MELD as an Adjunct for SEEG Trajectories

Epilepsy is a disorder of the brain that is associated with disabling seizures. It affects 100,000 children in the UK, 25-30% of whom will be classed as drug resistant.3 In these children, there is increasing evidence that resective epilepsy surgery in appropriate candidates can lead to seizure freedom and improve quality of life and cognitive outcomes.4-6 However, about 30% of children do not achieve seizure freedom following epilepsy surgery and data suggests that these figures are not improving over time despite increasing use of intracranial evaluation via stereoelectroencephalography (SEEG). 7 The planning of where to place SEEG electrodes currently relies on an expert multidisciplinary team consisting of neurologists, neurophysiologists and neurosurgeons. Information from multiple sources, mainly the patient's magnetic resonance imaging (MRI) scan and video-electroencephalography (EEG) recording, are used to generate hypotheses about the location of the clinical seizure onset zone (SOZ). Using this information, between 5-18 electrodes are implanted and the recordings continue for 5-15 days in hospital. In a retrospective review of 75 SEEG cases, a focus was identified in about 77% of cases which means that the focus is sometimes missed. This prospective single arm pilot study to aims assess a new automated lesion detection algorithm, MELD, designed to identify focal cortical dysplasias (the most common pathology associated with focal epilepsy in children) on otherwise 'normal' MRI scans.1 This algorithm was developed in-house by collaborators in this grant application. In our subsequent retrospective study of 34 SEEG patients, the algorithm colocalised with the SEEG-defined SOZ in 62% of all patients with a cortical SOZ and 86% of all patients with a histologically confirmed focal cortical dysplasia.2 Importantly, there were 3 patients whose SOZ was thought to be missed on SEEG who had MELD-identified lesions that were not implanted. In order to improve the algorithm, investigators have subsequently launched an international multicentre collaboration (https://meldproject.github.io//) to increase the number of lesion positive and control scans available to train the algorithm, improving its sensitivity, specificity and accuracy. This project has gathered over 550 lesional and 350 control scans, which will be used to train the algorithm. The prospective MAST Trial is therefore the ideal next step in the evaluating the utility of the MELD algorithm in identifying abnormal areas of the brain that could be responsible for seizures.