Introduction: 

The entorhinal cortex (EC) is one of the earliest cortical regions affected in Alzheimer’s disease (AD) and serves as a key target for magnetic resonance imaging (MRI) biomarkers. However, conventional segmentation pipelines based on the Desikan–Killiany atlas do not clearly distinguish the EC from the adjacent perirhinal cortex, leading to mixed labels and reduced diagnostic sensitivity.

Methods: 

To address these anatomical ambiguities, we developed a refined EC segmentation framework that combines expert-guided anatomical correction with deep learning. FreeSurfer-derived EC labels were manually refined by removing anterior perirhinal extensions and other anatomically inconsistent regions that are functionally distinct from the EC. These expert-corrected labels were then used to train a no-new-Net (nnU-Net) model on Alzheimer’s Disease Neuroimaging Initiative 1 (ADNI1) MRI data, enabling anatomically precise and scalable EC delineation across individuals and scanners.

Results: 

The refined EC segmentation preserved anatomically valid boundaries and demonstrated stronger group-level differentiation among cognitively normal, mild cognitive impairment, and AD groups. When incorporated into volumetric and classification analyses, it provided more specific imaging biomarkers of early neurodegeneration and improved discrimination between diagnostic stages. External validation further confirmed reliable generalization across datasets.

Discussion: 

These findings demonstrate that anatomically precise and expert-informed EC delineation improves the sensitivity of MRI-based biomarkers for early AD diagnosis. The proposed framework offers a practical and reproducible approach for studying subtle cortical changes that precede overt clinical symptoms.