Classification of dementia-type diseases using deep learning methods

Dementia is a complex group of neurological diseases leading to impairments in memory and activities of daily living. The increase in neurological diseases with the aging world population affects the quality of life of individuals and places a huge burden on healthcare systems. Technologies such as artificial intelligence and deep learning are leading promising developments in the early diagnosis and treatment of dementia. This thesis aims to classify dementia-type diseases from magnetic resonance images using deep learning techniques. In this study, the dataset named "Alzheimer Parkinson Disease" obtained from Kaggle was used. This dataset is divided into two different data subsets as "3 Class" and "4 Class". These datasets contain magnetic resonance images of Alzheimer's patients and non-Alzheimer's patients, Parkinson's patients and non- Parkinson's patients, and healthy individuals. These images were subjected to various data preprocessing techniques for diagnostic classification of Alzheimer's and Parkinson's diseases. Then, features were extracted from the EfficientNet-B0 model using transfer learning. These features are combined with other layers to create a customized classification model for the diagnostic classification task. With the proposed model, multiple classification of magnetic resonance images was achieved with 99.34% training accuracy and 95.15% test accuracy for the "4 Class" dataset and 99.06% training accuracy and 94.90% test accuracy for the "3 Class" dataset. In addition to the high accuracy rates obtained, in this study, experiments were conducted using different preprocessing techniques, machine learning methods, pre-trained networks and various activation functions and the performances of these methods were analyzed. We anticipate that the findings obtained can contribute to future studies by establishing an important reference point in the diagnosis and diagnostic classification of neurological diseases such as Alzheimer's and Parkinson's with deep learning techniques.