AUTOMATIC DETECTION OF SUBTLE FOCAL CORTICAL DYSPLASIA USING SURFACE-BASED FEATURES ON MRI Pierre Besson a,c , Olivier Colliot b , Alan Evans c and Andrea Bernasconi a,c a Neuroimaging of Epilepsy Laboratory, Montreal Neurological Institute, Montreal, Canada b Cognitive Neuroscience and Brain Imaging Laboratory, CNRS UPR 640-LENA, Université Pierre et Marie Curie - Paris 6, Hôpital de la Pitié-Salpêtrière, Paris, France c McConnell Brain Imaging Centre, Montreal Neurological Institute, Montreal, Canada ABSTRACT Focal cortical dysplasia (FCD) is an important cause of pharmacoresistant epilepsy. Small FCD lesions are difficult to distinguish from non-lesional cortex and remain often overlooked on radiological MRI inspection. This paper presents a method to detect small FCD lesions on T1-MRI relying on surface-based features: cortical thickness, gradient magnitude at the white-matter / grey-matter interface, cortical signal intensity, curvature and depth of inner-cortical surface. These features best describe the visual and morphometric characteristics of small FCD, and allow differentiating it from healthy tissues. The automatic detection was performed by a neural-network bagging trained on manual labels. The method was tested on 19 patients with small FCD and identified the lesion in 89% (17/19) of cases. Cluster analysis demonstrated that the lesional cluster was the largest in 76% (13/17) of identified cases. This new approach may assist the presurgical evaluation of patients with intractable epilepsy, especially those with “MRI-negative” epilepsy. Index Terms— Magnetic resonance imaging, Nervous system, Biomedical signal detection, Biomedical image processing, Neural network applications. 1. INTRODUCTION Malformations of cortical development (MCD) have been increasingly recognized as an important cause of pharmacoresistant epilepsy. Focal cortical dysplasia (FCD) [1], a malformation due to abnormal neuroglial proliferation, is the most frequent MCD in patients with intractable extra-temporal epilepsy [2]. Epilepsy surgery, consisting in the removal of the FCD lesion, is an effective treatment for these patients and magnetic resonance images (MRI) plays a pivotal role in presurgical evaluation [3]. Image analysis techniques were previously developed to detect FCD automatically on MRI, relying on different types of voxel-wise analysis [4-6]. In particular, we proposed computational models of FCD characteristics [7] and a Bayesian classifier for lesion detection [4]. While these approaches successfully identified FCD in a majority of patients, most of the lesions included in these studies were detected on routine radiological evaluation. On the other hand, the detection of small FCD lesions, which are overlooked in more than 80% of cases [8], is a much more difficult task and has never been addressed. This paper presents a new method for detecting small FCD lesions on T1-weighted MRI, relying on surface-based MR features of FCD. To increase the sensitivity of the automated method, we developed vertex-based analysis by projecting voxel-wise features onto the cortical surface. 2. METHODS 2.1. Image acquisition and preprocessing 3D MR images were acquired on a 1.5T scanner using a T1- fast field echo sequence (TR=18, TE=10, 1 acquisition average pulse sequence, flip angle=30°, matrix size=256×256, FOV=256, thickness=1mm) with an isotropic voxel size of 1mm 3 . All images underwent automated correction for intensity non-uniformity and intensity standardization [9], automatic registration into stereotaxic space [10], automatic tissue classification [11] and brain extraction [12]. 2.2. FCD features extraction To detect the lesion, five features were extracted from the MR images. These features correspond to visual characteristics – cortical thickening, a blurred transition between gray matter (GM) and white matter (WM), and hyperintensity signal within the displastic lesion [7] – or to morphological characteristics specific to small FCD – depth from the outer cortical surface and local curvature of the cortical surface [8]. 2.2.1. Extraction of cortical surfaces In each hemisphere, the inner and outer-cortical surfaces were computed using the CLASP (Constrained Laplacian Anatomical Segmentation using Proximities) algorithm [13]. The inner-cortical surface was extracted by inflating a 1633 978-1-4244-2003-2/08/$25.00 ©2008 IEEE ISBI 2008