A fixed-point iteration based constrained independent component analysis and its application in fMRI

Introduction. Independent component analysis (ICA) [1] is a useful tool for fMRI data analysis, but it suffers from the intrinsic order ambiguity, which makes it tricky to pick up the interested components across different runs and makes a problem of assessing the same or similar components across subjects. One solution is to incorporate prior information in the learning process as in the constrained ICA (cICA) [2,3]. However, the original cICA depends on a learning rate, which is not easy to be tuned. In this work, we developed a fast and learning rate free cICA algorithm and validated its performance for brain activation detection. Materials and Methods. CICA can be described as an inequality constrained problem (ICP)[2,3]: maximize an object function ) (y θ , subject to 0 ) ( ≤ y q