Learning Personal Representations from fMRI by Predicting Neurofeedback Performance

Abstract

We present a deep neural network method that enables learning of a personal representation from samples acquired while subjects are performing a self neuro-feedback task, guided by functional MRI (fMRI). The neurofeedback task (watch vs. regulate) provides the subjects with continuous feedback, contingent on the down-regulation of their Amygdala signal. The representation is learned by a self-supervised recurrent neural network that predicts the Amygdala activity in the next fMRI frame given recent fMRI frames and is conditioned on the learned individual representation. We show that our personal representation, learned solely using fMRI images, improves the next-frame prediction considerably and, more importantly, yields superior performance in linear prediction of psychiatric traits, compared to performing such predictions based on clinical data and personality tests. Our code is attached as supplementary and the data would be shared subject to ethical approvals.