Abstracts

Rationale: Hippocampal sclerosis (HS) is the most common cause of refractory temporal lobe epilepsy (TLE). Radiologically, it is recognized by hippocampal atrophy and increased T2-weighted signal, with quantitative imaging such as hippocampal volumetry and T2 relaxometry improving sensitivity. Following our automation of hippocampal volumetry (HippoSeg, https://hipposeg.cs.ucl.ac.uk) we now present an automated method to evaluate volumetric profiles along the anterior-posterior (AP) axis, to improve our sensitivity to detect “local atrophy”. Methods: 50 TLE-patients with radiologically-confirmed HS (mean age 42.9 years, 23 male) and 50 age-matched healthy controls underwent imaging on a 3T GE MR750 scanner with a 1mm isotropic 3D T1-weighted MPRAGE. Hippocampi were automatically segmented using HippoSeg. Subject scans were reoriented to have their hippocampi along the AP axis (found using principal component analysis) and cross-sectional area (CSA) was obtained at each slice and corrected for intracranial volume. The MNI-152 template and hippocampal delineations from the Harvard-Oxford atlas were similarly reoriented (template referred to as MNI-HC). To compare across subjects, all subjects’ scans were registered to this MNI-HC atlas with an initial affine alignment followed by a non-linear (b-spline) registration, with the matching between template and subject scan excluding the hippocampi. This modified matching, by only evaluating the cost-function in the registration in the brain excluding the hippocampal segmentations from the atlas, was done to avoid influence of atrophy or other pathology on the registration. CSA at each slice were transformed according to this registration and sampled along the centerline of the hippocampal segmentations from the Harvard-Oxford atlas. Statistical comparisons were done at each point using the 50 control subjects to generate a normative range (median±2*IQR). For any point along the AP axis, the CSA was said to be significantly smaller if that slice and the 2 slices either side of it (a 5 mm sliding window) were smaller than the normative range. Hippocampi with at least three of such points were termed to have “local atrophy”. Results: Volumetric values from HippoSeg showed significant global hippocampal atrophy in 83.6% of radiologically-confirmed sclerotic, and 2% of control subjects’, hippocampi. Including our presented local atrophy measure, keeping the control rate detection at 2%, increased automated detection to 85.9%. Fig. 1 shows an example where focal atrophy was detected in the anterior right hippocampus, showing the median (black plain line) and normative range (red area) from the 50 controls, and the patient’s profile (dashed blue line). Conclusions: Manual hippocampal volume profiling has previously shown potential to improve the distinction between healthy and sclerotic hippocampi but is very labor-intensive. Our work demonstrates the feasibility of automating this process to increase atrophy detection rates in a time- and cost-effective way. Funding: This work is funded by the National Institute for Health Research University College London Hospitals Biomedical Research Centre (NIHR BRC UCLH/UCL High Impact Initiative). We are grateful to the Wolfson Foundation and the Epilepsy Society for supporting the Epilepsy Society MRI scanner. GPW was supported by an MRC Clinician Scientist Fellowship (MR/M00841X/1).