InnerEye-HS: a disease-agnostic clinical tool for hippocampal segmentation.

Area of Science:

• Neuroimaging
• Medical AI
• Biomarker Discovery

Background:

• Hippocampal atrophy is a key feature in Alzheimer's disease and temporal lobe epilepsy.
• Accurate hippocampal volume measurement is crucial for early disease detection.
• Existing automated segmentation tools struggle with disease-related hippocampal changes.

Purpose of the Study:

• To develop and validate a robust deep learning tool, InnerEye hippocampal segmentation tool (InnerEye-HS), for accurate hippocampal segmentation.
• To assess InnerEye-HS performance against manual segmentation and other leading automated tools.
• To demonstrate the clinical utility of InnerEye-HS for disease biomarker analysis.

Main Methods:

• InnerEye-HS was trained on diverse MRI scans across the Alzheimer's disease spectrum.
• The model was validated on clinical dementia and epilepsy datasets against manual segmentations.
• Performance was compared to Automatic Segmentation of Hippocampal Subfields (ASHS), FreeSurfer, FastSurfer, and HIPPOSEG using Dice scores.

Main Results:

• InnerEye-HS achieved the best Dice scores on a hospital dementia dataset (mean = 0.85 ± 0.02).
• InnerEye-HS and ASHS showed comparable top performance on an epilepsy dataset (mean = 0.85 ± 0.02 and 0.84 ± 0.03, respectively).
• A high correlation (R² = 0.85) was observed between InnerEye-HS and ground-truth hippocampal volumes, indicating robustness across disease progression.

Conclusions:

• InnerEye-HS demonstrates superior or comparable performance to existing tools for hippocampal segmentation in neurodegenerative and epilepsy conditions.
• The tool's robustness across varying hippocampal sizes and topologies highlights its clinical potential.
• InnerEye-HS offers a valuable advancement for reliable hippocampal volumetry in clinical settings.