Hyo Min Lee | Montreal Neurological Institute and Hospital (The Neuro)
Hyo Min Lee, Fatemeh Fadaie, Ravnoor Gill, Benoit Caldairou, Viviane Sziklas, Joelle Crane, Seok-Jun Hong, Boris C Bernhardt, Andrea Bernasconi, Neda Bernasconi, Decomposing MRI phenotypic heterogeneity in epilepsy: a step towards personalized classification, Brain, 2021; awab425, https://doi.org/10.1093/brain/awab425
Advanced MRI analysis reveals markers for person-centered care of epilepsy patients
In temporal lobe epilepsy (TLE), precise predictions of clinical outcomes remain challenging. Analyzing differences between patients is increasingly recognized as a step towards person-centered care. Here, PhD. Student Hyo Min Lee and his colleagues at the Montreal Neurological Institute and Hospital analysed Magnetic Resonance Imaging (MRI) scans of brains of patients and found unique structural markers of disease that can be used to provide personalized prognostics.
The researchers utilized unsupervised machine learning to estimate disease factors from MRI features of whole-brain MRI scans in 82 TLE patients. They then assessed the specificity of these factors against healthy individuals and frontal lobe epilepsy patients. Moreover, they evaluated the data-driven disease factor model for individualized predictions of clinical outcomes.
They identified four distinct disease factors characterized by changes in brain structure in specific brain regions. Statistical analysis showed these factors to be stable and robust, meaning they were co-expressed in each TLE patient, but they were not expressed in healthy controls and only negligibly in disease controls, supporting specificity. When patients were classified based on these, the factors accurately predicted drug-response and postsurgical seizure in more than 80% of patients, in addition to being excellent predictors of specific verbal, memory and motor skills.
In summary, disease factors allow for a robust, fine-grained imaging-based characterization of interindividual variability in patterns of whole-brain structural makeup in patients with TLE. Innovating along conceptual and clinical domains, this work illustrates how technical literacy and computational proficiency are increasingly necessary for advancing knowledge in the era of artificial intelligence applied to clinical neurology.
Hyo Min Lee
This work was conducted as part of Hyo Min Lee’s PhD thesis in the Neuroimaging of Epilepsy Laboratory (NOEL), McConnell Brain Imaging Centre, Montreal Neurological Institute and Hospital, McGill University, under the supervision of Dr. Neda Bernasconi.