Development of new deep learning models for detection of alzheimer's disease in magnetic resonance images

Alzheimer's disease, which is a neurological disorder, usually affects old people. This disease can be diagnosed early and some preventive measures can be taken to improve patient comfort. At this point, effective Alzheimer disease detection methods have been developed with the use of artificial intelligence and image processing methods. In this thesis, deep learning-based methods were developed to detect disease stages by using an Alzheimer MRI dataset containing 3 stages of Alzheimer's disease. The developed methods use deep learning architectures. By using conventional convolution layer, ResNet50 and Inception V3 architectures together, 5 different deep learning models have been developed. In particular, hybrid deep learning models have been developed by optimizing the special block structures of the Resnet50 and Inception V3 architectures to work together. The layer structure of the models has been optimized to interpret Alzheimer's data. Model designs are made by considering the vanishing gradient problem. In addition, undesirable situations such as over fitting and computational complexity are taken into account in the ordering and modeling of special block structures. The developed methods are compared with current deep learning methods using different performance metrics. The experimental results show that the developed methods have obtained effective results. Keywords: Alzheimers's disease detection, deep learning, Inception, ResNet.