Abstract
Heterogeneous presentation of a neurological disorder suggests potentialdifferences in the underlying pathophysiological changes that occur in thebrain. We propose to model heterogeneous patterns of functional networkdifferences using a demographic-guided attention (DGA) mechanism for recurrentneural network models for prediction from functional magnetic resonance imaging(fMRI) time-series data. The context computed from the DGA head is used to helpfocus on the appropriate functional networks based on individual demographicinformation. We demonstrate improved classification on 3 subsets of the ABIDE Idataset used in published studies that have previously producedstate-of-the-art results, evaluating performance under a leave-one-site-outcross-validation framework for better generalizeability to new data. Finally,we provide examples of interpreting functional network differences based onindividual demographic variables.