Abstracts

Rationale: Visual identification of hippocampal sclerosis (HS) on MRI, characterized by atrophy and increased T2 signal intensity, allows a correct seizure focus lateralization in patients with drug-resistant temporal lobe epilepsy (TLE). However, HS visibility is variable depending on the degree of neuronal loss and gliosis(1), posing a significant presurgical challenge in 40%-60% of patients with so-called MRI-negative TLE. Here we developed a fully automated MRI-based clinical decision support system to lateralize seizure focus in TLE. Methods: We studied 69 TLE patients (34 male, age=34.10) and 40 matched controls (21 male, age=31.7). Based on clinical MRI evaluation, 37 patients had visible HS (MRI+), while 32 were deemed as MRI-negative (MRI-). Histopathology, available in 48 patients, showed combined neuronal loss and gliosis (n=34) and isolated gliosis (n=14). All subjects underwent 3T MRI using the HARNESS-MRI protocol (2) (3D T1-weighted, 1x1x1 mm3; 3D FLAIR, 0.9x0.9x0.9 mm3; 2D coronal T2, 0.4x0.4x0.2 mm3). Hippocampal subfields (CA1-3, CA4-DG) and subiculum were automatically segmented using SurfPatch (3). A medial surface sheet running along the central path of each subfield allowed for vertex-wise sampling of columnar volume and normalized T2 signal intensity (4). Moreover, to maximize the detection of HS features (increased FLAIR, decreased T1 signals), we created a synthetic contrast based on FLAIR/T1 ratio. Linear models assessed group differences. In regions of significant findings, we calculated an asymmetry index [2x(L-R)/(L+R)], which was fed to a supervised linear discriminant classifier for individual lateralization of the seizure focus; validation was done following a repeated 5-fold scheme. We assessed generalizability by testing our classifier on an independent dataset of 10 TLE patients who underwent 3T MRI on a different scanner (5 MRI+ and 5 MRI-). Results: Group analysis mapped the spatial distribution of anomalies across subfields (Fig. 1) in TLE patients relative to controls. Individual classifications are shown in Table 1. Considering all patients, columnar volume correctly lateralized 82%, T2 intensity 87%, and FLAIR/T1 ratio 88%. In MRI+ cases, FLAIR/T1 ratio provided 96% accuracy, columnar volume 95%, and T2 intensity 90%. In MRI- cases, T2 signal provided 81% accuracy, FLAIR/T1 ratio 78%, and columnar volume 67%. When tested on the independent dataset, the classifier correctly lateralized 90% using either T2 intensity or FLAIR/T1 and 60% using columnar volume. Conclusions: Automated analysis combining the recently-recommended high-resolution multicontrast HARNESS-MRI protocol 2 allows a reliable seizure focus lateralization irrespective of the degree of hippocampal damage, thus offering substantial gain in performance relative to visual rating. T1/FLAIR ratio is particularly sensitive as it captures co-occurring T1 and T2 signal anomalies, thus maximizing HS detection in MRI negative TLE.Reference1. Bluemcke, at al. Epilepsia (2011)2. Bernasconi A, et al. Epilepsia (2019)3. Caldairou B, et al. MICCAI (2016)4. Kim H, et al. MICCAI (2014) Funding: CIHR (MOP-57840, MOP-123520)