Johan Martijn Jansma1, Jacco A. de Zwart2, Peter van Gelderen2, Wayne C. Drevets1, Maura L. Furey
1SNMAP/NIMH, NIH, Bethesda, MD, USA; 2AMRI/NINDS, NIH, Bethesda, MD, USA
Because BOLD response timing can vary across regions and subjects, model free fMRI signal analysis through deconvolution (a.k.a. FIR) analysis can be preferred above methods that apply an a-priori BOLD model. A disadvantage is the larger number of regressors, which can increase regressor dependency. The effect of regressor dependency (expressed as the average tolerance or aTOL) on sensitivity and reliability of multi-regressor-based deconvolution analysis is investigated here. Comparison, through simulations and experimental data, of 128 random regressors with a pseudo-random m-sequence design indicated that performance of the design scaled with aTOL and that the m-sequence consistently outperformed all random designs.