Basics of Quantification of ASL

Transcription

1 Basics of Matthias van Osch Contents Thoughts about quantification Some quantification models Bloch equations Buxton model Parameters that need to be estimated Labeling efficiency M Bolus arrival time Bolus width T 1 of arterial blood/tissue T 2 of label Or use phase contrast MRA to calibrate CBF 2 2 Two kinds of quantification Multiplication factors to scale to absolute perfusion (ml/1ml/min) that are constant for the whole image Factors that differ within the ASL-image and therefore will change relative perfusion Transit time differences Labeling efficiency for different arteries Etc 3 3 1

2 Do we need absolute quantification? For many studies global scaling factors are not essential Scaling of ASL-images with respect to whole brain average How does disease affect blood flow distribution Relative perfusion of tumor compared to normal appearing GM/WM Comparison with contralateral hemisphere Correction for regional differences is often more important Absolute quantification is essential for whole brain diseases, like large vessel occlusive disease, neurodegenerative disease, sickle cell disease, etc Especially important when comparing patients to controls (differences in T 1, labeling efficiency, or brain perfusion?) 4 4 Arterial spin labeling 3) Acquire images of the brain 2) Wait 1-2 seconds for label to arrive in brain tissue 1) Label blood magnetically (inversion) Quantification LIBC of ASL 5 5 Monday, September 23, 213 Half-time of tracer Signal (%) % Time (s) The labeled spins decay with the longitudinal relaxation time T 1 (T 1,blood 165 3Tesla) 6 2

3 Arrival-time of label pcasl PASL 1-3 s Transport time to imaging slice is approximately 1 sec Probably takes another s to reach capillary bed Quantification Arterial spin labeling of ASL 7 Monday, September 23, Arrival-time of label vs decay of tracer 1 pcasl 1-3 s PASL Signal (%) Time (s) Wait long for arrival of spins in microvasculature Image as quickly possible due to loss of label Quantification Arterial spin labeling of ASL 8 Monday, September 23, Loss of label: longitudinal relaxation 1 ms 3 ms 6 ms 1 ms 15 ms 21 ms 26 ms 3 ms 35 ms Inflow of label Decay of label Angiogram Perfusion Noise The labeled spins decay with the longitudinal relaxation time T 1 (T 1,blood 165 3Tesla) 9 3

4 Principle of arterial spin labeling ASL is based on tracer kinetics to measure blood flow, similar to the nitrous oxide method, perfusion CT and DSC-MRI Use blood as a tracer by inverting its longitudinal relaxation: NO CONTRAST AGENT Half-time of tracer governed by the longitudinal relaxation time Freely diffusing tracer: accumulation of tracer reflects blood flow Kety, S. S., and C. F. Schmidt. The nitrous oxide method for the quantitative determination of cerebral blood flow in man: theory, procedure, and normal values. J. Clin. Invest. 27: , Quantification models I: Bloch Inflow of label Outflow of label Correction for differences in water content Under assumption of well mixed compartment Solution steady state: Quantification models II: Buxton Magn Reson Med Sep;4(3):

5 Quantification models II: Buxton Labeling efficiency Delivery function Normalized arterial concentration of magnetization at voxel Residue function Fraction of tagged spins arrived at t and still present at t Relaxation function Relaxation between t and t Formulation of ASL signal as a convolution Labeling efficiency Labeling efficiency of PASL Check inversion efficiency and profile in phantoms Redo these experiments in vivo, because B 1 -distribution is different in vivo than in phantoms Acquire images at different delay times (inversion recovery)

6 Labeling efficiency of PASL 3 25 MR-signal (a.u.) Inversion time (ms) Remember: switch off pre-saturation pulses and post-labeling saturation Labeling efficiency of PASL 3 25 MR-signal (a.u.) Inversion time (ms) Remember: switch off pre-saturation pulses and post-labeling saturation Remember: modulus data rectifies noise Labeling efficiency of PASL 3 25 MR-signal (a.u.) Inversion time (ms) Labeling efficiency can be very close to 1% for PASL, but also check control condition

