A Transfer Learning Approach for Early Diagnosis of Alzheimer's Disease with MobileNetV2 Architecture

Alzheimer's disease (AD) progresses rapidly, necessitating advanced imaging techniques and automated classification systems. Deep learning models, which extract features automatically using convolutional layers, have gained popularity for disease classification without manually crafted features. This study aims to develop an automated system for classifying different stages of mild cognitive impairment (MCI) using the ADNI database, explicitly focusing on Early MCI (EMCI), MCI, and Late MCI (LMCI). We enhanced MobileNet and MobileNetV2 architectures by incorporating a customized three-layer head and implemented transfer learning to preserve the pre-trained weights. A comparative analysis was conducted between the two architectures, considering 2 -way and 3 -way classifications, with a 70/30 train/test split. The results indicate that MobileNetV2 slightly outperforms MobileNet, achieving accuracies of 96.88 % for MCI vs. LMCI, 93.47 % for MCI vs. EMCI, and 93.31% for EMCI vs. LMCI in the 2-way classification. In the 3-way classification, MobileNetV2 achieves an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 7. 1 3 \%}$</tex>.