MRI-Based Radiomics Approach for Differentiating Juvenile Myoclonic Epilepsy from Epilepsy with Generalized Tonic-Clonic Seizures Alone
Abstract number :
1.459
Submission category :
5. Neuro Imaging / 5A. Structural Imaging
Year :
2023
Submission ID :
1258
Source :
www.aesnet.org
Presentation date :
12/2/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Kyung Min Kim, MD – Yonsei University College of Medicine
Soomi Cho, MD – Yonsei University College of Medicine; Bo Kyu Choi, MD – Yonsei University College of Medicine; Seungwon Song, MD – Yonsei University College of Medicine; Jungyon Yum, MD – Neurology – Yonsei University College of Medicine
Rationale: This study focuses on two types of idiopathic generalized epilepsy: juvenile myoclonic epilepsy (JME) and epilepsy with generalized tonic-clonic seizures Alone (GTCA). While conventional MRI scans usually show no abnormalities, recent advanced MRI analyses have revealed structural and functional changes in JME & GTCA patients. The study aims to develop and validate MRI-based radiomics models that can effectively distinguish between JME and GTCA, offering valuable insights for clinical decision-making and prognosis.
Methods: A total of 164 patients (127 with JME and 37 with GTCA) with 3D T1-weighted images were enrolled. The patients were divided into training and test sets in a 7:3 ratio. From the 17 region-of-interest masks, which have clinical evidence of association with JME and GTCA, a total of 1,581 radiomics features were extracted. Various machine-learning combinations of oversampling, feature selection, and classification algorithms were explored to develop the optimal radiomics classification model using the training set. The performance of the best radiomics model was evaluated in the test set by measuring the area under the receiver operating curve (AUC), accuracy, sensitivity, and specificity. SHapley Additive exPlanations (SHAP) analysis was performed to estimate the contribution of each radiomics feature to the prediction of the classification model.
Results: The best radiomics classification model was developed using a machine-learning combination of Synthetic Minority Oversampling Technique-Edited Nearest Neighbors, Least Absolute Shrinkage and Selection Operator, and eXtreme Gradient Boosting. The AUC of the best radiomics classification model was 0.861 (95% confidence interval [CI]: 0.737‒0.985) in the training set and 0.800 (95% CI: 0.617‒0.982) in the test set. At the optimal cutoff point of 0.476, the accuracy, sensitivity, and specificity of the best model were 80.0%, 81.8%, and 79.5%, respectively. SHAP analysis revealed that the texture features of the caudate, thalamus proper, cerebral white matter, and putamen had the highest importance in the best radiomics model.
Conclusions: MRI-based radiomics models can assist in differentiating between JME and GTCA.
Funding: No funding was used to support this study.
Neuro Imaging