Independent component imaging of disease signatures

This paper describes a neural computation approach to independent component imaging of disease signatures. The novel feature is to separate mixed imagery sources blindly over an informative index subspace. The recovery of patterns is achieved by independent component analysis, whose parameters are estimated using the infomax principle. We discuss the theoretic roadmap of the approach, and its applications to the partial volume correction in cDNA microarray expression and the neuro-transporter binding separation in positron emission tomography.