Wednesday 18th December 2019
Ground Floor Seminar Room
25 Howland Street, London, W1T 4JG
Simultaneous Non-Gaussian Component Analysis (SING) for Data Integration in Neuroimaging
Authors: Benjamin Risk and Irina Gaynanova
As advances in technology allow the acquisition of complementary information, it is increasingly common for scientific studies to collect multiple data sets. Large-scale neuroimaging studies often include multiple modalities (e.g., functional MRI, diffusion MRI, and/or structural MRI) and behavioral data, with the aim to understand the relationships between data sets. Classical approaches to data integration utilize transformations that maximize covariance or correlation. In neuroimaging, a popular approach uses principal component analysis for dimension reduction prior to feature extraction via a joint independent component analysis (ICA). We introduce an unsupervised learning method based on non-Gaussianity for data integration. In SImultaneous Non-Gaussian component analysis (SING), dimension reduction and feature extraction are achieved simultaneously, and shared information is captured via subject scores. We apply our method to a working memory task and resting-state correlations from the Human Connectome Project. Our method finds meaningful joint structure across a working memory task and resting-state correlations, and we discover biological meaning in the joint subject scores by relating them to a measure of fluid intelligence.