Planning sEEG implantation using automated lesion detection: retrospective feasibility study

Abstract

Objective: A retrospective, cross-sectional study to evaluate the feasibility and potential benefits of incorporating deep-learning on structural MRI into planning stereoelectroencephalography (sEEG) implantation in paediatric patients with diagnostically complex drug-resistant epilepsy. This study aims to assess the degree of co-localisation between automated lesion detection and the seizure onset zone (SOZ) as assessed by sEEG. Methods: A neural network classifier was applied to cortical features from MRI data from three cohorts. 1) The network was trained and cross-validated using 34 patients with visible focal cortical dysplasias (FCDs). 2) Specificity was assessed in 20 paediatric healthy controls. 3) Feasibility for incorporation into sEEG implantation plans was evaluated in 38 sEEG patients. Coordinates of sEEG contacts were coregistered with classifier-predicted lesions. sEEG contacts in seizure onset and irritative tissue were identified by clinical neurophysiologists. A distance of <10mm between SOZ contacts and classifier-predicted lesions was considered co-localisation. Results: In patients with radiologically-defined lesions, classifier sensitivity was 74% (25/34 lesions detected). No clusters were detected in the controls (specificity 100%). Of 34 sEEG patients, 21 patients had a focal cortical SOZ. Of these there was co-localisation between classifier output and SOZ contacts in 62%. The algorithm detected 7/8 histopathologically-confirmed FCDs (86%). Conclusions: There was a high degree of co-localisation between automated lesion detection and sEEG. We have created a framework for incorporation of deep-learning based MRI lesion detection into sEEG implantation planning. Our findings demonstrate that automated MRI analysis could be used to plan optimal electrode trajectories.

Machine learning applications on neuroimaging for diagnosis and prognosis of epilepsy: A review

“…T1w MRI gray matter volume [30] cortical thickness, intensity at the grey-white matter contrast, curvature, sulcal depth, intrinsic curvature [37] T2w-MRI volume and intensity sampled on the medial sheet of hippocampus [38] FLAIR-MRI intensity sampled at 25%, 50% and 75% of the cortical thickness and at the grey-white matter boundary [37] DTI MD and FA [39] DKI MK [39] MD, FA, MK and the fusion of FA and MK [40] fMRI fALFF [30] BOLD signal values in different regions, hemispheres and tasks [41] PET hemisphere symmetry tensor [42] Abbrev : MD, Mean diffusion; FA, Fractional Anisotropy; MK, Mean Kurtosis; fALFF, fractional amplitude of low-frequency fluctuation.…”

“…Three studies chose the shallow neural network as the classifier in compositional method [20,89,37].…”

“…and FLAIR MRI, effectively localizing the FCD with 73.7% sensitivity and 90% specificity [89]. And the morphological and intensity features could be extracted to localize the FCD regions with up to 73.5% sensitivity and 100% specificity by wagsty et al [37].…”