Multi Time-point Arterial Spin Labeling Arterial Transit Time, Arterial Blood Volume,...

Transcription

1 Multi Timepoint rterial Spin Labeling rterial Transit Time, rterial lood Volume,... Esben Thade Petersen Department of Radiology and Department of Radiotherapy, University Medical Center Utrecht, The Netherlands cquisition, Processing and Limitations Control? Label M CF cquisition: olus width and delay Sequence differences: Mainly in the label (pseudo) Continuous Pulsed (Slab selective) Velocity Selective New cceleration Selective Readouts: Everything can be used! EPI the standard Spiral 2/3D 3D GRSE Post labeling delay Single (TI) the standard Multiple (TI) With or without vascular crushers ackground suppression: Static tissue nulling Reduced physiological noise 1

2 Kinetic modeling: Voxel (Label) (Control) rterial lood CSF (Partial volume) : rteries : Capillaries C: Veins Kinetic modeling arteries/arterioles capillaries venules/veins Tracer Delivery: Inverted blood Efflux of Unused Tracer water (Restricted) exchange rain tissue Hypothesis: H 2 O is in the fast exchange limit with tissue Time per Experiment (Label/Control): 15 s MTT ~ 37s & lood T1 ~ 3T Efflux of Unused Tracer ~ (1% exchange volume) Processing: The model uxton s Model: t M ( t) = 2 α M f c( τ ) r( t τ ) m( t τ ) dτ a, Tissue signal = rterial signal Tissue response uxton et al., MRM, 4:383(1998) 2

3 Processing: The model uxton s Model: M ( t) = 2 α M f c( τ ) r( t τ ) m( t τ ) dτ a, t Tissue signal = rterial signal Tissue response uxton et al., MRM, 4:383(1998) Processing: The model uxton s Model: M ( t) = 2 α M f c( τ ) r( t τ ) m( t τ ) dτ a, t Tissue signal = rterial signal Tissue response uxton et al., MRM, 4:383(1998) Processing: The model uxton s Model: M ( t) = 2 α M f c( τ ) r( t τ ) m( t τ ) dτ a, t Tissue signal = rterial signal Tissue response m = T1 decay uxton et al., MRM, 4:383(1998) 3

4 Processing: The model ssuming plug flow and a single compartment:, t < τ a R1 a t α e ( PSL), τ a t < τ d c( t) = R1 a τ a α e ( CSL), τ a t < τ d, t τ d r( t) = e m( t) = e f t λ R1 t t uxton et al., MRM, 4:383(1998) Processing: Single TI PSL ssuming plug flow and a single compartment:, t < τ a 2 M a, f 1 ( ) ( t ) e R a t R a ( 1 e δ t τ M = ), τ a t < τ d δ R 2 M a, f R1 a d δ R ( τ d τ a ) R1 app ( t τ e τ d ) ( 1 e ) e, t τ d δ R uxton et al., MRM, 4:383(1998) cquisition: Single TI PSL rterial Input Function Tissue Signal 4

5 cquisition: Single TI pcsl rterial Input Function Tissue Signal Processing: Single TI PSL Parameters to be assumed or measured: 1. lood T1 2. Inversion efficiency 3. Uniform plug flow (dispersed model exist) 4. Single compartment (Kety model) or multiple compartments 5. loodtissue partition coefficient 6. Transit time from label to image slice 7. Tissue T1 8. lood M Processing: Multiple TI PSL ssuming plug flow and a single compartment:, t < τ a 2 M a, f 1 ( ) ( t ) e R a t R a ( 1 e δ t τ M = ), τ a t < τ d δ R 2 M a, f R1 a d δ R ( τ d τ a ) R1 app ( t τ e τ d ) ( 1 e ) e, t τ d δ R Traditionally Least Mean Square fitting is used, but ayesian methods gain more and more attention. uxton et al., MRM, 4:383(1998) 5

6 cquisition: Multi TI PSL rterial Input Function Tissue Signal Günther M et al., MRM, 46: 974 (21) cquisition: Repeated Single TI PSL rterial Input Function Tissue Signal Günther M et al., MRM, 54: 491 (25) Processing: Multiple TI PSL Parameters to be assumed or measured: 1. lood T1 2. Inversion efficiency 3. Uniform plug flow (dispersed model exist) 4. Single compartment (Kety model) or multiple compartments 5. loodtissue partition coefficient 6. Transit time from label to image slice 7. Tissue T1 8. lood M 6

