Improved BOLD Detection with Sliced Inverse Regression

Abstract

Functional magnetic resonance imaging (fMRI) has been effective in linking task-related brain responses to changes in blood oxygen level density (BOLD). However, its reliance on BOLD measurements makes it vulnerable to artifacts and false-positive signals. Commonly, researchers use many regressors in a General Linear Model to filter true signals, but this adds noise and complicates interpretation. In this paper we suggest using Sliced Inverse Regression (SIR) to simplify covariate dimensionality and identify relevant regressors. We compare a general linear model applied to both original and SIR-adjusted data, demonstrating that SIR improves signal detection, reduces noise, and yields statistically significant results even with conservative measures.