Deep Transfer Learning to Predict Alzheimer Disease Status

Alzheimer disease (AD) is a significant global health concern, and early diagnosis is crucial for effective intervention. In this study, we explore the use of deep transfer learning convolutional neural networks to classify AD status based on brain magnetic resonance imaging (MRI) data. Specifically, we employ two pre-trained CNN architectures: VGG19 and DenseNet201 to extract features from MRI scans of the Kaggle AD dataset. The dataset used in this study includes a diverse collection of brain MRI scans, accurately classifying individuals into different categories, including those with AD at various stages such as mild, very mild, and moderate, as well as non-AD subjects. We employ these pre-trained CNN architectures to extract features from the Kaggle AD dataset. We employed the SMOTE technique to address the dataset’s class imbalance problem. We applied deep transfer learning to customize these pre-trained models for the AD classification task. Our results display the effectiveness of transfer learning in AD classification, achieving the following accuracies: VGG19 (0.93) and DenseNet201 (0.96), remarkably, the DenseNet201 model emerged as the top performer, achieving an accuracy of 0.96.