7 Processing: Multiple TI IF Non crushed data Crushed data (V enc = 3cm/s) rterial Input Function [ms] cquired data: Post processing (Deconvolution) : Result: Max = CF = 75 ml/min/1g Model! Petersen et al., MRM, 55: 219 (26) Processing: Multiple TI IF Parameters to be assumed or measured: 1. lood T1 2. Inversion efficiency 3. Uniform plug flow (dispersed model exist) 4. Fast exchange/single compartment (Kety model) some propose two compartments 5. loodtissue partition coefficient 6. Transit time from label to image slice 7. Tissue T1 8. lood M Results: verage CF, av & TT CF [ml/1g/min] av [ml/1g] Transit time [s] 7

8 Summary Multi Timepoint rterial Spin Labeling allow: Measurement of CF, av, rrival Times Improve quantification e aware of long bolus arrival times No flow or just late arrival? Slow flow did the labeled blood clear from feeding vessels? The minus is: Longer scan time or lower coverage More post processing With the risk of introducing additional noise The effect of bolus length and dispersion on rterial Spin Labeling flow quantification ackground: Global IF Target IF E.T. Petersen, X. Golay, Workshop on Cerebral Perfusion and rain Function, Salvador da ahia, razil 27 8

9 Methods: QUSR acquisition Non crushed data Crushed data (V enc = 3cm/s) rterial Input Function [ms] cquired data: Post processing: Results: Dispersion map Global bolus length Hrabe et al, JMRI 24;167:4955 Parallel Fitting Global: bolus length Local: dispersion, arrival time, amplitude Methods: QUSR TestRetest 28 Sites worldwide, TestRetest, 284 Subjects (96 scans) 7 slices (6 mm / gap=2 mm); matrix = 64x64 FOV = 24 mm; α = 11.7 & 35 ; 12 & 72 averages TR/TE = 4/23 ms; SENSE = 2.5 Time points = 13; sampling int. = 3 ms τ b = 64 ms (QUIPSSII bolus saturation) Labeling slab = 15 mm; gap = 15 mm V enc = [,4 cm/s] Results: Dispersion & bolus length Dispersion std. [ms] 9

10 Results: Site variability Site avg. GM CF Site avg. bolus length r=.63, p<.1 Consequences: CF quantification Multi TI PSL CF Error [%] ssumptions: No dispersion Right: with dispersion (σ=2ms), corresponding to what is often seen in stenotic patients. Noise free ideal situation, with known blood and tissue parameters! Consequences: CF quantification Single TI PSL CF Error [%] ssumptions: No dispersion, TT =.8s, olus length =.6s Right: with dispersion (σ=2ms), corresponding to what is often seen in stenotic patients. Noise free ideal situation, with known blood and tissue parameters! 1

11 Consequences: CF quantification (p)csl CF Error [%] ssumptions: No dispersion, TT =.8s Right: with dispersion (σ=2ms), corresponding to what is often seen in stenotic patients. Noise free ideal situation, with known blood and tissue parameters! Consequences: CF quantification CF Error [%] General simulation parameters (results are average errors within these ranges): lood T1=1.65s, tissue T1=1.2s, λ=.9, CF=26ml/1g/min, transit time=.41.s, bolus duration=.5.7s (left), dispersion std.=..2s (right). Consequences: cetazolamide challenge Pre cetazolamide GM CF = 4 ml/1g/min, bolus = 64 ms GM CF = 41 ml/1g/min, bolus = 34 ms GM CF = 47 ml/1g/min, bolus = 64 ms GM CVR = 18% Post cetazolamide (1g) GM CF = 62 ml/1g/min, bolus = 34 ms GM CVR = 51% Nyssa Craig et al., University of Sheffield, ISMRM 21 11

12 Summary on bolus duration and dispersion! Dispersion: Good or bad? Potential of introducing CF quantification errors! lso in other modalities: DSC etc? Peter Gall et al. ISMRM 21 clinical marker for vascular disease? olus duration Labeling in IC (Mean velocity ~4cm/s)Use short bolus (.5s) Careful in vascular reactivity challenges (cetazolamide/co 2 ) Variations between sites! In conclusion: Watch your bolus! It might be more dispersed than expected! It might be shorter than anticipated! M a, estimation from Multi Timepoint SL CF quantification Tissue signal rterial signal Tissue response = t a, c τ τ m( M ( t) = 2 M f ( ) r( t ) t τ ) dτ uxton et al., MRM, 4:383(1998) 12

