Surface-based multi-template automated hippocampal segmentation: application to temporal lobe epilepsy.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Epilepsy Research

Background:

• Accurate hippocampal segmentation on MRI is vital for surgical planning in drug-resistant temporal lobe epilepsy (TLE).
• Existing methods struggle with hippocampal atrophy and malrotation, common in TLE patients.
• Volume-based segmentation techniques are negatively impacted by these anatomical variations.

Purpose of the Study:

• To introduce SurfMulti, a novel hippocampal segmentation algorithm designed for improved accuracy in TLE.
• To evaluate SurfMulti's performance against established methods like FreeSurfer and Vol-multi.
• To assess the algorithm's robustness in the presence of hippocampal atrophy and malrotation.

Main Methods:

• SurfMulti integrates deformable parametric surfaces, vertex-wise texture/shape modeling, and multiple templates.
• A library from 80 healthy and 288 TLE hippocampi was used to handle inter-subject variability.
• 3D models were generated for manual labeling and collateral sulci extraction to quantify malrotation.

Main Results:

• SurfMulti achieved high accuracy in TLE patients (Dice=86.9%), comparable to healthy controls (Dice=87.5%).
• Its performance was independent of atrophy and malrotation, unlike FreeSurfer and Vol-multi.
• SurfMulti provided hippocampal atrophy measurements closest to manual volumetry.

Conclusions:

• SurfMulti demonstrates superior and robust hippocampal segmentation performance in TLE.
• The algorithm is unaffected by common TLE-related anomalies, offering reliable results.
• SurfMulti is a promising tool for accurate hippocampal volumetry and surgical target definition in epilepsy.