Localization of Focal Cortical Dysplasias
Focal cortical dysplasias (FCDs) are congenital disruptions of neuronal migration and constitute a major cause of therapy-refractory focal epilepsy. If amenable to epilepsy surgery, more than 80% of patients reach seizure-freedom after a complete resection of the lesion. However, FCDs may be easily overlooked in conventional visual assessment of MRI. Automated approaches have been proven useful for the detection of FCDs and their use may enhance the postoperative outcome. Nevertheless, as even most elaborate approaches do not yield satisfactory sensitivity, the development of novel approaches is warranted on clinical grounds.
Our aim is to develop tools that may be used in routine clinical assessment based on state-of-the-art deep learning methodology, namely generative adversarial and convolutional neural networks, that robustly detect MR-negative FCD cases. In first pilot studies, our current approaches reach detection sensitivity-scores above 90% while maintaining a near perfect specificity of ~96%. Most importantly our results are invariant to different scanner manufacturers, MR-sequence parameters, and variable data quality.