Investigating conversion from mild cognitive impairment toAlzheimer's disease using latent space manipulation

Alzheimer's disease (AD), a progressive neurological disorder, is the most common cause of dementia, affecting millions worldwide. It severely affects cognitive abilities, such as speech and memory. Due to its multi-modal effects, the disease has a heterogeneous progression pattern, which makes tracking the patient's cognitive impairment level challenging. Mild Cognitive Impairment (MCI) is considered an intermediate stage before AD. Early prediction of the conversion from MCI to AD is crucial to taking necessary precautions for decelerating the disease progression and developing suitable treatments. The problem of early prediction of AD among MCI patients has often been posed as a classification problem. However, the classification models may not be sufficient to disclose the underlying factors of the transition from MCI to AD. This thesis proposes an intuitive framework to investigate the underlying causes and risk factors of the conversion from MCI to AD. The proposed deep learning-based framework employs latent space manipulation techniques to obtain principal directions toward AD diagnosis for an MCI patient and analyze the changes the patient's attributes undergo during disease progression. The predictive ability of the proposed framework is evaluated by correlating the magnitude of the manipulation in the latent space of a variational auto-encoder trained with MCI and AD patients with the possibility of conversion in later stages. Experimental results show promising quantitative and qualitative outcomes on two publicly available and commonly used AD neuroimaging data sets.