A novel convolutional neural network architecture for Alzheimer’s disease classification using magnetic resonance imaging data

Abstract

Accurate categorization of Alzheimer’s disease is crucial for medical diagnosis and the development of therapeutic strategies. Deep learning models have shown significant potential in this endeavor; however, they often encounter difficulties due to the intricate and varied characteristics of Alzheimer’s disease. To address this difficulty, we suggest a new and innovative architecture for Alzheimer’s disease classification using magnetic resonance data. This design is named Res-BRNet and combines deep residual and boundary-based convolutional neural networks (CNNs). Res-BRNet utilizes a methodical fusion of boundary-focused procedures within adapted spatial and residual blocks. The spatial blocks retrieve information relating to uniformity, diversity, and boundaries of Alzheimer’s disease, although the residual blocks successfully capture texture differences at both local and global levels. We conducted a performance assessment of Res-BRNet. The Res-BRNet surpassed conventional CNN models, with outstanding levels of accuracy (99.22%). The findings indicate that Res-BRNet has promise as a tool for classifying Alzheimer’s disease, with the ability to enhance the precision and effectiveness of clinical diagnosis and treatment planning