A Deep Learning Approach for Alzheimer's Disease Classification from Brain MRI
Abstract
Alzheimer's Disease (AD) is a degenerative neurological disease that impacted cognition and memory and hence early and correct diagnosis was critical. The focus of this research was on classifying brain MRI images to find the severity of Alzheimer. A publicly sourced MRI dataset obtained from Kaggle that comprised augmented as well as original images used in this research. Three Neural Networks architectures were evaluated in this research that are VGG19, ResNet50, and a custom CNN. VGG19 and ResNet50 with transfer learning were used with pre-trained ImageNet weights. On the unseen original test set, the VGG19 model obtained the best classification accuracy of 98.95%, and the custom CNN obtained 97.70%, and ResNet50 95.98%. The results illustrated the efficiency of transfer learning and the excellent performance of custom architectures in medical image classification. The research also revealed the significance of applying augmented data in training and validation to enhance model generalization and robustness.