Serial correlations of fMRI time series are altered at faster TRs (<1s) due to reduced signal to noise ratio per time frame. Thus, caution is advised when utilizing serial correlation models describing long TR (2~3s) conditions with fast acquisitions. Here, we show that statistical models alternative to the commonly used first order auto-regressive (AR) model – AR2 and AR1moving-average(MA)1 model —can achieve reasonable fitting for short TR data and improve the accuracy of activation estimates. Potential model bias can be further reduced by low-pass filtering.
This abstract and the presentation materials are available to members only; a login is required.