Automated and Interpretable Detection of Hippocampal Sclerosis in Temporal Lobe Epilepsy: AID-HS

Abstract

Objective: Hippocampal sclerosis (HS), the most common pathology associated with temporal lobe epilepsy (TLE), is not always visible on magnetic resonance imaging (MRI), causing surgical delays and reduced postsurgical seizure-freedom. We developed an open-source software to characterize and localize HS to aid the presurgical evaluation of children and adults with suspected TLE.

Methods: We included a multicenter cohort of 365 participants (154 HS; 90 disease controls; 121 healthy controls). HippUnfold was used to extract morphological surface-based features and volumes of the hippocampus from T1-weighted MRI scans. We characterized pathological hippocampi in patients by comparing them to normative growth charts and analyzing within-subject feature asymmetries. Feature asymmetry scores were used to train a logistic regression classifier to detect and lateralize HS. The classifier was validated on an independent multicenter cohort of 275 patients with HS and 161 healthy and disease controls.

Results: HS was characterized by decreased volume, thickness, and gyrification alongside increased mean and intrinsic curvature. The classifier detected 90.1% of unilateral HS patients and lateralized lesions in 97.4%. In patients with MRI-negative histopathologically-confirmed HS, the classifier detected 79.2% (19/24) and lateralized 91.7% (22/24). The model achieved similar performances on the independent cohort, demonstrating its ability to generalize to new data. Individual patient reports contextualize a patient's hippocampal features in relation to normative growth trajectories, visualise feature asymmetries, and report classifier predictions.

Interpretation: Automated and Interpretable Detection of Hippocampal Sclerosis (AID-HS) is an open-source pipeline for detecting and lateralizing HS and outputting clinically-relevant reports. ANN NEUROL 2024.

FIGURE 1

AID‐HS overview. The T1w scan (A) is used as input in HippUnfold, which generates hippocampal segmentations and mesh surfaces that can be visualized flat or folded (B). Surface‐based features undergo preprocessing to remove outliers, adjust for site‐based batch effects, and account for inter and intra‐individual differences (C). Affected hippocampal features are compared to (i) normative developmental trajectories of hippocampal features generated from the healthy controls, and (ii) contralateral hippocampi to characterize the asymmetries (D). Asymmetries are used to train a logistic regression model to predict the likelihood of an individual having a left HS, right HS or no HS. These scores are used to detect and lateralize HS (E). AID‐HS outputs individualized patient reports that detail HS detection and lateralization predictive scores as well as hippocampal feature asymmetries and characterizations of hippocampal abnormalities against normative trajectories (F). AID‐HS, Automated and Interpretable Detection of Hippocampal Sclerosis; CA, cornu ammonis; DG, dentate gyrus; HS, hippocampal sclerosis; SRLM, stratum radiatum, lacunosum, and moleculare. [Color figure can be viewed at www.annalsofneurology.org]

FIGURE 2

Healthy hippocampal anatomy. (A) Folded and flat maps of HippUnfold‐derived surface‐based thickness, curvature and gyrification in our study's healthy controls. Correlation of surface‐based features extracted from our study's healthy controls compared to the HCP cohort. (B) Normative growth charts of harmonized features in healthy male and female hippocampi for the 5^th, 50^th and 95^th percentiles of the population. (C) Coefficients from the linear regression model testing the effect of hemisphere, sex and age on the harmonized features and colored by their significance (p‐values). HCP, Human Connectome Project; GAM, generalized additive model; OLS, ordinary least squares. [Color figure can be viewed at www.annalsofneurology.org]

FIGURE 3

Characterization of morphological abnormalities in HS. (A) Distribution of harmonized features in the ipsilateral hippocampi of HS patients (purple), contralateral hippocampi of HS patients (orange), and hippocampi of disease controls (green) plotted against normative trajectories derived from the 5^th and 95^th percentiles of the healthy male (blue dashed line) and female (red dashed line) controls' hippocampi. Histograms of the percentage of each group falling within each centile of normative curves are reported on the right axis. Features from ipsilateral HS consistently fell outside of the 5^th/95^th centiles for all features. (B) Boxplots of z‐score asymmetries in ipsilateral hippocampi of HS patients compared to healthy controls and disease controls. Statistically significant differences in distributions between each group were assessed using the Welch T‐test (p^W) for normal distributions and the Mann–Whitney test (p^M) for non‐normal ones. HS, hippocampal sclerosis; GAM, generalized additive model. [Color figure can be viewed at www.annalsofneurology.org]

FIGURE 4

Examples of AID‐HS reports for 2 patients with MRI‐negative right HS (example 1) and left HS (example 2). (A) Automated hippocampal segmentation and reconstructed hippocampal surfaces using HippUnfold, alongside automated quality control of the segmentation. (B) Individual hippocampal features compared to normative trajectories (with 25th – 75th percentiles in dark green, 5th – 25th and 75th – 95th percentiles in light green, patient's left hippocampus in blue and patient's right hippocampus in pink). (C) Asymmetry scores against left and right abnormality thresholds and automated lateralization scores from the AID‐HS classifier, indicating the probability that hippocampal feature asymmetries are consistent with left or right HS or that there is no asymmetry. AID‐HS, Automated and Interpretable Detection of Hippocampal Sclerosis; CA, cornu Ammonis; HS, hippocampal sclerosis; MRI, magnetic resonance imaging; SRLM, stratum radiatum, lacunosum, and moleculare. [Color figure can be viewed at www.annalsofneurology.org]