A NOVEL artificial intelligence (AI) model, MELD Graph, successfully identified 64% of epilepsy-causing focal cortical dysplasia (FCD) lesions that had been previously missed by radiologists, demonstrating strong potential for clinical application.
Focal cortical dysplasia (FCD) is a leading cause of drug-resistant focal epilepsy and is often challenging to detect on MRI scans. Many cases are considered MRI-negative, making diagnosis and surgical planning difficult. This study, part of the Multicenter Epilepsy Lesion Detection (MELD) Project, evaluated the use of a graph neural network (MELD Graph) to improve the identification of these subtle brain lesions. The algorithm aims to support radiologists by providing interpretable reports detailing lesion location, size, and morphological characteristics, alongside confidence scores to enhance clinical decision-making.
Researchers analysed retrospective MRI data from 703 patients with FCD-related epilepsy and 482 controls, collected from 23 epilepsy centres worldwide. Data from 20 centres were used for model training and testing, while three centres provided independent validation data. The MELD Graph was trained to detect FCD lesions using 34 surface-based MRI features, with performance compared against an existing automated algorithm. In the main test dataset, the model demonstrated a sensitivity of 81.6% in histopathologically confirmed seizure-free patients and 63.7% in MRI-negative patients. The positive predictive value (PPV) for identified lesions was 67% (70% sensitivity; 60% specificity), significantly outperforming the baseline algorithm, which had a PPV of 39% (67% sensitivity; 54% specificity). In an independent test cohort, the MELD Graph achieved a PPV of 76% (72% sensitivity; 56% specificity), compared to 46% for the existing algorithm. Importantly, the AI-generated reports provided key lesion characteristics to facilitate integration into clinical workflows.
These findings suggest that MELD Graph represents a significant advancement in AI-assisted radiology for epilepsy diagnostics. By offering improved lesion detection with interpretable outputs, the algorithm has the potential to enhance early diagnosis and optimise neurosurgical planning for patients with FCD-related epilepsy. Future research should explore prospective validation in clinical settings and assess its impact on patient outcomes, further paving the way for AI-driven tools in epilepsy care.
Katrina Thornber, EMJ
Reference
Ripart M et al. Detection of epileptogenic focal cortical dysplasia using graph neural networks: a MELD study. JAMA Neurol. 2025;DOI10.1001/jamaneurol.2024.5406.
