Introducing a Deep Learning-Based Model for Classification of Alzheimer’s Disease

Alzheimer's disease, the most common cause of dementia, is a growing global health concern. Accurate and timely diagnosis of this disease, especially in the early stages, plays an important role in managing symptoms and improving the quality of life of patients. In recent years, machine learning algorithms, deep learning, and in particular convolutional neural networks, have shown great potential for automated analysis of neuroimaging with high accuracy. However, the validity of many previous studies has been questioned due to data leakage errors. Aiming to overcome the challenge of data leakage, this study presents a valid and efficient deep learning framework for distinguishing patients with Alzheimer's disease from control subjects using structural magnetic resonance images from ADNI dataset. The main innovation of this study is the design of a dedicated architecture. Instead of using pre-trained models, this architecture is designed from the ground up, using depthwise separable convolutions to strike an optimal balance between computational efficiency and feature extraction power. To validate the results, this research has avoided any data leakage. Evaluation of the final model on the test dataset showed excellent performance, achieving 95.12% accuracy and an area under the ROC curve of 0.94. These results indicate that a carefully designed architecture and a valid methodology can become a robust and reliable tool to aid in the early diagnosis of Alzheimer's disease.