Neural Network Embedding for RS-fMRI ghraps and phenotype annotations

In this work, a computational intelligence method is developed to represent brain connectivity networks of patients with autism spectrum disorders through the use of resting magnetic resonance imaging. A neural embedding network is proposed that allows vectorial representation of connectivity networks by fusion of data from two sources, images and annotations of phenotypes. The ABIDE II database is used for embedding network training. The performance of the proposed model is presented by quantifying the ability of the vectors generated to discriminate between subjects with disorder and those without. The results showed F1- score 0.86 and average AUC 0.94 in adults. This surpasses the results shown in the literature. In the case of all subjects (adults and children), yields of F1-score 0.79 and AUC average 0.77 were reached.