Visual Analysis of Brain Networks Using Sparse Regression Models.

Lei Shi,Hanghang Tong,Madelaine Daianu,Feng Tian,Paul M. Thompson

DOI: https://doi.org/10.1145/3023363

2018-01-01

Abstract:Studies of the human brain network are becoming increasingly popular in the fields of neuroscience, computer science, and neurology. Despite this rapidly growing line of research, gaps remain on the intersection of data analytics, interactive visual representation, and the human intelligence—all needed to advance our understanding of human brain networks. This article tackles this challenge by exploring the design space of visual analytics. We propose an integrated framework to orchestrate computational models with comprehensive data visualizations on the human brain network. The framework targets two fundamental tasks: the visual exploration of multi-label brain networks and the visual comparison among brain networks across different subject groups. During the first task, we propose a novel interactive user interface to visualize sets of labeled brain networks; in our second task, we introduce sparse regression models to select discriminative features from the brain network to facilitate the comparison. Through user studies and quantitative experiments, both methods are shown to greatly improve the visual comparison performance. Finally, real-world case studies with domain experts demonstrate the utility and effectiveness of our framework to analyze reconstructions of human brain connectivity maps. The perceptually optimized visualization design and the feature selection model calibration are shown to be the key to our significant findings.