Comparison of the Performance of CNN Transfer Learning in the Classification of Alzheimer's Disease

Alzheimer's disease is a progressive neurodegenerative condition followed by a psychiatric, cognitive, and structural decline, accounting for 60%–80% of all dementia cases. The disease can be diagnosed using imaging studies, clinical assessments, and neuropsychological tests. One of the technological developments that can diagnose Alzheimer's disease with imaging is deep learning. In general, our contribution to this study is to use the CNN algorithm in classifying Alzheimer's disease by focusing on selecting the transfer learning model used. The stages of this research include dataset processing, training and test data distribution, and creation of CNN transfer learning models such as InceptionV3, Inception ResNet, VGG16, ResNet50, and Exception to evaluate the models made. As a result, at the learning stage, it can be seen that the AUC value of each model is above 0.85, which means that the model has learned well to classify Alzheimer's disease. The highest loss value was issued by the ResNet50 model of 0.875, and the lowest loss value was issued by the Inception ResNet model of 0.283. accuracy value at the testing stage, the ResNet Inception model has the highest accuracy value of 0.894, and the ResNet50 model has the lowest accuracy value of 0.603. Meanwhile, the highest AUC value was obtained by the InceptionV3 model at 0.977, and the lowest AUC value was obtained by the ResNet50 model at 0.858. At the model testing stage, it can be seen that the VGG16 model ranks with the highest accuracy value with a value of 0.927, while the lowest accuracy value belongs to the ResNet50 model.