Alzheimer's Disease Detection Using Ensemble Techniques on MRI Images

Alzheimer's disease is a significant diagnostic challenge as well as the leading cause of dementia. MRI scans can reveal structural brain changes, but the manual interpretation of these scans is both subjective and time intensive. This work therefore explores an automated approach using deep learning to detect Alzheimer's from MRI images. It proposes an ensemble of three models - EfficientNet, Vision Transformer (ViT), and SqueezeNet which were each chosen for its unique strengths in image analysis. EfficientNet excels in feature extraction, ViT captures long-range dependencies within images, and SqueezeNet offers computational efficiency. These models were then trained to identify four stages of cognitive decline: Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented. By utilizing the complementary strengths of these architectures, our ensemble approach achieved an impressive 90.50% accuracy on a publicly available Kaggle dataset, outperforming individual models. This paper demonstrates both the potential of combining diverse deep learning architectures and also presents a promising step towards enhancing both the accuracy and efficiency of Alzheimer's detection from MRI images.