Table 2.

Comparison of detrending performance of in vivo real-time data. In general, and for the right VOI in particular, we observed a better performance of iGLM/iGLM^window compared to the EMA method. Parameter estimates (ß) and t-values resulted from GLM estimation (MATLAB's fitlm using the experimental paradigm and constant term as design matrix). Average runtime was measured for the first 10 and all 150 TRs of a single time series and a volume with 90 × 90 × 15 voxels.

Method	left VOI	right VOI	Timeseries n = 10, 150 TRs	Volume n = 10, 150 TRs
EMA	ß = +1.58, t = 10.64	ß = +0.51, t = 2.97	<1 ms, <1 ms	289 ms, 760 ms
iGLM	ß = +1.68, t = 11.88	ß = +1.27, t = 5.43	3 ms, 4 ms	456 ms, 890 ms
iGLMwindow	ß = +1.85, t = 13.73	ß = +1.32, t = 6.29	1 ms, 1 ms	482 ms, 508 ms
MATLAB Detrend	ß = +1.78, t = 11.12	ß = +1.21, t = 4.80	<1 ms, <1 ms	>1000 ms
SPM8 Detrend	ß = +1.83, t = 12.86	ß = +1.23, t = 5.32	<1 ms, <1 ms	472 ms, 884 ms