An Over-Complete Independent Component Analysis (ICA) Approach to Magnetic Resonance Image Analysis

This paper presents a new application of independent component analysis (ICA) in magnetic resonance (MR) image analysis. One of most successful applications for ICA-based approaches in MR imaging is functional MRI (fMRI) which basically deals with one-dimensional temporal signals. The ICA approach proposed in this paper is rather different and considers a set of MR images acquired by different pulse sequences as a 3-dimensional image cube and performs image analysis rather than signal analysis. One major difference between the fMRI- based ICA approaches and our proposed ICA-based image analysis is that the ICA used in the former is under-complete as opposed to the latter which uses over-complete ICA. Such a fundamental difference results in completely different applications.