Parallel Pathway Convolutional Neural Network with Low-rank Fusion for Brain Age Prediction

Abstract

Brain age prediction is important for understanding normal brain development and aging processes in humans, and may has the potential as a biomarker for identifying brain diseases. By now, how to accurately estimate brain age from functional magnetic resonance imaging (fMRI) data is still a challenging problem. In this paper, we propose a novel deep learning-based model to predict the brain age based on fMRI data. Specifically, we first designed a parallel pathway convolutional neural network, which uses two pathways to extract different features from fMRI data. Then, we fused these features by using a low-rank fusion module. Finally, we fed these fused features into a fully-connected layer to predict the brain age. We achieved better performance, compared with representative methods.