Comprehensive Evaluation of Pre-trained Neural Networks for Alzheimer’s Disease Classification Using Transfer Learning

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that requires early and accurate diagnosis for effective intervention. Recent advancements in deep learning and convolutional neural networks (CNNs) have shown significant potential in magnetic resonance imaging (MRI-based) AD classification. This study aims to evaluate the classification performance of three pre-trained CNN architectures, Visual Geometry Group 16-layer network (VGG16), densely connected convolutional network with 121 layers (DenseNet121), and residual neural network with 50 layers (ResNet50), in distinguishing AD stages using T1-weighted MRI data and transfer learning (TL). MRI scans were obtained from the Alzheimer’s Disease Neuroimaging Initiative, covering five classes: cognitively normal, early mild cognitive impairment, mild cognitive impairment, late mild cognitive impairment, and AD. After preprocessing and data augmentation, models were trained and validated using stratified fivefold cross-validation. VGG16 achieved the highest accuracy (95.0%), followed by DenseNet121 (90.4%) and ResNet50 (62.0%). Evaluation metrics included precision, recall, F1-score, confusion matrix, and receiver operating characteristic-area under the curve. The results support the effectiveness of TL for AD classification, particularly with VGG16. Future research may explore ensemble learning (e.g., VGG16 + DenseNet121), multimodal data fusion, and deployment in clinical settings.