Mental Disorder Detection using Non-Orthogonal Adaptive Constrained IVABMGGMM

The constrained independent component analysis (cICA) algorithm has proven remarkable effectiveness in integrating prior information into the framework of independent component analysis (ICA). However, the necessity of orthogonality for decoupling the constraints on individual sources restricts the optimization space and consequently the separation performance of cICA algorithms. To overcome this limitation, we propose a decoupling method called non-orthogonal adaptive constrained independent vector analysis with a bounded multivariate generalized Gaussian mixture model (non-orthogonal acIVABMGGMM), designed to address the drawbacks of independent component analysis in modeling multivariate data. This structure has the capability to incorporate prior knowledge regarding the sources or the mixing coefficients into the cost function of IVA. A robust decoupling strategy is needed to enable this integration. We also identify the brain regions impacted by Alzheimer’s disease by using the extracted components from our non-orthogonal acIVABMGGMM model. We demonstrate the superior performance of non-orthogonal acIVABMGGMM over the basic IVA model through experiments conducted on both simulated and real medical imaging data.