Lecture Goals Other non-bold techniques (T2 weighted, Mn contrast agents, SSFP, Dynamic Diffusion, ASL) Understand Basic Principles in Spin labeling : spin inversion, flow vs. perfusion ASL variations Quantification of perfusion from ASL data Functional ASL analysis: Detection and Quantification Some Examples The niche for functional ASL
Dispersion: (t e -kt ) Relaxation: (e -t/t1a ) The ASL signal Arterial signal Parenchy mal signal Venous signal Input Function (Label) Capillary Bed Retention (f e -(f/l)t ) Relaxation (e -t/t1 )
Two Compartment Model Magnetic Resonance in Medicine 50:599 607 (2003
Quantification of perfusion dm tissue (t) dt (the model) = f M art (t) f λ M tissue(t) M tissue(t) T 1 M tissue accumulated tissue magnetization tag (subtraction of control and tagged images) M art incoming arterial magnetization tagmust account for inversion efficiency and decay (2a e -(D/T1art) M art (0)) l: blood brain partition coefficient f: perfusion rate T1 : longitudinal relaxation rate of the tissue Williams DS, Proc. Natl. Acad. Sci., 89:212-216, 1992.) Another way to look at it M tissue (t) = f M art (t) *[r(t)m(t)] r(t) : tissue retention function ( e -ft ) m(t): T1 decay of tag in the tissue (e -t/t1 ) Buxton, MRM, 40: 383-396 (1998)
Quantification of perfusion Continuous ASL: steady state solution yields a simple equation Pulsed ASL : not as simple. You really need fit the equation for the different stages of the passage... There are approximations, such as assuming that you sample during the uptake portion of the passage: references Williams Proc. Natl. Acad. Sci., 89:212-216, 1992 Kim, MRM, 34: 293-301, 1995. Buxton, MRM, 40: 383-396 (1998) Yang, MRM, 39, 1998, p.825 f = λ T 1app M ss1 ss2 ( b M b ) M ss2 b + 2 α$ 1 ( ) M b ss1 ( f = M M control tag)λ ΔR 1 2M 0 1 e (t τ )/ ΔR 1
Lecture Goals Other non-bold techniques (T2 weighted, Mn contrast agents, SSFP, Dynamic Diffusion, ASL) Understand Basic Principles in Spin labeling : spin inversion, flow vs. perfusion ASL variations Quantification of perfusion from ASL data Functional ASL analysis: Detection and Quantification Some Examples The niche for functional ASL
Example: Functional ASL Raw (unsubtracted) time course
A Linear Model for ASL Base image BOLD effects ASL at rest ASL during activation
Analyzing ASL based FMRI signals It is very common to difference the data and/or apply filters. To difference or not to difference? Warning: AR noise is still present Differencing Amplifies (high frequency noise) transitions. Short Answer: It is preferable to model the sources than to modify the data. Greater estimation efficiency and power. Mumford, J.A., Hernandez-Garcia, L., Lee, G.R., Nichols, T.E., 2006. Estimation efficiency and statistical power in arterial spin labeling fmri. Neuroimage 33, 103-114.
Analyzing ASL based FMRI signals Differencing the timecourse: perfusion is proportional to the difference between control and tag Differencing schemes help you clean up the data pairwise y 2 [m] = y c [m] y t [m] " $ D 2 = $ $ # $ 1 1 0 1 1 0 1 1 % & running y 3 [n] = ( 1) n (y[n] y[n + 1]) surround " $ $ D 3 = $ $ $ # y 4 [n] = ( 1) n (2y[n] y[n + 1] y[n + 2]) 1 1 0 1 1 1 1 0 1 1 " $ $ D 4 = $ $ $ # % & 1 2 1 1 2 1 1 2 1 1 2 1 % &
Effect of Differencing ASL Data (frequency content)
The GLM Equations in ASL Simple GLM: Y = Xb + e Differencing the data (and the design matrix) DY = DXb + De Prewhitening : Generalized Least Squares WY = WXb + We
GLS Analysis of un-differenced data vs. OLS Analysis of differenced data
Quantifying dynamic perfusion Traditional Approach: changes 1. Calculate perfusion from each subtraction pair using tracer kinetics Wang J, et al. (2002). Magn. Reson. Med., vol. 48, no. 2, pp. 242-254. 2. Calculate difference in perfusion between conditions. 3. Calculate means and variances GLM Approach: 1. Translate GLM parameters (betas) into perfusion. 2. Translate variance estimates (sigmas) into perfusion effects variance Hernandez-Garcia L, et al. (2010), Magn. Res. Imaging, 28 (7), Pages 919-927.
Quantifying dynamic perfusion changes ˆf effect f = λ ΔM ˆβ effect R 1app R 1app %% M ˆβ ( & e TR R 1 ) * α e δ R 1a (δ w) R 1app (δ τ w) R 0 ( 1app 1 e TR R 1 * & ) 2 α e δ R 1a e (δ w) R 1app e (δ τ w) R 1app (( ) f effect is the perfusion change due to the effect of interest, a is the inversion efficiency, b effect is the parameter estimate of the regressor representing the effect of interest, l is the blood brain partition coefficient, R 1, R 1a, R 1app are longitudinal relaxation rates of arterial blood, tissue, and tissue in the presence of perfusion. d is the arterial transit time, TR, w, and t are repetition time, post labeling delay, and labeling duration.
