An early classification of Alzheimer’s Disease with deep Features using Advanced Deep Learning Method (Graph Convolutional Neural Networks)

Abstract

Earlier and precise classification of Alzheimer’s Disease (AD) disease is one of the most urgent issues in neuroimaging and computational neuroscience. Traditional deep learning models like the Convolutional Neural Networks (CNNs), are useful in extracting features of medical images, but cannot often capture the intricate inter-regional interactions of the human brain. To find a solution to this constraint, this paper presents a sophisticated deep learning model based on Graph Convolutional Neural Networks (GCN) to detect and classify the Alzheimer Disease at an early stage. The method proposed creates a brain connectivity graph, each node of which corresponds to a region of interest (ROI) computed based on MRI data, and the edge corresponds to an anatomical or functional correlation between the regions. Deep feature representations are learned by using multiple layers of GCN, to learn both local and global topological statistics of disease progression. Experimental testing on the Open Access Series of Imaging Studies (OASIS) and transformed into a 4D format to a 2D format dataset shows that the proposed GCN-based algorithm is much more effective in improving the accuracy of the classification, in comparison to the traditional CNN and machine learning algorithms. The model is also biologically explanatory in that it points to significant areas of the brain that play a role in the prediction of diseases. This study has highlighted the possibility of the graph-based deep learning in improving early diagnosis and clinical decision making in the case of Alzheimer.