Abstract
Functional medical imaging promises powerful tools for the visualization and elucidation of important disease-causing biological processes in living tissue. Recent research aims to dissect the distribution or expression of multiple biomarkers associated with disease progression or response, where the signals often represent a composite of more than one distinct source independent of spatial resolution. Formulating the task as a blind source separation or composite signal factorization problem, we report here a statistically principled method for modeling and reconstruction of mixed functional or molecular patterns. The computational algorithm is based on a latent variable model whose parameters are estimated using clustered component analysis. We demonstrate the principle and performance of the approaches on the breast cancer data sets acquired by dynamic contrast-enhanced magnetic resonance imaging.