Automated Segmentation of Focal Cortical Dysplasias: A Meld Study.

Introduction Focal cortical dysplasia (FCD) is a common cause of drug-resistant epilepsy, and accurate detection on MRI is critical for presurgical planning. However, MRI identification of these subtle lesions remains a challenge. This Multicentre Epilepsy Lesion Detection (MELD) project study aimed to develop a whole-brain graph neural network (GNN) for segmenting FCDs.

Methods The MELD cohort includes 618 patients with FCD and 397 controls, used to train and test a novel GNN model. The cortical mesh was represented as a graph, treating vertices as nodes connected to neighbouring vertices by edges, enabling the network to learn spatial relationships between brain regions. The model was trained to segment vertices into lesional and non-lesional. We combined four tasks: a lesion segmentation task, a task to predict distance to the lesion, and a task to classify whether a lesion is present. These last two tasks were designed to mitigate uncertainty in manually delineated lesion masks.

Results On a withheld test cohort, the GNN model achieved a sensitivity of 67% in patients, with a specificity of 76% in controls, a significant gain in specificity in controls against patch-based approaches on the same dataset (sensitivity 67%, specificity 49%). The GNN model decreased the number of false positive clusters from a median of 1 per patient [IQR:0-3] to 0 [IQR:0-1].

Conclusions Our study demonstrates the GNNs for FCD segmentation, substantially reducing false positives. This improvement in specificity is vital for clinical integration of lesion-detection tools into radiological workflows, increasing clinical confidence in AI radiological adjuncts.