In this study, the author propose a novel multi-channel machine learning based classification framework to identify the infants at high-risk for ASD at 6 months old. The major contribution of the study include: firstly, we develop a novel brain parcellation strategy to partition a publicly available atlas “AAL” (automatic anatomical labeling) into anatomically meaningful ROIs with adaptive sizes; secondly, unlike Ingalhalikar et al., 2012, we propose to use the features from a hierarchical set of whole-brain WM connectomes, instead of conventional region-based features, to identify ASD infants; finally, we utilize an effective two-stage feature selection scheme and multi-kernel SVM classifier that can incorporate the complementary information from multiple sources to improve the classification accuracy.