13 CF quantification Tissue signal rterial signal Tissue response = M t M ( t) = 2 ( τ ) ( τ ) ( τ ) τ, f c r t m t d a uxton et al., MRM, 4:383(1998) lood M Estimation Curie s Law M 2 2 γ ħ = ρ 4kT Distribution Volume Ratio λ = ρ ρ, t, b M = M, t, b M a, estimate, a common problem! Original Methods Sagittal CSF, GM Sinus or WM Signal ROI Correction for difference steady between state rterial and Venous Often assume lood T1 2 and λ Problems User dependent Hard to reproduce (partial volume/location) Solution utomatic/user independent lood M estimation 13

14 dditional scaling parameters! TE t 2 * T M ( t) = 2 α Scoil e M f ( τ ) ( τ ) ( τ ) τ, c r t m t d a α is a global parameter and typically estimated once! PSL ~ 95% pcsl ~ 85% CSL ~ 7% ackground suppression reduce α further M a, is a global parameter S coil is NOT a global parameter Coil design and load spatial dependence T 2 * is NOT a global parameter Tissue and spatial susceptibility differences Together they are preferably estimated on a voxel by voxel basis! Wang et al., MRM, 53:666(25) Petersen et al. ISMRM 7 Dai et al. MRM, 66: (211) QUSR Readout (LookLocker) L = Label experiment C = Control experiment Non crushed data Crushed data (V enc = 35 cm/s) Low flip angle (1/3 of nominal) Dynamic: L C L C L C L C L C L C L C L Improved lood M Estimation M t,eff,high M t High α R 1eff,high R 1 Low α R 1eff,low ctual α 14

15 Improved lood M Estimation Mt λmap λgm =.98 λwm =.82 (literature) R1 M,blood R1 histogram fit Influence of the chosen approach Method 1: Single global Ma, Method 2: Estimate λ[x,y] Ma,[x,y] = Mt,[x,y]/λ[x,y] Method 3: Ma,[x,y] = Mt,[x,y]/λ Summary Data scaling is needed for CF quantification Ma, Tricky as other spatial variations needs to be incorporated Especially important in D where atrophy can influence your method Standard is division with proton density map (Mt,[x,y]/λ) vailable information about T1 llow tissue type classification within data This can potentially be used for partial volume correction (FRSIER). 15

16 Improve SNR and/or reduce scan time in Multi Timepoint SL The cool stuff! pcsl and time encoded pcsl label img. standard pcsl control img. label img. C control img. D label img. E control img. F label img. G control img. H ISMRM Perfusion 212and permeability 47 pcsl and time encoded pcsl label control img. img. standard pcsl img img img. C img. D img. E img. F img. G img. H 4 4 time encoded pcsl References: M.Gunther, ISMRM workshop Salvador de ahia J.Wells et al. MRM 63: (21) ISMRM Perfusion 212and permeability 48 16

17 pcsl and time encoded pcsl label control img. img. standard pcsl img img img. C img. D img. E img. F img. G img. H 4 4 img. img. time encoded pcsl = equal SNR ISMRM Perfusion 212and permeability 49 pcsl and time encoded pcsl label control img. img. standard pcsl img img img. C img. D img. E img. F img. G img. H 4 4 time encoded pcsl ISMRM Perfusion 212and permeability 5 pcsl and time encoded pcsl img img img. C img. D img. E img. F img. G img. H = nd... 1 img. 1 img. 1 img. C 1 img. D 1 img. E 1 img. F 1 img. G 1 img. H ISMRM Perfusion 212and permeability 51 17

18 pcsl and time encoded pcsl img img img. C img. D img. E img. F img. G img. H = nd... 2 img. 2 img. 2 img. C 2 img. D 2 img. E 2 img. F 12 img. G 2 img. H ISMRM Perfusion 212and permeability 52 pcsl and time encoded pcsl img img img. C img. D img. E img. F img. G img. H = nd... 3 img. 3 img. 3 img. C 3 img. D 3 img. E 3 img. F 13 img. G 3 img. H ISMRM Perfusion 212and permeability 53 pcsl and time encoded pcsl img img img. C img. D img. E img. F img. G img. H = nd... 4 img. 4 img. 4 img. C 4 img. D 4 img. E 4 img. F 14 img. G 4 img. H ISMRM Perfusion 212and permeability 54 18

19 pcsl and time encoded pcsl img img img. C img. D img. E img. F img. G img. H = nd... 5 img. 5 img. 5 img. C 5 img. D 5 img. E 5 img. F 15 img. G 5 img. H ISMRM Perfusion 212and permeability 55 pcsl and time encoded pcsl img img img. C img. D img. E img. F img. G img. H = nd... 6 img. 6 img. 6 img. C 6 img. D 6 img. E 6 img. F 16 img. G 6 img. H ISMRM Perfusion 212and permeability 56 pcsl and time encoded pcsl img img img. C img. D img. E img. F img. G img. H = 7 img. 7 img. 7 img. C 7 img. D 7 img. E 7 17 img. F img. G 7 img. H ut then in oneseventh of the total scan time ISMRM Perfusion 212and permeability 57 19