7 Labeling efficiency of CASL and pcasl Gradient More difficult to measure, because inversion is achieved by flowdriven (pseudo-)adiabatic inversion Labeling efficiency therefore frequently determined by means of simulation of Bloch equations Which you will be doing this afternoon M M is the equilibrium signal (TR=, TE=) in a voxel containing 1% arterial blood Is therefore proportional to voxel volume Take care of scaling between scans on your scanner (M vs ASL) Measure M with PD-weighted sequence in vein Use reference ROI (WM or CSF) Use separate M scan 2 2 Separate M -scan PD-weighted sequence with similar readout as ASL Correct for differences in water content of tissue and blood (λ WM.82; λ GM.98) 1 Perform voxel-wise correction (=division) Also possible to use control image (be aware of T 1 -effects and should not be combined with background suppression!) 1 Herscovitch P, Raichle ME. J Cereb Blood Flow Metab Mar;5(1):

8 Separate M -scan Automatically correction for regional differences in T 2 (*) Assuming that T 2 (*) of ASL signal equals T 2 (*) of PD-signal (tissue dominated) 22 Errors in single echo ASL when assuming uniform T2* error map Mean relative error in S,s.e. (%) slice thickness difference ROI 5 mm 8 mm p-value frontal sinus 47 * 54 *.2 nasal cavity 37 ** 5 **. petrous bones grey matter white matter Table shows relative error in S calculation from single echo approach assuming a uniform T 2* =5 msec as compared to dual echo approach. Errors can amount to more than 5%. In areas close to air containing cavities the error is significantly reduced with smaller slice thickness. Relative error in grey and white matter at a higher level is not significant. 23 May Separate M -scan Automatically correction for regional differences in T 2 (*) Assuming that T 2 (*) of ASL signal equals T 2 (*) of PD-signal (tissue dominated) Also correction for B 1 -inhomogeneities and other scaling factors is achieved (i.e. parallel imaging) 24 8

9 Correcting B 1 -inhomogeneities 12 1 Error (%) % higher B 1 1% lower B Excitation angle (degrees) Error in CBF as a function of readout flipangle 25 7T B 1 map With correction for B T B 1 map Without correction for B

10 Separate M -scan Automatically correction for regional differences in T 2 (*) Assuming that T 2 (*) of ASL signal equals T 2 (*) of PD-signal (tissue dominated) Also correction for B 1 -inhomogeneities and other scaling factors is achieved (i.e. parallel imaging) Without correcting for voxel-wise differences in lambda, some errors are made 1 Herscovitch P, Raichle ME. J Cereb Blood Flow Metab Mar;5(1): Choice of TR TR = 2 ms 29 TR = 1 ms 29 M in CSF 25 2 MR-signal (a.u.) Inversion time (ms) Same readout as ASL Inversion pulse of background suppression Perform fit 3 1

11 M in CSF 25 2 MR-signal (a.u.) Inversion time (ms) TE TE * * T 2, CSF T 2, blood. M, CSF blood M, blood M, CSF e 87 CSF e TE TE * * T 2, CSF T 2, blood 31 Comparison of M -methods pcasl FAIR ROI in CSF Voxelwise M -map ROI in WM Voxelwise M -map correction for λ WM and λ GM Control image as estimate for M Courtesy: Chen, Wang, Detre, ISMRM 211, abstract nr 3 32 Quantification models II: Buxton Delivery function Normalized arterial concentration of magnetization at voxel

12 Input function for PASL PASL Early arrival Little relaxation of label Later arrival More relaxation of label t Transit delay Bolus width Quantification Perfusion and of permeability ASL 34 Input function in PASL For PASL a large part of the vasculature is labeled. The amount of labeled spins is dependent on the volume of the arteries in the labeling plane (e.g. curved vessels, collateral pathways, etc). Will be different for different vessels (especially anterior vs posterior) Quantification Perfusion and of permeability ASL 35 Monday, September 23, Bolus width ( Determined by width of labeling slab Or by extent of RF-coil Or is set by QUIPSS-time (temporal cut-off of the bolus) Can be measured by fitting multi-ti data Trailing edge Courtesy: Bokkers et al, AJNR Am J Neuroradiol. 28 p

13 Choice of QUIPSS time (ms, TI =24 ms) No quipps Quantification Perfusion and of permeability ASL 37 Monday, September 23, Input function for PASL and CASL PASL (p)casl t Transit delay t = Labeling duration Quantification Perfusion and of permeability ASL 38 Input function for PASL and CASL PASL (p)casl c e t T1, blood ( t) t t t t t t t c e t t T1, blood ( ) t t t t t t t Quantification Perfusion and of permeability ASL 39 13

