210 likes | 385 Views
A Signal Processing Model for Arterial Spin Labeling Perfusion fMRI. Thomas Liu and Eric Wong Center for Functional Magnetic Resonance Imaging University of California, San Diego. Wait. Tag by Magnetic Inversion. Acquire image. Wait. Control. Acquire image. Arterial Spin Labeling (ASL).
E N D
A Signal Processing Model for Arterial Spin Labeling Perfusion fMRI Thomas Liu and Eric Wong Center for Functional Magnetic Resonance Imaging University of California, San Diego
Wait Tag by Magnetic Inversion Acquire image Wait Control Acquire image Arterial Spin Labeling (ASL) 1: 2: Control - Tag µ CBF
Goal: Accurately measure dynamic CBF response to neural activity From C. Iadecola 2004
Example: Perfusion and BOLD in primary and supplementary motor cortex. Measured with PICORE QII with dual-echo spiral readout. Obata et al. 2004
ASL Data Processing • CBF = Control - Tag • An estimate of the CBF time series is formed from a filtered subtraction of Control and Tag images. • Use of subtraction makes CBF signal more insensitive to low-frequency drifts and 1/f noise.
Pairwise subtraction example Control Tag +1 -1 +1
Surround subtraction TA = 1 to 4 seconds Control Control Control Control Tag Tag Tag +1/2 -1 +1/2 -1/2 1 -1/2 Perfusion Time Series
Generalized Running Subtraction ycontrol +1 yperf Low Pass Filter Upsample ytag 1.0
Questions • What is the difference between the various processing schemes? • How do they effect the estimate of CBF? • What are the noise properties of the estimate?
=1 presaturation applied • = 0 No presat Tag : n even Control: n odd is the inversion efficiency ideal inversion: =1
Tag : n even Control: n odd Pairwise Subtraction Surround Subtraction Sinc Subtraction
Demodulate Modulate
Perfusion Estimate Demodulated and filtered perfusion component Modulated and filtered BOLD component Modulated and filtered noise component
Perfusion Component BOLD Component
Summary • For block designs with narrow spectrum, use surround subtraction or sinc subtraction • For randomized designs with broad spectrum, use pair-wise subtraction. • To minimize noise autocorrelation use pair-wise or surround subtraction. • General framework can be used to design other optimal filters.