Alzheimer Disease Diagnosis Using Deep Learning Techniques

Student thesis: Master's Thesis

Abstract

This research develops a classification system for the early prediction of Alzheimer's disease to aid patients and health practitioners in preparing optimal treatment plans and carrying out clinical studies. This is achieved through the use of automatic feature extraction using 3D convolutional neural networks trained on structural MRI volumes of brain regions of interest. The predictions of the region-based CNNs (ROI CNNs) are then used to train a Support Vector Machine to perform the final classification. The proposed system achieves an accuracy of 92.62% on the diagnosis of Alzheimer's disease. This approach also predicts the conversion of Mild Cognitive Impairment to Alzheimer's disease within 24 months with an accuracy of 82.22%. Further analysis of results identifies regions which play a significant role in the progression of Alzheimer's disease, including the amygdala, hippocampus, entorhinal area, lateral ventricle, and parahippocampal gyrus.
Date of AwardJul 2021
Original languageAmerican English

Keywords

  • Alzheimer’s disease prediction
  • deep learning
  • convolutional neural networks
  • segmented structural MRI
  • ensemble learning.

Cite this

'