Fully automatic segmentation of T2-weighted MR Imaging of medial temporal lobe structures as a biomarker in prodromal Alzheimer’s disease

Background: Alzheimer’s disease (AD) is a progressive neurodegenerative disorder with great impact on patients, their families and our society. To date, there is no cure available; therefore, a better understanding of its onset and disease progression is very important. The hippocampus and other medial temporal lobe structures, such as entorhinal cortex and perirhinal cortex, play an important role in memory and are affected by AD and other neurodegenerative diseases, such as semantic dementia (SD). We aimed to develop a new atlas for fully automatic segmentation of medial temporal lobe structure subfields in a clinical population. Clinical application of this atlas could help to discriminate AD, especially its prodromal form (Mild Cognitive Impairment) from a group of patients with non-neurodegenerative diseases (NND). Hippocampal subfield Cornu Ammonis 1 (CA1) and extrahippocampal structures such as entorhinal cortex (ERC) and perirhinal cortex (PRC) are thought to be mostly affected by AD. Also, changes in declarative memory related to hippocampal subfield and PRC atrophy in MCI and AD patients are investigated. Methods: In the current study, a new atlas for fully automatic segmentation of hippocampal formation body structures was created, based on a template of 40 manual segmentations of T2 weighted MR Imaging. Next, MTL structure subfields in T2 scans from AD, amnestic MCI, NND and SD patients were fully automatically segmented, and differences in subfield volume and thickness measurements were compared between groups. Finally, correlations between cognitive memory testing and subfield atrophy were determined. Results: Intra-rater reliability and Dice overlap between manual and automatic segmentations were measured and confirmed a high reliability and accuracy of the newly developed atlas, especially for large subfields such as CA1 and Dentate Gyrus (DG). Results from fully automatic segmentations of MTL subfields in T2 MR Imaging of 17 NND patients, 26 amnestic MCI patients, 33 AD patients and 22 SD patients showed significant artrophy of CA1 in aMCI, AD and SD, especially in the left hippocampus. ERC showed significant atrophy in the right hemisphere in AD patients. Perirhinal cortices were affected only in SD patients. Atrophy in CA1 was associated with impairment in episodic memory. Semantic memory impairment correlated most with PRC. Conclusion: Results of this study demonstrate that the newly created atlas could potentially be used for diagnostics of AD and its prodromal form in clinical settings. It reliably identifies especially big hippocampal subfield structures and is able to discriminate accurately between groups. The major finding of the clinical approach was significant group discrimination in CA1 subfields, especially in AD and its prodromal form. Furthermore, artrophy in this subfield correlated significantly with impairment in episodic memory.