Agglomeration of deep learning networks for classifying binary and multiclass classifications using 3D MRI images for early diagnosis of Alzheimer’s disease: a feature-node approach

Alzheimer’s disease is a degenerative brain condition causing memory loss in the elderly. Existing machine learning methods often yield low classification accuracy due to evaluating single modality features. This paper presents a novel approach that combines Graph Attention Networks and Deep Convolutional Graph Neural Networks to leverage 3D 1.5 T and 3 T T1-weighted MRI images as nodes, enabling faster feature extraction. Three Graph Convolutional Network layers are introduced to improve the classification accuracy for three binary classifications (AD vs. CN, MCI vs. CN, and MCI vs. AD) and multiclass classification (AD vs. CN vs. MCI). The model is optimized for weight updates using the Adaptive Stochastic Gradient Descent technique. Comparative analysis with efficient 3DNET, Squeeze3DNET, and GoogLENET demonstrates superior performance of the proposed DCGNN network. Furthermore, evaluations against four state-of-the-art techniques for binary and multiclass classifications show its potential in diagnosing the early stages of Alzheimer’s disease. The developed model exhibits promise as an effective tool for diagnosing Alzheimer’s disease at its early stages.