Abstracts

Rationale: Focal cortical dysplasia (FCD) type II is a highly epileptogenic developmental malformation characterized by intracortical dyslamination and dysmorphic neurons, either in isolation (FCD-IIA) or together with balloon cells (FCD-IIB). Recent functional MRI studies suggest abnormal local functional profiles and peri-lesional connectivity [1, 2]. Beyond these properties, however, FCD connectivity to the rest of the brain and its interplay with large-scale network organization remains unknown. Variability in lesion location, size and histological subtypes challenge the classical group-based approach to study brain connectivity. Thus, we propose a novel MRI framework that systematically probes the connectivity of FCD lesions using hierarchical clustering of resting-state functional connectivity. Methods: Fig 1 summarizes our method. We studied 27 patients with histologically-verified FCD and 34 healthy controls examined with a 3T MRI. After subdividing every lesion into a set of similarly-sized cortical patches, we computed seed-based resting-state functional connectivity. Based on the canonical community structure of the human functional connectome [3], we dichotomized connectivity profiles of lesional patches into those belonging to the same functional community as the lesion (intra-community) and other communities (inter-community). We then applied hierarchical clustering to group lesions based on the similarity of their patch-based connectivity profiles. To evaluate the clinical yield of our approach, we employed an ensemble classifier [4] in that each base learner was trained with connectome profiles from one of the identified clusters, to predict histological grade (i.e., IIA vs. IIB) and response to surgery (i.e., seizure-free vs. non-free). A 100´10-fold cross-validations tested the accuracy and generalizability.  Results: 96 Normal 0 false false false EN-US KO X-NONE /* Style Definitions */table.MsoNormalTable {mso-style-name:"Table Normal"; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-style-noshow:yes; mso-style-priority:99; mso-style-parent:""; mso-padding-alt:0cm 5.4pt 0cm 5.4pt; mso-para-margin:0cm; mso-para-margin-bottom:.0001pt; mso-pagination:widow-orphan; font-size:10.0pt; font-family:"Cambria",serif;} Our connectome-based biotyping identified three distinct lesion classes (Fig 2A-B) with decreased intra- and inter-community connectivity (class 1), decreased intra-community, but normal inter-community connectivity (class 2), and increased intra- and inter-community connectivity (class 3). Machine learning informed by these classes predicted histological grading and post-surgical outcome with excellent accuracy (≥84%; permutation test: p < 0.01), outperforming learners operating without class information (Fig 2C).  Conclusions: Our data provides evidence for substantial heterogeneity in the relation of FCD lesional connectivity to whole-brain functional networks. Given the irreversible nature of surgery, identifying pathology- and outcome-specific imaging signatures on pre-operative MRI may optimize surgical planning and patient counseling.[1] Besseling RM, et al. Abnormal Profiles of Local Functional Connectivity Proximal to FCD. PLoS One 2016;11:e0166022.[2] Hong SJ,  et al. Multimodal MRI profiling of FCD type II. Neurology 2017;88:734-742.[3] Yeo BT, et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol 2011;106:1125-1165.[4] Dietterich TG. Ensemble methods in machine learning. Multiple Classifier Systems 2000;1857:1-15. Funding: 2017 Canadian League Against Epilepsy Postgradudate Training Fellowship