Abstract
OBJECTIVE: Focal cortical dysplasia (FCD) and dysembryoplastic neuroepithelial tumor (DNET) are two major causes of intractable epilepsy, often with confusing imaging findings. This study aimed to develop magnetic resonance imaging (MRI)-based radiomics models for preoperative differentiation between FCD and DNET. METHODS: This study included 169 patients who underwent epilepsy surgery and were pathologically diagnosed with FCD (n = 96) or DNET (n = 73). Conventional brain T1-weighted (T1WI), T2-weighted (T2WI) and T2-fluid-attenuated inversion recovery (T2-FLAIR) images were acquired from all cases. The whole dataset was randomly divided into a training set and a test set at a ratio of 3:1. PyRadiomics software was used for feature extraction and selection. The final features were determined using the least absolute shrinkage and selection operator (LASSO) algorithm. A support vector machine (SVM) was used to establish radiomics models based on individual sequences or fusion of these sequences. The performance of each model was evaluated by the area under the receiver operating characteristic curve (AUC), and the optimal model was also compared with the radiologists' assessment results. RESULTS: The fusion radiomics model exhibited the best performance in differentiating between FCD and DNET, with an AUC of 0.894 (95% CI: 0.799-0.968) and an accuracy of 82.0%, which were superior to the individual models based on T1WI, T2WI, or T2-FLAIR images. In addition, the diagnostic performance of the fusion radiomics model was superior to that of the junior radiologist and comparable to that of the senior radiologist. CONCLUSION: The fusion radiomics model based on multi-sequence MRI can successfully differentiate FCD from DNET preoperatively, which contributes to appropriate surgical planning and satisfactory treatment outcomes.