Classifying schizophrenic and controls from fMRI data using graph theoretic framework and community detection

• Published in: Network modeling and analysis in health informatics and bioinformatics (Wien) Vol. 12; no. 1; p. 19
• Main Authors: Nallusamy, Kalaichelvi, Easwarakumar, K. S.
• Format: Journal Article
• Language: English
• Published: Vienna Springer Vienna 04.04.2023; Springer Nature B.V
• Subjects: Algorithms; Applications of Graph Theory and Complex Networks; Bioinformatics; Brain; Brain mapping; Brain research; Classification; Computational Biology/Bioinformatics; Computer Science; Deep learning; Disease; Disintegration; Functional magnetic resonance imaging; Graph theory; Graphical representations; Health care; Health Informatics; Labeling; Magnetic resonance imaging; Medical imaging; Mental disorders; Neural networks; Neuroimaging; Neurosciences; Original Article; Public health; Schizophrenia; Similarity; Social networks; Social organization; Community detection; Schizophrenia; Graph similarity; Social network; Connectome; Brain network
• ISSN: 2192-6670; 2192-6662; 2192-6670
• DOI: 10.1007/s13721-023-00415-4

Schizophrenia is a psychiatric disorder characterized by symptoms such as disorganized thinking, hallucinations, disintegration of reality perception, and delusions, among others. Resting-state functional magnetic resonance imaging is a promising method for studying changes in functional brain networks in schizophrenic patients. Graph theoretic representations can effectively distinguish between healthy and schizophrenic subjects. The process of grouping users with similar interests in social networks, which can also be used to group diseased subjects, is known as community detection. In this paper, we propose a method for classifying schizophrenia and normal subjects from fMRI images by employing graph similarity and community detection algorithms. The fMRI images are first preprocessed to remove noise, and then the automated anatomical labelling atlas is used to divide the human brain into 116 regions. Following that, a region connectivity matrix is constructed, and a weighted undirected graph is generated from the connectivity matrix. The graph similarity algorithm is then used to determine the similarity between each graph or subject. Then, a network of networks is built, which is a weighted network in which each graph is a node, and the top k (threshold) similarity scores between the graphs form the graph’s edges. On the newly constructed weighted graph, a community detection algorithm is used to detect communities that classify schizophrenia and normal subjects. We applied this proposed method to the COBRE dataset, which is publicly available and consists of 72 schizophrenic patients and 74 healthy subjects. We achieved an accuracy of 86.5% and compared it to other graph-based methods.