A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Data Science and Artificial Intelligence, School of Engineering and Technology

Thesis (M. Sc.) - Asian Institute of Technology, 2023

As the longevity is increasing globally, aging and senile disease have become common natural processes. Unfortunately, one of the major health issues due to aging is dementia, with more than 60% of cases being attributed to Alzheimer{u2019}s Disease(AD). AD is a clin ical manifestation of irreversible neurological deficits and congnitive impairment which affect the daily living of a person. Patients may initially present with mild congitive impairment(MCI) which can progress over time to AD. Clinicians primarily depend on clinical findings and cognitive functional assessments for the diagnosis of MCI and AD. However, in real-world practice, acquiring clinical data involves a series of processes that demand considerable time and effort. Moreover, biomarkers such as structural magnetic resonance imaging(sMRI), cerebrospinal fluid(CSF) examination and positron emission tomography(PET) also play a supportive role in characterizing AD and other dementia diseases alongside clinical data. Nowadays, deep learning and artificial intelligence(AI) have been emerging in the medical field especially in biomedical imaging and AD classi fication using brain sMRI. Interestingly, some physical changes in the brain tissues such as atrophy in the area of hippocampus is one of the significant findings in diagnosing AD, however the whole brain may possibly include the contribution of the disease cau sation. Although there are several open source datasets available for automatic disease classification, the majority of them comprises MCI and AD patients who are diagnosed based on clinical findings and neuropsychological assessments rather than sMRI ab normalities. In addition, some previous works provide limited information regarding the data preprocessing method and there is a lack of explicit mention of the validation data whether they come from same distribution or not. This paper presents a system atic approach on data preprocessing and highlights the benefits of data preprocessing on biomedical imaging analysis. Finally, by utilizing the independent validation set, DenseNet121 achieves a classification accuracy of 63.44% for distinguishing between Normal Cognitive(NC), MCI and AD in a 3-way classification scenario. In 2-way clas sifications, the model achieves 93.49% in NC vs AD, 69.11% in NC vs MCI and 79.1% in MCI vs AD.