Quantifying the variance of those dynamic perfusion changes ˆσ 2 f effect $ ˆf 2 = effect ˆσ ˆβ0 & % ˆβ 0 ) ( 2 $ ˆf 2 + effect ˆσ ˆβeffect & % ˆβ effect ) + 2 COV ˆβ 0, ˆβ effect ( ( ) ˆf effect ˆβ effect = & ˆβ 0 ( 1 e TR R 1 λ R 1app ) + * 2 α e δ R 1a e (δ w) R 1app e (δ τ w) R 1app ( ) ˆf effect ˆβ 0 = ˆβ 0 2 & ( 1 e TR R 1 λ ˆβ effect R 1app ) + * 2 α e δ R 1a e (δ w) R 1app e (δ τ w) R 1app ( )
Maps of Perfusion Effects and their Variance
Estimates and their Variance
The whole ASL Processing Stream
Software you can use BASIL: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/basil f-asl http://www.eecs.umich.edu/~hernan/public/programs ASLtbx https://cfn.upenn.edu/~zewang/asltbx.php
Lecture Goals Other non-bold techniques (T2 weighted, Mn contrast agents, SSFP, Dynamic Diffusion, ASL) Understand Basic Principles in Spin labeling : spin inversion, flow vs. perfusion ASL variations Quantification of perfusion from ASL data Functional ASL analysis: Detection and Quantification Some Examples The niche for functional ASL
Real Time F-ASL (pcasl) Motor Visual Paradigm Current Raw image Current Subtracted image Statistical Map Time courses
Example: Working memory training study M. Buschkuehl, et al Neural effects of short-term training on working memory. Cogn. Affect. Behav. Neurosci., pp. 147 160, 2014. How does cognitive training affect performance? What happens to your brain when you practice? Resting Activation 52
Working memory training study: the experiment Paradigm: blocks of rest, 4-back and 1-back Imaging with pseudo-continuous ASL Working Memory Load: 4-back 1-back 1. Test Performance (while collecting ASL images) 2. Train for one week, 20 mins. per day 3. Test Performance (while collecting ASL images) 53
Working memory training study: Effect of training: bigger activations 54
Working memory training study Effect of training: Resting Perfusion Change Note: there was a correlation between performance and resting CBF in frontal regions! 55
Lecture Goals Other non-bold techniques (T2 weighted, Mn contrast agents, SSFP, Dynamic Diffusion, ASL) Understand Basic Principles in Spin labeling : spin inversion, flow vs. perfusion ASL variations Quantification of perfusion from ASL data Functional ASL analysis: Detection and Quantification Some Examples The niche for functional ASL
Why ASL? Quantification. Quantification: BOLD : a complicated mix of parameters Perfusion: Single physiological parameter Real Time neuro-feedback : How do you evaluate whether subject s brain is doing the right thing? 1. Brain activity matches a specific spatial pattern 2. Brain activity matches a specific temporal pattern 3. Amount of activity in a given region
Why ASL? The BOLD Signal Drift Autoregressive structure BOLD signals are inherently drifty because: Physiological effects Respiration Heart beat Scanner effects Temperature
ASL Perfusion vs. BOLD Very Low Task Frequency Wang, J., G. Aguirre, et al. (2003). "Arterial spin labeling perfusion fmri with very low task frequency." Magn Reson Med 49(5): 796-802. Treatment effects START in this range!
Event Related BOLD
Blocked Design BOLD: Slow Paradigm
Blocked Design FASL: Slow Paradigm
Blocked Design FASL: Slow Paradigm
ASL Perfusion fmri vs. BOLD (sensitivity) Aguirre, G., J. Detre, et al. (2002). "Experimental design and the relative sensitivity of BOLD and perfusion fmri." Neuroimage 15(3): 488-500. Single Subject Group Analysis (Random Effects)
ASL s niche Current ASL techniques have lower SNR and Speed than BOLD BOLD breaks down in slow paradigms because of autoregressive drifts Applications Drug Studies Attention Cognitive Training Slow paradigms Mental State studies Population Studies anything SLOW or requiring Quantification!
Some References D. C. Alsop, et al., Magn. Reson. Med., vol. 73, no. 1, pp. 102 116, Apr. 2014. Aguirre, G., Detre, J., Zarahn, E., & Alsop, D. (2002) Neuroimage, 15(3), 488-500. Mumford, J. A., Hernandez-Garcia, L., Lee, G. R., & Nichols, T. E. (2006). Neuroimage, 33(1), 103-114. Liu, T., & Wong, E. (2005). Neuroimage, 24(1), 207-215. Parkes, L. M., Rashid, W., Chard, D. T., & Tofts, P. S. (2004). Magnetic Resonance in Medicine, 51(4), 736-743. Petersen, E. T., Mouridsen, K., & Golay, X. (2010). Neuroimage, 49(1), 104-113. Murphy, K., Harris, A. D., Diukova, A., Evans, C. J., Lythgoe, D. J., Zelaya, F., et al. (2011). Magnetic Resonance Imaging, 29(10), 1382-1389.