Abstract:
Alzheimer's Disease (AD) is a pervasive neurodegenerative ailment, affecting a vast
population globally and posing challenges for early and precise diagnosis within medical
image analysis. Although machine learning and deep learning have emerged as competent
methodologies for AD detection, several obstacles persist, especially with imbalanced
datasets and convolutional effectiveness. This research thesis is using deep learning models
empowered with transfer learning to efficiently detect the Alzheimer’s disease classes. More
precisely, by employing fine tuning a pre-trained VGG16 and Inception V3 model is
investigated for multi-class classification. The study's paramount objective is to enhance AD
detection via a custom fine-tuning framework. In this study CNN deep learning models VGG
16 (sequence 1) and Inception V3 (sequence 1 & 2) are proposed for classifying AD in to
four stages i.e. Non-demented, Very Mild-demented, Mild-demented and Moderate-demented
using brain MRI scans. Then, their performance is evaluated using certain metrics such as
accuracy, loss, precision, recall, f1-score, Matthew’s correlation coefficient and balanced
accuracy. The results showed that the proposed models, VGG16 (sequence 1) and Inception
V3 (sequence 1 & 2) outperformed many of the state-of-the-art models by achieving testing
accuracies of 93.9% and (93.87%, 93.09%) for Kaggle MRI dataset.