14 How to determine t? Multi TI PASL Average arterial transit map (284 subjects) Courtesy: Petersen, Mouridsen, Golay. The QUASAR reproducibility study; NeuroImage Volume 49, Pages Input function for PASL and CASL PASL (p)casl c e t T1, blood ( t) t t t t t t t c e t t T1, blood ( ) t t t t t t t Quantification Perfusion and of permeability ASL 41 T 1, blood 4.7Tesla 3.Tesla Silvennoinen,et al, Magn Reson Med. 23,p Lu et al, Magn Reson Med. 24,p Frequently, the value is taken from literature experiments in bovine blood 42 14

15 T 1, blood 43 T 1, blood Varela et al. NMR Biomed. 211 Jan;24(1): T 1, blood measurements Males Females 21 T 1, blood (ms) T 3T 7T Zhang, Petersen, Ghariq, De Vis, Webb, Teeuwisse, Hendrikse, van Osch. Magn Reson Med 212 Quantification T1 measurement of ASL

16 Quantification models II: Buxton label r(t)=exp(-ft/λ) Residue function Fraction of tagged spins arrived at t and still present at t Tissue Outflow: f M v =f M b /λ Assumption of well-mixed compartment Quantification models II: Buxton Which T 1? T 1 of blood or of tissue? Relaxation function Relaxation between t and t Where is my label? Does it exchange immediately? 47 Exchange of label? Monitor T 2 -changes as a function of delay time GM WM 48 16

17 Two compartment model Conclusion: T 1,blood is best candidate, especially since moment of exchange is often unknown 49 Quantification models II: Buxton 1 PASL.65 CASL.82 pcasl PASL exp(-ft/λ) exp(-t/t 1,bl ) CASL 5 5 Buxton model: putting everything together Transit time Inflow of label Decay and outflow of label

18 Buxton model results CBF 18; TA=.7; tau=.8 CBF 8; TA=.5; tau=1.2 CBF 88; TA=.5; tau= Single TI Influence of width of input function.7.6 Tau =.3 s Tau =.6 s Buxton fit to multi-ti data CBF 18; TA=.7; tau=.8 CBF 8; TA=.5; tau=1.2 CBF 88; TA=.5; tau=

19 Post-processing of (multi-ti) PASL QUASAR: deconvolution with input function Voxel-wise or ROI-wise fitting of Buxton model Fixation of parameters (especially, by means of QUIPSS) See Esben s presentation of tomorrow AIF WM GM residue Petersen et al, MRM 55: (26) 55 What is recommended? White paper recommendation

20 White paper recommendation Table 3 of white paper Parameter Value λ (blood-brain partition coefficient).9 ml/g (1) T 1, blood at 3. Tesla 165 ms (2) T 1, blood at 1.5 Tesla 135 ms (3) α (labeling efficiency) for PCASL.85 (4) α (labeling efficiency) for PASL.98 (5) 1) Herscovitch P. JCBFM1985;5(1): ) Lu H. Magn Reson Med 24;52(3): ) Lu H. Magn Reson Med 23;5(2): ) Dai W. Magn Reson Med 28;6(6): ) Wong E. Magn Reson Med 1998;4(3): White paper recommendation 6 6 2

21 Recommendation Conversion to ml/1ml/min M (ASL-signal) ASL is based on inversion Labeling efficiency M : proton density scan corrected for difference in proton content between water and tissue Recommendation,, =,,,, Amount of label,, =,,,,,,, =, 1, T 1 -decay during delay time labeling 63 PLD imaging 63 21

22 Quantification too difficult? Use PC-MRI! Measure blood flow in the internal carotid arteries and calibrate perfusion maps based on a tissue segmentation of a 3D-T1 scan Aslan, et al. Magn Reson Med. 21 Mar;63(3): Is this all about quantification? No!! Think about: Dispersion of the bolus of labeled spins Dependency on cardiac cycle Motion artifacts Scanner problems (spikes, drift, instabilities) Intersubject differences Etc, etc, etc More research needed! 65 Acknowledgements Leiden University Medical Center Wouter Teeuwisse Sophie Schmid Xingxing Zhang Jeroen van der Grond Mark van Buchem University College London,UK Xavier Golay Amsterdam Medical Center, NL Sanna Gevers Aart Nederveen University Medical Center Utrecht Jeroen Hendrikse Reinoud Bokkers Esben Petersen UT Southwestern, Dallas, USA Hanzhang Lu Kiel University, Germany Michael Helle