Detection of Epileptogenic Focal Cortical Dysplasia Using Graph Neural Networks: A MELD Study

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Epilepsy Research

Background:

• Focal cortical dysplasia (FCD) is a leading cause of drug-resistant epilepsy, often difficult to detect on MRI.
• Existing automated detection methods for FCD have limitations, including high false-positive rates, hindering clinical use.

Purpose of the Study:

• To assess the effectiveness and interpretability of graph neural networks for automated FCD detection in MRI scans.
• To compare the performance of a novel graph neural network (MELD Graph) against existing algorithms.

Main Methods:

• A multicenter study involving retrospective MRI data from 23 epilepsy centers.
• Training a graph neural network (MELD Graph) on surface-based MRI features from 20 centers, with independent testing on data from 3 centers.
• Evaluating performance using sensitivity, specificity, and positive predictive value (PPV), with feature analysis for interpretability.

Main Results:

• MELD Graph demonstrated improved PPV compared to a baseline algorithm in both test datasets (67% vs 39% and 76% vs 46%).
• The network achieved a sensitivity of 81.6% in histopathologically confirmed patients and 63.7% in MRI-negative FCD patients.
• Interpretable reports detailing lesion location, size, confidence, and salient features were generated.

Conclusions:

• MELD Graph is a state-of-the-art, interpretable tool for FCD detection on MRI, offering significant PPV improvements.
• Its clinical implementation could enhance early diagnosis and management of focal epilepsy, potentially improving patient outcomes.