Authors :
Mathilde Ripart, MSc – UCL Great Ormond Street Institute of Child Health, London, UK
Jordan DeKraker, PhD – McGill University
MELD Project, n.a. – UCL Great Ormond Street Institute of Child Health
Ali Khan, PhD – Western University
Torsten Baldeweg, MD – University College London
Sophie Adler, MBPhD – UCL Great Ormond Street Institute of Child Health
Presenting Author: Konrad Wagstyl, MBPhD – School of Biomedical Engineering & Imaging Sciences, King's College London, UK
Rationale: Hippocampal Sclerosis (HS) can elude visual detection on MRI scans of patients with temporal lobe epilepsy (TLE), causing delays in surgical treatment and reducing the likelihood of postsurgical seizure-freedom. We developed AID-HS, an open-source software to characterise and localise HS to aid the presurgical evaluation of children and adults with suspected TLE. AID-HS has been validated on a large heterogeneous multi-centre cohort.
Methods: AID-HS was developed on an international multicentre cohort of 154 patients with HS, 90 disease controls with focal cortical dysplasia (FCD), and 121 healthy controls, from 4 sites (main cohort, Table 1). The classifier was validated on an independant multicentre test cohort of 275 patients with HS and 161 disease and healthy controls from 14 epilepsy centres worldwide (independant test cohort, Table 1). We used the open-source software HippUnfold
(DeKraker et al. 2023) to extract morphological surface-based features and volumes of the hippocampus from T1w MRI scans. These features were normalised by controls and asymmetries were computed to enhance differences between and within subjects. We characterised pathological hippocampi in patients, by comparing them to normative growth charts and analysing within-subject feature asymmetries. Feature asymmetry scores were used to train a logistic regression classifier to detect and lateralise HS and output predicted scores. The classifier was evaluated for its sensitivity in detecting HS (i.e. differentiating HS patients from controls) and lateralizing HS ( i.e. detecting the side of the abnormality) in patients as well as for its specificity in controls (i.e. differentiating controls from patients). AID-HS outputs individualised patient reports that present detailed HS detection and lateralisation predicted scores as well as hippocampal feature asymmetries and characterisations of hippocampal abnormalities against normative trajectories (Figure 1).
Results: HS was characterised 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 the main cohort (Table 1). 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 generalise to new data. An example of an individual report is presented for a patient with a right HS, previously missed on inspection of MRI scan but lateralized with intracranial EEG (Figure 1).
Conclusions: AID-HS is an open-source pipeline, for accurately detecting and lateralizing HS and outputting interpretable reports. It has been validated in a large, independent, heterogeneous, multi-centre, cohort of paediatric and adult patients with HS.
Funding: The MELD Project, M.R. and S.A. are funded by the Rosetrees Trust (A2665) and Epilepsy Research Institute (P2208). K.W. is funded by the Wellcome Trust.