deepFCD

This study developed a multicenter-validated deep convolutional neural network (CNN) to detect MRI-negative focal cortical dysplasia (FCD) using 3D T1 and FLAIR MRI. The CNN achieved 93% sensitivity overall, including 85% in MRI-negative cases, and incorporated Bayesian uncertainty to guide clinical confidence. With strong generalizability across sites, ages, and scanners, this method supports presurgical evaluation in challenging epilepsy cases.