Diagnostic classification of ASD improves with positively sparsified DTI inter-regional correlation and...

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by impairments in social interaction and behavioral functioning. ASD diagnosis primarily relies on behavioral assessments and can vary across individuals. In this study, an ASD diagnostic model was developed using diffusion tensor imaging (DTI) derived inter-regional correlations, graph-theoretical analysis and artificial intelligence. Initially, the DTI data from 283 individuals (154 ASD and 129 typically developing) were preprocessed using a standard pipeline. Subsequently, DTI-derived parameters were extracted from 50 white matter (WM) regions. Inter-regional correlations were computed between DTI-derived parameters using Pearson correlation coefficient (PCC), Biweight midcorrelation coefficient, and Spearman rank correlation coefficient. The resulting fully correlated matrices were subsequently decomposed into positive and negative correlation matrices. These matrices were sparsified at different threshold levels to construct graph networks. Nine graph-theoretical features were calculated for each WM region, yielding 450 features, which were fed as input to artificial intelligence models, including logistic regression, support vector machine (SVM), random forest, extreme gradient boosting, and a one-dimensional convolutional neural network. The highest average 5-fold classification accuracy of 90.47% was achieved using 75 features derived from a positively correlated PCC network thresholded at 0.2 and SVM. The most discriminative features were eigenvector and pagerank centrality of the cingulum (cingulate gyrus) left, and eigenvector centrality of the superior corona radiata left, inferior cerebellar peduncle right, and cingulum (cingulate gyrus) right. These findings indicate that positively sparsified DTI inter-regional correlation networks combined with graph-theoretical analysis and artificial intelligence provide a promising framework for accurate ASD classification.