Sparse multiway decomposition for analysis and modeling of diffusion imaging and tractography

The number of neuroimaging data sets publicly available is growing at fast rate. The increase in availability and resolution of neuroimaging data requires modern approaches to signal processing for data analysis and results validation. We introduce the application of sparse multiway decomposition methods (Caiafa and Cichocki, 2012) to linearized neuroimaging models. We show that decomposed models are more compact but as accurate as full models and can be successfully used for fast data analysis. We focus as example on a recent model for the evaluation of white matter connectomes (Pestilli et al, 2014). We show that the multiway decomposed model achieves accuracy comparable to the full model, while requiring only a small fraction of the memory and compute time. The approach has implications for a majority of neuroimaging methods using linear approximations to measured signals.