20 pplied time encoding scheme Fixed block duration (327 ms) cquisitions C D E F G H I J K L 14% 8% ISMRM Perfusion 212and permeability 58 Time encoding T1 decay compensation Fixed block duration C D E F G H I J K L djusted block duration C D E F G H I J K L Standard pcsl ISMRM Perfusion 212and permeability 59 Comparing tepcsl and pcsl Fixed block duration C D E F G H I J K L djusted block duration C D E F G H I J K L Standard pcsl ISMRM Perfusion 212and permeability 6 2

21 Results, rterial Transit Time maps N = 5 Fixed block dur TT (ms) 15 tepcsl, fixed block duration tepcsl, adjusted block duration 5 25 QUSR ISMRM Perfusion 212and permeability 61 Results, perfusion images tepcsladj (PLD in ms) 1225 ISMRM Perfusion 212and permeability 62 Results, tepcslfix versus tepcsladj Fixed djusted Time (ms) ISMRM Perfusion 212and permeability 63 21

22 Discussion, tsnr in (te)pcsl tsnr pcsl Fixed djusted ISMRM Perfusion 212and permeability 64 Discussion, caveat for block summation lock 1 = C D E F G H I J K L C D E F G H I J K L ISMRM Perfusion 212and permeability 65 Discussion, caveat for block summation lock 1 = C D E F G H I J K L lock 2 = C D E F G H I J K L C D E F G H I J K L ISMRM Perfusion 212and permeability 66 22

23 Discussion, caveat for block summation lock 1 = C D E F G H I J K L lock 2 = C D E F G H I J K L lock 3 = C D E F G H I J K L C D E F G H I J K L ISMRM Perfusion 212and permeability 67 Discussion, caveat for block summation lock 1 = C D E F G H I J K L lock 2 = C D E F G H I J K L lock 3 = C D E F G H I J K L Sum = 3 C D E F 3G H I 3J K L C D E F G H I J K L ISMRM Perfusion 212and permeability 68 Discussion, caveat for block summation Fixed block duration C D E F G H I J K L lock 1 = C D E F G H I J K L lock 6 = C D E F G H I J K L Sum = 6 2 2E 2F 4G2H 2 L ISMRM Perfusion 212and permeability 69 23

24 C D E F G H I J K L Discussion, tsnr in (te)pcsl tsnr rel. tsnr (%) 1 pcsl Fixed djusted pcsl Fixed djusted Theoretical tsnr, taking noise dependency in subtracted images into account ISMRM Perfusion 212and permeability 7 Free lunch! Standard pcsl wait wait wait wait wait wait Time encoded pcsl Gray matter perfusion map (a.u.) rterial Transit Time map (ms) ISMRM Perfusion Perfusion and permeability workshop 71 msterdam 212 Extracting the arterial signal CF, no vascular crushing arterial signal No vascular crushing: mixed signal CF, vascular crushing subtraction rterial signal With vascular crushing: tissue signal ISMRM Perfusion Perfusion and permeability workshop 72 msterdam

25 Perfusion signal in different ROI s ISMRM Perfusion Perfusion and permeability workshop 73 cquisition: Multi TI TurboQUSR rterial Input Function Tissue Signal cquisition: Multi TI TurboQUSR 25

26 cquisition: Multi TI TurboQUSR CF TT av cquisition: Multi TI TurboQUSR Turbo QUSR CF Pseudo CSL CF Conclusions: Multi TimePoint rterial Spin Labeling provides: Quantitative CF, av and rrival Times way to asses long bolus arrival times No flow or just late arrival? Slow flow did the labeled blood clear from feeding vessels? Plan your study according to subjects and scan time available White paper recommend single time point pcsl Testretest have showed reasonable reproducibility across sites making it suitable in large multicenter trials Clinical awareness of SL Is growing with the recent introduction of clinical SL packages from the vendors Multi Timepoint SL (LL) is not standard from the vendors 26

27 cknowledgments: University Medical Center Utrecht Jill De Vis Nolan Hartkamp C..T. van den erg Jeroen Hendrikse Leiden University Medical Center Matthias J.P. van Osch Wouter Teeuwisse Sophie Schmid Xing Xing Zhang University College London Xavier Golay 27