| dc.description.abstract | Alzheimer’s disease poses a critical challenge to public health with an increasing prevalence among the aging population worldwide. The research question is whether machine
learning-based solutions could be a reliable, cost-effective, and non-invasive alternative to
existing biomarker tests. This thesis presents two machine learning-based approaches to
diagnosing Alzheimer’s disease using Magnetic Resonance Imaging (MRI) and blood-based
biomarkers. The first approach aims to train machine learning models on volumes of brain
regions from MRI to classify patients into three classes: Alzheimer’s Dementia (AD), Mild
Cognitive impairment (MCI) and Normal Control (NL). Pretrained weights of a well-known
CNN-based brain segmentation model were used in segmenting the hippocampal, parahippocampal, ventricles, entorhinal and cerebral white matter from MRI of patients, and their
volumes were estimated. The volumes and demographic data of the patients were subsequently trained on SVM and KNN models, and their performance was recorded.
The second approach aims to design efficient feature selection methods to identify relevant feature panels to identify individuals in the early stages of Alzheimer’s accurately. Two
feature selection methods were introduced. The first method ranks features according to
their dependence on the diagnosis, determined using metrics such as Mutual Information,
Symmetric Uncertainty and Cramer’s V. Panels are formed in this method by iteratively
selecting the top features and increasing the panel size. The second method filters out irrelevant features using the Euclidean distance between the class means of each feature and
applying a threshold. [...] | en_US |
