Authors :
Presenting Author: Jessica Royer, B.Sc., Psy.D. – Montreal Neurological Institute and Hospital
Sara Lariviere, Ph.D. – Harvard Medical School; Judy Chen, B.Sc. – McGill Univeristy; Raul Rodriguez-Cruces, MD/PhD – McGill University; Jordan DeKraker, Ph.D. – McGill University; alexander Ngo, B.Sc. – McGill University; Hans Auer, B.Sc. – McGill University; Ella Sahlas, B.Sc. – McGill University; Dewi Schrader, MHSc, MBBS, FRCPC – University of British Columbia; Andrea Bernasconi, MD – McGill University; Neda Bernasconi, MD/PhD – McGill University; Birgit Frauscher, MD/PD – McGill University; Boris Bernhardt, PhD – McGill University
Rationale:
The identification of a lesion in magnetic resonance imaging (MRI) significantly increases the odds of seizure freedom following resective surgery in patients with drug-resistant focal epilepsy. We describe a new open access image processing and analysis environment to identify patient-specific alterations in brain morphology and microstructure and showcase its utility in a cohort of patients diagnosed with drug-resistant mesio-temporal lobe epilepsy (mTLE). This tool emphasizes clinical translation of MRI post-processing findings via detailed reports of structural feature asymmetry and regional changes.
Methods:
A cohort of 31 mTLE patients (15 women; age 37.81±11.03 years) and 89 healthy controls (38 women; age 31.06±8.47 years) underwent high-resolution T1-weighted, quantitative T1 (qT1), diffusion-weighted, and fluid-attenuated inversion recovery (FLAIR) imaging at 3T. Multimodal MRI data were processed using micapipe (Rodriguez-Cruces, Royer, et al., 2022; Figure 1A) and hippunfold (DeKraker et al., 2022), and analyzed using an automated clinical case report pipeline (
https://github.com/MICA-MNI/z-brains). Z-brains generates patient-specific z-score measures of cortical thickness/regional volume (CT/RV), qT1 intensity, apparent diffusion coefficient (ADC), fractional anisotropy (FA), and FLAIR intensity within cortical and subcortical grey matter, and standardizes individual patient data against a user-defined normative cohort. In addition to individual features, the pipeline also allows for the generation of multivariate deviation maps combining available structural modalities.
Results:
Feature asymmetry could correctly identify disease laterality in most patients from (i) neocortical data (percentage of patients with consistent lateralization from telemetry: CT: 64.00%; qT1: 66.67%; ADC: 67.86%; FLAIR: 45.45%; multivariate: 64.00%), (ii) regionally-sampled subcortical and hippocampal metrics (RV: 80.00%; qT1: 85.71%; ADC: 93.75%; FLAIR: 100.00%; multivariate: 92.31%), and (iii) an unfolded model of the hippocampus (CT: 67.86%; qT1: 60.00%; ADC: 81.25%; FLAIR: 75.00%; multivariate: 61.11%). Moreover, the 10% most atypical vertices within multivariate maps in nearly every patient (96.78%) overlapped, to some extent, with temporal neocortices or hippocampi (mean overlap: 42.26%±30.56%). Individual-specific findings are provided in reports illustrating regions of significant intensity deviation from a normative group composed of healthy controls (z >1.96 or z< -1.96; threshold can be modified by user). Results are then projected to each patient’s native space for visualization of volumetric results (Figure 1B).