Introduction

Focal cortical dysplasia (FCD) is a relatively common cause of drug-resistant focal epilepsy, and particularly of MRI negative (MRI-) epilepsy.  Approximately 15-25% of patients being evaluated for epilepsy surgery have apparently normal MRIs; FCD is diagnosed pathologically in 25-50%  of these patients \cite{Blumcke2017,Lerner2009,Bien2009}.  Because post-operative seizure outcomes are significantly worse in MRI- compared to MRI positive (MRI+) patients \cite{Téllez-Zenteno2010}, improved identification of subtle FCDs is of great clinical importance as it may lead to better surgical outcomes.
FCDs are variable in their histopathological and radiographic appearances.  Pathological abnormalities range from cortical dyslamination in type I, to the presence of large dysmorphic neurons in type IIa (FCDIIa), and balloon cells and more prominent dysmyelination in type IIb (FCDIIb) \cite{Blümcke2011}.  FCD type I lesions are typically difficult to identify radiologically, most often consisting of cortical thinning and lobar hypoplasia, at times with blurring of the gray-white junction and changes in T1- or T2-weighted image intensity.  FCD type II lesions are more easily identified on MRI, with typical findings consisting of increased cortical thickness, blurring of the gray-white junction, FLAIR/T2 hyperintensity (including the transmantle sign in FCD type IIb), and alterations in gyrification patterns \cite{Adler2017a,Kini2016}.  These findings, however, appear inconsistently within and across lesions \cite{Hong2017} and are often subtle, with up to 80% of small bottom-of-sulcus dysplasias being missed on routine visual inspection \cite{Besson2008}.
A number of post-processing methods have therefore been developed to aid in FCD detection.  These range from creation of synthetic contrasts to highlight areas of interest (such as the Morphometric Analysis Program (MAP) \cite{Huppertz2005}) to machine-learning based fully automated detection methods  \cite{Adler2017,Ahmed2015,Hong2014}.  Across these methods, two key challenges have been 1) selection of optimal features to describe FCD lesions, and 2) accounting for variability within and across FCD lesions, as well as in normal cortex.  While a wide variety of features have been investigated, most methods have used features derived from either voxel-based morphometry \cite{Huppertz2005,Martin2017} or surface-based morphometry (SBM) (see review in \cite{Kini2016}).  It has been challenging to compare the efficacy of these methods,  as most algorithms are not easily implemented across centers, lesions studied vary in their underlying histology as well as radiological appearance (MR negative vs positive), and particularly for supervised machine-learning methods, performance also appears to improve significantly with increasing size of the healthy volunteer and patient training sets, likely allowing for improved modeling of both pathological and normal cortical variability \cite{Jin2018}.  
In this work, we describe a novel approach for describing the normal variability observed along the cortical sheet in healthy volunteers.  We use this representation to then identify outliers, first in healthy volunteers to identify atypical normal cortical regions, and then in patients to identify focal cortical dysplasias.  Our model is based on an implementation of 3D multiscale rotationally-invariant local image features across multiple MR contrasts, similar to those that have been previously shown to efficiently represent the local statistics of natural images \cite{Simoncelli2001}.  We then created a latent representation of this normative data that allows for straightforward outlier detection in our multivariate feature space.  We demonstrate that this model identifies several atypical cortical regions in healthy volunteers as outliers, corresponding to regions known to have atypical underlying cytoarchitecture and myelination patterns.  In patients, most FCD lesions also appear as outliers, but are similar in their underlying features to some normally atypical regions.  Local normalization corrects for the expected appearance at any given cortical location, and allows for automated detection of FCDs.  

Materials and Methods

Study Participants

From our surgical epilepsy imaging database, we retrospectively identified 15 consecutive patients undergoing presurgical evaluation for drug-resistant focal epilepsy from 2014-2019 with: 1) radiologically apparent (MRI+) or histologically proven (MRI+ or MRI-) FCDs; and 2) our standard 3T MRI epilepsy structural imaging protocol.  Patients were excluded if they underwent other MR imaging protocols or had low image quality on visual inspection. The control group consisted of 30 healthy volunteers (HVs) scanned using the same imaging protocol with no previous history of neurologic, psychiatric, or other significant medical illnesses that may affect the central nervous system. Data were collected at the National Institutes of Health (NIH) Clinical Center (Bethesda, MD). All participants were enrolled in an Institutional Review Board-approved research protocol; informed consent was obtained from all participants.

Lesion Labels

For MRI+ patients, lesions were traced in the volume using the Analysis of Functional NeuroImages (AFNI) software package \cite{Cox1996a} by an experienced neurologist using the Tweighted image, informed by the T2 and FLAIR images when necessary. For MRI- patients, the postoperative T1 was registered to the preoperative T1 in the same manner as described below for T2 and FLAIR images; the resected region was manually traced using AFNI. Lesion masks were mapped onto the smoothed white matter surface using AFNI's 3dVol2Surf function.

MRI Acquisition Protocol

All participants were scanned on a Philips Achieva 3T MRI scanner in the NIH Clinical Center Radiology Department as follows: 1) 3D T1 weighted MPRAGE (T1): TR = 6.8–7.2, TE = 3.2 ms, TI = 900ms, flip angle = 90, voxel size = 0.75 x 0.75 x 0.8, acceleration factor 2 in slice direction, acquisition time = 7:02 min; 2) 3D T2 weighted FSE (T2): TR = 2500, TE = 225–245, voxel size = 1 x 1 x 2 or 1 x 1 x 1, acceleration factor 2 in slice and phase directions, acquisition time = 5:03 min; and 3) 3D FLAIR: TR = 4800, TE = 271–415, TI = 1600, voxel size = 0.9 x 0.9 x 1, acceleration factor 2 in slice direction, 2.6 in phase direction, acquisition time = 6:10 min. All scans were acquired sagittally.

Image Preprocessing

For each individual subject, T2 and FLAIR images were co-registered to the T1 with an affine transformation using a normalized mutual information cost function and resampled to the T1 grid using AFNI.  Registered images were visually assessed for alignment.  Cortical reconstruction was performed using T1 and T2 images as input to FreeSurfer's standard processing pipeline with  FreeSurfer v6.0.0  \cite{Dale1999,Fischl1999}.  Results were visually inspected and manually corrected as needed. Cortical surfaces were resampled to a standard mesh using the AFNI SUMA package \cite{Cox1996a} to allow comparison of corresponding vertices across individuals.