Developing an anatomically valid segmentation protocol for anterior regions of the medial temporal lobe

Abstract Background The anterior portion of the MTL is one of the first regions targeted by pathology in sporadic Alzheimer’s disease (AD) indicating the potential for imaging metrics from this region to serve as valuable imaging biomarkers. However, most existing automated approaches for MTL segmentation do not incorporate anterior MTL subregions, and the few that do fail to account for its complex anatomical variability. Leveraging a unique postmortem dataset consisting of histology and structural MRI scans we aimed to develop an anatomically valid segmentation protocol for anterior entorhinal cortex (ERC), Brodmann Area (BA) 35, and BA36 and apply it for automated MTL segmentation of in vivo 3 tesla (T) MRI. Method We included 20 cases between 61 to 97 years of age (50% females) with and without neurodegenerative diseases (11 vs. 9 cases) to ensure broad generalizability of the developed protocol. Postmortem digitized MTL Nissl‐stained coronal histology serial sections from these cases were registered to same‐subject 0.2×0.2×0.2‐mm 3 9.4T postmortem MRI and annotated by an expert neuroanatomist. To develop the segmentation protocol, we determined the location of the histological borders of interest in relation to anatomical landmarks observable on in vivo MRI. The protocol was first applied manually to 29 3T in vivo MRI scans and then used to train an automatic segmentation method T1‐ASHS (Automatic Segmentation of Hippocampal Subfields). Intra‐rater reliability of a manual rater and five‐fold cross‐validation accuracy of T1‐ASHS were assessed with the Dice Similarity Index (DSI). Result Segmentation rules for the borders of ERC, BA35 and BA36 based on systematic analysis of inter‐landmark distances on histological sections are shown in Figure 1. Intra‐rater reliability for the manual rater applying these rules to 15 in vivo 3T MRI scans was high (Table‐1; Figure‐2). Comparing manual segmentations with the automated ones generated by T1‐ASHS showed moderate reliability, reflecting the challenging anatomy of this region. However, segmentation accuracy for the whole MTL including the newly added region was comparable to the previously reported accuracy for MTL without this region (Table‐1). Conclusion Future work will examine trhe utility of morphometric measures of anterior MTL regions enabled by this protocol for early AD.