Volumetric Measures of the Medial Temporal Structures and their Neuropsychological Correlates in Alzheimer’s Disease

Abstract

Diagnosing Alzheimer’s disease (AD) in its earliest phases is difficult. A combination of standardised cognitive testing with AD-related biomarkers such as medial temporal atrophy observed on neuroimaging scans has been proposed to improve the diagnostic process of AD detection. The availability of automated volumetric segmentation software presents as a promising avenue for the incorporation of quantitative volumetric measures into routine clinical workup. However, the performance of these automated protocols in a clinical context remains to be clarified. This thesis aimed to examine the use of automated segmentation in a “real-world” clinical setting against the gold standard of manual segmentation, assess the difficulties surrounding reliable automated measurements, and evaluate the relationship between volume, diagnosis, and memory function.

Methods utilised to address the research questions included a review of the diagnostic accuracy of various diagnostic tools, in-depth examination of structural variations that could undermine automated segmentation efforts, a series of validation studies, and assessment of medial temporal volumes and memory function in samples of clinical and healthy controls.

The results indicate that a multimodal approach to diagnosis that includes a combination of cognitive assessment and other biomarkers would likely yield a more accurate diagnosis. Although automated segmentation of the medial temporal structures was promising for the hippocampus, the entorhinal and transentorhinal cortices were inherently more challenging to delineate and produced much poorer reliability measures. Medial temporal structural volumes contributed little beyond what was already provided by comprehensive memory assessment for distinguishing between the clinical and healthy control samples. Taken altogether, these findings highlight the need for additional refinement of automated algorithms and further investigation into the ideal combination of cognitive testing and other biomarkers.