Feature study of conversion blindness on functional network with aggregation of local key information

It remains unclear whether brain networks are altered during conversion blindness in electroencephalogram (EEG) representation, which is of significance both on improving clinical management of conversion blindness, and providing objective evidence for judicial disputes. Functional brain network was constructed on coherence extracted from scalp EEGs, and conventional network metrics were analyzed in various frequency bands. Performance indices using relevant global features were evaluated with nonlinear and linear classifiers. To complement the incompetence of global feature in differentiating conversion blindness from control, local characteristics were selected by algorithm of minimum redundancy maximum relevance. Global network features were found most pronounced in the alpha band. Further consideration of local characteristics fused with global feature for differentiating conversion blindness from control, performance elevation was achieved with total accuracy of 92.32%, sensibility of 91.29%, and specificity of 93.36%. Experimental results demonstrated the potential of the proposed approach to represent brain network of conversion blindness with symptom of complete bilateral visual loss, it also provided new perspective to understand conversion blindness.