AUTHOR=Salvador Raymond , Fuentes-Claramonte Paola , García-León María Ángeles , Ramiro Núria , Soler-Vidal Joan , Torres María Llanos , Salgado-Pineda Pilar , Munuera Josep , Voineskos Aristotle , Pomarol-Clotet Edith TITLE=Regularized Functional Connectivity in Schizophrenia JOURNAL=Frontiers in Human Neuroscience VOLUME=16 YEAR=2022 URL=https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2022.878028 DOI=10.3389/fnhum.2022.878028 ISSN=1662-5161 ABSTRACT=

Regularization may be used as an alternative to dimensionality reduction when the number of variables in a model is much larger than the number of available observations. In a recent study from our group regularized regression was employed to quantify brain functional connectivity in a sample of healthy controls using a brain parcellation and resting state fMRI images. Here regularization is applied to evaluate resting state connectivity abnormalities at the voxel level in a sample of patients with schizophrenia. Specifically, ridge regression is implemented with different degrees of regularization. Results are compared to those delivered by the weighted global brain connectivity method (GBC), which is based on averaged bivariate correlations and from the non-redundant connectivity method (NRC), a dimensionality reduction approach that applies supervised principal component regressions. Ridge regression is able to detect a larger set of abnormally connected regions than both GBC and NRC methods, including schizophrenia related connectivity reductions in fronto-medial, somatosensory and occipital structures. Due to its multivariate nature, the proposed method is much more sensitive to group abnormalities than the GBC, but it also outperforms the NRC, which is multivariate too. Voxel based regularized regression is a simple and sensitive alternative for quantifying brain functional connectivity.