Abstracts

Focal Cortical Dysplasia Type-II: MRI-based Profiling and Subtype Prediction

Abstract number : 2.202
Submission category : 5. Neuro Imaging
Year : 2015
Submission ID : 2326949
Source : www.aesnet.org
Presentation date : 12/6/2015 12:00:00 AM
Published date : Nov 13, 2015, 12:43 PM

Authors :
S. Hong, B. C. Bernhardt, D. Schrader, N. Bernasconi, A. Bernasconi

Rationale: Focal cortical dysplasia (FCD) Type-II, a highly epileptogenic malformation, is characterized by cortical dyslamination and dysmorphic neurons, either in isolation (IIA) or with balloon cells (IIB). Compared to Type-IIB, IIA has been suggested to present with subtle MRI signatures (Blumcke et al., 2011) and less discernible boundaries (Fauser et al., 2004, Lawson et al., 2005), which may explain less favorable surgical outcome (Tassi et al., 2012). Given the irreversible nature of surgery, it is important to predict FCD subtypes in vivo. Here, we propose a multiparametric MRI description of the FCD lesional spectrum and histological subtype prediction using machine learning.Methods: We studied 33 patients with histologically verified FCD (9 Type-IIA, 24 IIB) and 41 age and gender matched controls using 3.0 Tesla multimodal MRI (T1w, FLAIR, DWI, and rs-fMRI). Two experts manually segmented FCD lesions. Using MRI postprocessing, we generated multiple surfaces running through the cortical mantle and the underlying white matter. In addition to cortical morphometry (cortical thickness, sulcal depth), we sampled texture (intra- and sub-cortical intensity, and its gradient vertical and tangential to the cortical surface), diffusion (fractional anisotropy, mean diffusivity), and functional features (amplitude of low frequency fluctuation, regional homogeneity). We compared patient groups (i.e., IIA, IIB) to controls and to each other using two-sample t-tests, corrected at FDR<0.05. To address histological putative peri-lesional anomalies (Cohen-Gadol et al., 2004), we systematically examined features in the lesional perimeter using geodesic distance mapping. Support vector machines were employed to automatically discriminate Type-IIA, IIB, and control cortex in individual cases.Results: FCD Type-IIB lesions revealed marked alterations in morphology and texture relative to corresponding regions in controls; conversely, abnormalities in IIA were subtle and mainly restricted to FLAIR intensity (Fig. 1A). Subtypes also diverged in regional homogeneity in functional signals, showing a significant decrease in Type-IIB and marginal increases in IIA. While Type-IIB impacted features across virtually all intra- and sub-cortical levels, anomalies in Type-IIA were found mainly in the proximity of the cortical interface (Fig. 1B). Intensity and gradient features extended beyond the primary lesion in both subtypes (up to 6mm). A classifier combining multimodal, multi-surface and peri-lesional features predicted FCD subtypes with 91% accuracy (Fig. 2), outperforming those using unimodal features that are averaged across surfaces (McNemar's test: p<0.025).Conclusions: Our results emphasize the utility of integrating metrics of structure and function to dissociate neuroimaging signatures across histological spectrum of FCD. Peri-lesional abnormalities beyond the visible FCD support a role for quantitative modeling to improve surgical target definition.
Neuroimaging