Predicting Progression to Alzheimer’s Disease Using FLAIR MRI Biomarkers

<p dir="ltr">Alzheimer’s Disease (AD) is the most prevalent form of dementia, and its effects are irreversible. The prodromal phase is known as Mild Cognitive Impairment (MCI) and presents as an opportunity for early intervention for AD. Individuals with MCI experience cognitive deficits, but have not yet reached the severity of AD. MCI subjects who progress to AD are known as converting MCI (cMCI), while those who are mildly impaired but do not develop AD are classified as stable MCI (sMCI). Fluid Attenuated Inversion Recovery (FLAIR) Magnetic Resonance Imaging (MRI) is routinely used in clinical practice due to its high potential for translation. Various technologies have investigated the identification of sMCI and cMCI using T1-weighted MRI, Cerebrospinal Fluid (CSF), cognitive, and multi-modal biomarkers. This thesis focuses on the development of quantitative FLAIR-only biomarkers to facilitate the identification of cMCI subjects in a large multi-centre cohort, both cross-sectionally and longitudinally. Initially, segmentation of the Lateral Ventricle Volume (LVV) was performed. The optimal model employed a 2D U-Net that was first trained on Silver Standard (SS) masks generated using an image-processing technique specific to the target dataset. The model was then fine-tuned using gold standard (GS) masks from the source dataset. This model outperformed the U-Net trained solely with GS masks, achieving a mean Dice Similarity Coefficient (DSC) of 0.89 and a coefficient of variation of 0.05. Additionally, it demonstrated a lower Average Volume Difference (AVD) and a higher correlation when comparing the predicted and ground truth LVVs. Next, six volume, three texture, and three intensity biomarkers, along with two composite indices and a brain age biomarker, were computed. Each biomarker underwent validation, showing significant correlation with existing AD conversion indicators and/or significant differences between cognitive groups. Finally, the biomarkers were utilized for cognitive classification, longitudinal analysis, and survival analysis to assess their ability to differentiate between sMCI and cMCI subjects. The FLAIR MRI biomarkers achieved accuracies of 86.2% and 69.2% in classifying between normal control and AD, and sMCI and cMCI, respectively. Longitudinal analyses revealed several biomarkers exhibiting significantly greater rates of change in the cMCI group compared to the sMCI group. Kaplan-Meier curves generated for survival analysis demonstrated that the biomarkers exhibited divergent trends between the groups up to four years prior to AD conversion. This same trend was observed when comparing the Brain Age Gap Estimation (BrainAGE) between the sMCI and cMCI subjects at one-year intervals. Overall, this work highlights the utility of FLAIR MRI and its unique insights into neurodegeneration and the conversion from MCI to AD.</p>