Performance analysis of transfer learning based deep neural networks in Alzheimer classification

Abstract

Medical image analysis using deep learning techniques found good attention to diagnose critical diseases within a shorter time and recommendable performance in the identification of disease conditions. Early detection of this disease has a way of doing the treatment effectively if it is possible to identify it before the symptoms appear. In this study, different methods are being proposed with their performance analysis using deep neural networks to diagnose the different stages of Alzheimer’s disease. The dataset used in this study was collected from the kaggle repository and consists of 3 different classes of Alzheimer’s disease which include Very Mild Demented, Mild Demented and Non Demented. In this study, VGG19 and ResNet50 pre-trained models with fine-tuning were used to classify different stages of the disease, alongside other two deep neural networks used where these VGG19 and ResNet50 pre-trained models were used as a feature extractor. Finally, an AlzheimerNet proposed, which outperformed previously mentioned methods that achieved 96.41% accuracy, 97% precision, 96% recall and F1- score. The current findings of the study indicate deep learning-based method achieved significant improvement in classifying Alzheimer’s disease in its early stage.