcm_ClusterShade is the T2Flair wavelet HLL gray-level co-occurrence matrix ClusterShade; T2Flair_wavelet-HLH_glszm_Siz eZoneNonUniformity is the T2Flair wavelet HLH grayscale and band matrix SizeZoneNonUniformity; T2Flair_wavelet-HLH_glrlm_GrayLevelNonUniformityNormalized is the T2Flair wavelet HLH grayscale run-length matrix GrayLevelNonUniformityNormalized; T2Flair_square_glszm_SmallAreaEmphasis is the T2Flair square grayscale and band matrix SmallAreaEmphasis; T2Flair_wavelet-HHH_glrlm_LowGrayLevelRunEmphasis is the T2Flair wavelet HHH grayscale run-length matrix LowGrayLevelRunEmphasis; A Rad-score greater than 0 in radiomics is considered an epileptogenic nodule; a Rad-score less than 0 in radiomics is considered a non-epilepsogenic nodule.

2. The epileptogenic nodule localization system based on MRI radiomics according to claim 1, characterized in that: The original MRI T2Flair images that conform to radiomics analysis are the original preoperative MRI images of TSC patients who have completed surgery or preoperative comprehensive assessment. Specifically, they include epileptic EEG seizures monitored by video EEG, lesion structural changes clearly identified by MRI and PET imaging, and TSC1 / 2 gene mutations confirmed by genetic testing. The preprocessing of the acquired raw imaging images is as follows: the acquired raw imaging images are divided into training set and test set, and a prediction model is built based on the training set data; raw MRI images with clear images and less interference are selected. Since cortical epileptogenic nodules present high signal on MRI T2Flair images, cortical nodules with high signal on T2Flair images of TSC patients are selected for processing and analysis.

3. The epileptogenic nodule localization system based on MRI radiomics according to claim 2, characterized in that: The delineation module is used to delineate epileptogenic and non-epileptogenic nodules on T2Flair images, specifically as follows: Based on the comprehensive evaluation results of epileptogenic nodules in TSC patients before, during and after the operation, the localization of epileptogenic and non-epileptic nodules was clarified. The training set of MRI T2Flair images of patients was imported into the 3DSlicer software for processing and target segmentation. Regions of interest (ROIs) of epileptogenic and non-epileptic nodules were delineated and set as epileptogenic nodule ROIs and non-epileptic nodule ROIs, respectively.

4. The epileptogenic nodule localization system based on MRI radiomics according to claim 3, characterized in that: The specific steps for extracting radiomics features from images of epileptogenic nodules and non-epileptogenic nodules are as follows: using the Pyradiomics package on the Python platform, radiomics features of the ROIs of epileptogenic nodules and non-epileptogenic nodules are extracted respectively.

5. The epileptogenic nodule localization system based on MRI radiomics according to claim 1, characterized in that: The specific steps for training and validating the prediction model are as follows: The radiomics features from the test set are input into the prediction model, the model is evaluated using ROC curves, the accuracy of the prediction model is tested, and the parameters of the prediction model are optimized based on the test results to continuously improve the accuracy of the trained model in predicting epileptogenic nodules. For newly admitted TSC patients, the prediction model is used to predict the epileptogenic nodules based on their MRI T2Flair images. Combined with comprehensive preoperative clinical assessment methods, the accuracy of the epileptogenic nodules predicted by the prediction model is evaluated, and the radiomics features and coefficients of each radiomics feature included in the radiomics score are dynamically adjusted to achieve a prediction model accuracy of 90%.

Images