Abstracts

Rationale: Epilepsy is associated with progressive brain atrophy over and above of normal ageing. We aimed to determine whether machine-learning algorithm can identify groups of patients with epilepsy showing different trajectories of progressive atrophy within cortical and subcortical regions.

Methods: We included T1-weighted MRI scans of 729 subjects: 418 with focal epilepsies, 193 with idiopathic generalised epilepsies and 118 healthy controls. We calculated cortical thickness and subcortical volumes related to epilepsy and expressed each regional measurement as a z-score relative to control population. Because regional brain volumes decrease over time the z-scores become negative with disease progression. For simplicity we took the negative value of the z-scores so that the z-scores would increase as the brain volumes became more abnormal. We used a machine-learning technique—Subtype and Stage Inference (SuStaIn)—able to uncover data-driven disease phenotypes with distinct temporal progression patterns from cross-sectional data. In each subject, SuStaIn provides a data-driven taxonomy (set of subtypes and stages), as well as detailed pictures of the progression of neurodegeneration within each of the data-driven subgroups.

Results: In focal epilepsy, SuStaIn identified three subtypes: (i) within the “cortical” subtype (49% of all patients), the first regions to become atrophic were superior- and transverse-temporal regions, followed by superior frontal and precentral regions, then precuneus and posterior cingulate cortex; (2) within the 2^nd subtype (“hippocampus,” 37% of all patients), the hippocampus was the first atrophic brain region; (3) within subtype “basal ganglia and thalamus” (14%), the first affected regions were basal ganglia, followed by thalamus and patients.

The most severely affected patients with the most seizures belonged to subtype 2 (hippocampus) with subtype 3 (basal ganglia and thalamus) showing a higher frequency of generalised tonic and clonic seizures (GTCS). SuStaIn stages of progression in focal epilepsies were related to illness duration (r=0.1, p=0.03).

In IGE, SuStaIn identified two subtypes: “cortical” and “basal ganglia and thalamus.” The sequence of atrophy is similar to findings focal epilepsy. 69% of IGE patients in the “cortical” subtype had juvenile absence epilepsy or juvenile myoclonic epilepsy patients, 68% of patients in “basal ganglia and thalamus” subtype had mainly GTCS. SuStaIn stages of progression in IGE were not related to illness duration.

Conclusions: Our findings suggest that data-driven MRI-based subtypes correlate with syndromic epilepsy diagnoses and predict cortical and subcortical atrophy progression. MRI-based classification may be used to define groups of patients in interventional trials to disrupt or slow progressive atrophy.

Funding: Please list any funding that was received in support of this abstract.: Newton International Fellowship.