CC- GCN: A novel graph-based approach for identification and detection of autism spectrum disorder

The brain is one of the most complex and essential organs in the human body, and any damage to it can have significant consequences on a person’s health and well-being. Neurological disorders are medical conditions affecting the brain, nerves, and spinal cord. They can manifest in a variety of ways, including changes in behavior and cognitive function, making them particularly challenging to diagnose and treat. Researchers have developed deep learning and machine learning frameworks to analyze brain data and predict/ classify diseases. Graph Neural Networks (GNNs) are powerful deep learning techniques for processing graphical datasets. One of the GNN techniques, namely Graph Convolutional Networks (GCNs) outperforming other state-of-the-art algorithms in many tasks, such as node classification, link prediction, and graph classification. In this study, the focus is on detecting Autism based on MRI data. Graphs are constructed where each node represents the MRI data of a patient, and edges are constructed based on MRI feature similarities. A novel training sample selection technique is proposed, which is based on Clustering coefficient scores instead of traditional random selection. The proposed clustering coefficient Graph Convolution Network (CC- GCN) method generates more structurally rich computational graphs for GCN models, leading to better training, node embeddings, and hence, better outcomes for node classification. The proposed method is compared with the other state-of-the-art approaches. CC- GCN model’s performance has been evaluated over two benchmark Autism datasets, ABIDE-1 and ABIDE-2, in which \({\textbf {96.85}}\%\) $\mathbf{96.85}\%$ and \({\textbf {99.1}}\%\) $\mathbf{99.1}\%$ classification accuracy has been achieved, respectively. These results outperform state-of-the-art GCN-based methods and can be applied to other disease classification tasks. The proposed CC-GCN approach has shown a lot of promise in the research of brain illnesses, and it has the potential to help in the creation of personalized diagnostic and therapeutic approaches for people with these disorders.