Machine learning methods not only can identify the discriminative features/biomarkers of brain connectome dataset between patients and healthy controls, but also can identify the affected features/biomarkers between the administration of an active drug and a placebo. However, generic machine learning algorithms (e.g., SVM, logistic regression, etc.) often ignore the structured information of domain knowledge. Thus, simple applications of these methods cannot yield desirable sensitivity or specificity.

　　In a recent study, post-doctoral fellow PU Jian, together with research staff Dr. WANG Jun from the Institute of Data Science and Technology of Alibaba Group, under the supervision of Dr. WANG Zheng at the Institute of Neuroscience, Chinese Academy of Sciences, proposed a novel method to unveil the intrinsic features of functional brain connectivity matrices of discriminative power for group comparison. Morespecifically, by encouraging co-selection of edges connected to the same node in the connectivity matrix, the discriminative edges are preserved to maximumextent.

　　The results obtained from both synthetic data and real benchmarks demonstrated that the proposed approach outperformed the competing methods under various settings. In addition to increasing the F-measure of feature selection, this approach captured the endogenous, discriminative patternsof the brain network, consistent with recent findings in biomedical literature. This data-driven technique paves a new avenue of enquiry into the inherent nature of network models for functional brain connectomes.

　　This work entitled “Discriminative Structured Feature Engineering for Macroscale Brain Connectomes” was published online in IEEE Transactions on Medical Imaging on May8, 2015. 

　　Figure (a) is the performance comparison of the proposed method and three representative approaches widely used in machine learning field;

　　Figure (b) illustrates the discriminative features of functional brain connectome between OCD patients and healthy subjects;

　　Figure (c) illustrates the discriminative features of functional brain connectomein monkeys between the administration of ketamine and saline.