ASL Methods and Poten/al for phmri. Michael Kelly, PhD

Transcription

1 ASL Methods and Poten/al for phmri Michael Kelly, PhD

2 Layout Introduc/on to cerebral perfusion Arterial spin labelling Basics Techniques (con/nuous ASL, pulsed ASL, pseudo- con/nuous ASL) Quan/fica/on of CBF (Buxton model, alterna/ves) Typical setup for pcasl experiment Current applica/ons (at FMRIB) Poten/al of ASL for phmri Advantages Confounds Recent phmri ASL studies

3 Introduc/on Cerebral perfusion the process involved in the delivery of nutri/ve blood to the brain /ssue capillary bed Various modali/es: PET, SPECT, CeCT, Doppler ultrasound. Many parameters: CBF (ml/100g /ssue/min), CBV (ml/100g), MTT (s), blood velocity (cm/s).. Two main quan/ta/ve MRI techniques: DSC- MRI & ASL

4 Introduc/on Why might we want to measure / quan/fy cerebral perfusion? Altered in various disease states Stroke - possible to iden/fy ischaemic core and penumbra of possibly salvageable /ssue: (1) CBF map from stroke survivor shows ischaemic region (measured by ASL) Hyperperfusion in tumours: (2) CBF map of glioblastoma. Hyper- perfusion within periphery (measured by ASL) (1) Brumm et al, Neuroimage, (2010) 51(3): (2) Deibler et al, AJNR, (2008) 29:

5 Introduc/on Func/onal MRI BOLD signal is a combina/on of changes in CBF, CBV and CMRO 2 BOLD signal % - not a quan/ta/ve physiological parameter Direct measures of CBF and CBV may be more closely localised at the site of neuronal ac/vity Combined BOLD & ASL CMRO 2 (and more )

6 MRI perfusion techniques DSC- MRI Capture first pass of intravenously injected bolus of paramagne/c contrast agent (Gadolinium) Contrast is either relaxivity (T1) / suscep/bility (T2 and T2*) based Østergaard, JMRI, (2005) 22:

7 MRI perfusion techniques Note the units of the perfusion parameters! Drawbacks: Requires injec/on of exogenous contrast agent Quan/fica/on requires accurate measurement of arterial input func/on Not suitable for longitudinal experiments / mul/ple measurements Not suitable in certain disease states Non- invasive quan/ta/ve MRI method?? ASL (1) Ohtonari et al, Neurol Med Chir, (2008) 48:

8 ASL - Basics

9 Arterial Spin Labelling (ASL) Acquire tag image inflowing arterial blood water labelled by magne/c inversion: Time delay between 1 and 2: Labelled water reaches capillary bed and is exchanged with water molecules in the /ssue signal change Acquire control image inflowing blood is not labelled: Difference between control and tag image is perfusion- weighted:

10 Arterial Spin Labelling: Tag Imaging slice Inflowing arterial blood water

11 Arterial Spin Labelling: Tag Imaging slice Inflowing arterial blood water Inversion

12 Arterial Spin Labelling: Tag CBF (inverted spins) arteries veins capillaries

13 Arterial Spin Labelling: Tag CBF arteries veins capillaries

14 Arterial Spin Labelling: Tag CBF arteries veins capillaries

15 Arterial Spin Labelling: Control Imaging slice Inflowing arterial blood water

16 Arterial Spin Labelling: Control Imaging slice Inflowing arterial blood water Inversion

17 Arterial Spin Labelling: Control CBF arteries veins capillaries

18 Arterial Spin Labelling: Control CBF arteries veins capillaries

19 Arterial Spin Labelling: Control CBF arteries veins capillaries

20 Arrival of labelled blood Flow Arterial Arrival Time

21 Whole- brain CBF maps CBF (ml / 100 g /ssue / min)

22 Whole- brain AAT maps AAT (seconds)

23 ASL - Techniques (Con/nuous ASL, Pulsed ASL, Velocity- selec/ve ASL)

24 Con/nuous Arterial Spin Labelling (CASL) First ASL technique (2) Control Label 3 4 sec RF pulse with magne/c field gradient in direc/on of arterial flow: Δr M control M label M Label: inflowing spins inverted by principle of flow- driven adiaba/c inversion Control: f l - f l (or G - G): inflowing spins are unaffected Adapted from (1) Host sequence (e.g. EPI, GRASE) played arer /me delay ( post labelling delay ) ΔM=M c - M l perfusion weighted image (1) Petersen et al, Br J Radiol, (2006) 79: (2) Williams et al, Proc Natl Acad Sci, (1992) 89:

25 Adiaba/c inversion RF pulses Before understanding how flowing spins are inverted in CASL, we need to look briefly at adiaba/c inversion pulses Amplitude and frequency of adiaba/c inversion pulses depends on /me: A(t) and ω rf (t) The net magne/za/on vector, M, precesses around B eff in the rota/ng frame Principle of adiaba/c fast passage M follows B eff provided direc/on of B eff does not vary too much during one precession of M about B eff By modula/ng A(t) and ω rf (t), B eff can be swept from +z to z and if adiaba/c condi/on is sa/sfied, M will follow and be inverted (1) M.A. Bernstein. Handbook of MRI Pulse Sequences. Oxford: Academic Press, 2004

26 Adiaba/c inversion RF pulses (Used to invert large volume of spins in PASL)

27 Flow- driven Adiaba/c Inversion Back to CASL apply fixed RF (no amplitude or frequency modula/on) for 2-4 secs: We know this pulse cannot adiaba/cally invert sta/onary (/ssue) spins (Why?) Recall that a gradient G=G z is applied along with the RF: No /me- dependency on either A or ω G z z For flowing spins, B 0 is swept due to G z Spins flowing with constant velocity along z experience a linearly varying magne/c field and a resultant sweep of the resonant frequency (1) M.A. Bernstein. Handbook of MRI Pulse Sequences. Oxford: Academic Press, 2004

28 Flow- driven Adiaba/c Inversion v r >> r 0 r > r 0 Spins moving towards r 0 from r(t)<<r 0 experience a /me- varying frequency offset: Labeling plane r = r 0 r < r 0 And an effec/ve magne/c field given by: r << r 0 If velocity of flowing spins is within the correct range, this frequency offset acts on flowing spins like an adiaba/c pulse as B eff is swept by changing Δω: (1) M.A. Bernstein. Handbook of MRI Pulse Sequences. Oxford: Academic Press, 2004

29 Flow- driven Adiaba/c Inversion v r >> r 0 (1) z r > r 0 Labeling plane r < r 0 r = r 0 y r << r 0 x Spins far from labeling plane Δω is large (propor/onal to r(t) r 0 )

30 Flow- driven Adiaba/c Inversion (2) z y x Spins approaching labeling plane Δω decreasing (propor/onal to r(t) r 0 )

31 Flow- driven Adiaba/c Inversion (3) z y x Spins at the labeling plane Δω goes to zero (propor/onal to r(t) r 0 )

32 Flow- driven Adiaba/c Inversion (4) z y x Spins moving away from labeling plane Δω starts to increase but is now posi/ve (propor/onal to r(t) r 0 )

33 Flow- driven Adiaba/c Inversion (5) z y x Spins flowing into brain Δω is large (propor/onal to r(t) r 0 ) and flowing spins are inverted!

34 Magne/za/on Transfer Effects CASL apply long off- resonance RF pulse Schema/c diagram of absorp/on spectra of free water and macromolecule protons (1) : ASL labelling pulse At the imaging plane Free water protons at the imaging plane are largely unaffected (due to narrow RF absorp/on spectrum) Protons bound to macromolecules at imaging plane are inverted by the off- resonance labelling pulse (due to broad spectrum of immobile protons) Chemical exchange between the two proton pools magne/za/on transfer Used as a posi/ve contrast mechanism (MTI magne/za/on transfer imaging) but problema/c in ASL.why? (1) Grossman et al, Radiographics, (1994) 14:

35 Magne/za/on Transfer Effects Water in sta/c /ssue appears to be inverted due to magne/za/on transfer This adds to the blood water signal that we want to measure with ASL but is not true perfusion signal Overes/mate ASL difference signal

36 Magne/za/on Transfer Effects Label off- resonance pulse (f L ) To compensate for MT effect in ASL control image acquired with frequency offset f c = - f L Control off- resonance pulse (- f L ΔM=M ) c - M l MT contribu/on should be the same in both the tag and control image removed by subtrac/on BUT Assumes MT effect is symmetrical about ω 0 ΔM/M 0 is the frac/onal signal difference between an image acquired with frequency offset +δω 2 and one acquired with with frequency offset - δω 2 MT effect is not symmetrical this label / control scheme (+/- f l ) s/ll leads to an error in subsequent perfusion es/ma/on (1) Pekar et al, MRM, (1996) 35:70-79

37 Magne/za/on Transfer Effects Best solu/ons to MT problem - use dedicated labelling coil (1) Small sensi/ve region of coil no satura/on of macromolecules at the imaging region Can be used to easily selec/vely label main feeding arteries (2) : (1) Zhang et al, MRM, (1995) 33: (2) Werner et al, MRM, (2004) 52:

38 Pseudo- con/nuous ASL CASL drawbacks: Magne/za/on transfer (as we have just described) Inversion efficiency (<< 100%) Requires applica/on of long RF pulse ( con/nuous mode opera/on ) Pseudo- conjnuous ASL (pcasl) (a) Label: B 1ave 0, G ave 0 (b) Control: B 1ave =0, G ave =0 Magne/za/on transfer is matched between tag and control High inversion efficiency (typically 90% increased SNR) Can be implemented on clinical systems that do not support con/nuous RF (1) Dai et al, MRM, (2008) 60:

39 Pseudo- con/nuous ASL Sweeping the phase shir accumulated between RF pulses results in an inversion of flowing spins Control (doked line): phase shir for all posi/ons (alternate sign of RF pulse + zero net gradient) Label: Imbalance in gradients Posi/on dependent phase shir Changing gradient imbalance sweep ϕ from to causes inversion of flowing spins (1) Garcia et al, Proc ISMRM, (2005) 13:37 (2) Dai et al, MRM, (2008) 60:

40 Pseudo- con/nuous ASL Mul/slice perfusion weighted images from pcasl experiment

41 Vessel- encoded pcasl pcasl labelling scheme (RF and G z ) Add transverse gradient blips (G xy ) can label selec/vely (e.g. along line A) RF pulse phase cycled such that spins in the tag vessel (e.g. A) are inverted while spins along control vessel (e.g. B) are unaffected Different tag / ctrl vessel combina/ons are possible Perfusion territory image Vessel- encoded dynamic angio (1) (1) Okell et al, MRM,(2010) 64:

42 Pulsed Arterial Spin Labelling ConJnuous and pseudo- conjnuous ASL (CASL and pcasl): Label for a given /me at a specific loca/on (i.e. in the neck) Pulsed ASL (PASL): Label a large volume of blood with short RF pulse Typically use adiaba/c RF pulse (described earlier) Wait for labeled volume (or bolus) to flow into capillary bed and exchange with /ssue Two families of PASL sequence: Echo planar and signal targe/ng with alterna/ng RF (EPISTAR) Flow sensi/ve alterna/ng inversion recovery (FAIR) Quan/ta/ve imaging of perfusion using a single subtrac/on (QUIPSS)

43 EPISTAR (1) Saturate spins at the imaging loca/on f l Dephase transverse magne/za/on prior to labeling f c =- f l Spa/ally selec/ve adiaba/c inversion pulse ΔM EPISTAR =M C - M L Doesn t account for MT asymmetry problem PICORE drop the slab selec/on gradient in the control phase (problem with this?) Other EPISTAR type sequences: STAR, STAR- HASTE, TILT. (1) Edelmann et al, Radiology,(1994) 192:

44 FAIR (1) Slice selec/ve gradient: pulse only inverts spins at the imaging loca/on No slice selec/ve gradient: pulse inverts spins in the en/re sensi/ve region ΔM FAIR =M L M C No off resonance RF not sensi/ve to MT effects Other FAIR- type sequences: UNFAIR, FAIRER (1) Kwong et al, Radiology,(1994) 192:

45 Quan/fica/on? In order to quan/fy, generally vary post labelling delay and image mul/ple /me- points (usually called inversion /me (TI)= labelling dura/on + post labelling delay) Track /me- course of flowing spins at the imaging loca/on (more on this in quan/fica/on sec/on ) Voxel with arrival Time of ~200ms Voxel with arrival Time of ~500ms Transit /me sensi/vity: ΔM depends on what TI is used (in a single TI experiment) CBF quan/fica/on from single TI? Are rela/ve perfusion values between regions valid?? (1) Petersen et al, Br J Radiol,(2006) 79:

46 QUIPSS Reduces the transit /me sensi/vity of single- TI PASL techniques (1) Inplane satura/on applied arer TI 1 removes contribu/on to the final difference image of spins that arrived before TI 1 Only labelled blood that arrives at the imaging loca/on between TI 1 and TI 2 contributes to the perfusion weighted signal Reduces transit /me sensi/vity more confidence in quan/fica/on QUIPSS2 apply satura/on pulse at the labeling plane instead (1) Wong et al, MRM,(1998) 39:

47 Velocity Selec/ve ASL (1) CASL and PASL label spins based on their loca/on (spa/al selec/vity) VS- ASL modulates longitudinal magne/za/on of flowing spins based on their velocity Uses a velocity selec/ve scheme Gradients induce a velocity- dependent phase shir between each pair or RF pulse train elements: These parameters define a cut- off velocity, V c Only spins with V > V c are labeled Image acquisi/on only includes spins with V < V c Only decelera/ng spins are included arteriole / capillary specific (excludes venous blood) (1) Wong et al, MRM,(2006) 55:

48 ASL - Quan/fica/on

49 ASL Quan/fica/on Mul/- TI ASL experiment ΔM(t) (ASL difference signal is measured as a func/on of /me) pcasl data: Labelling dura/on=1.4s, 6 x PLDs Mathema/cal model for ΔM(t), containing flow term, T1, describing exchange of labelled water between blood and /ssue, etc. QuanJfy CBF Fit model to data Quan/fy physiological parameters (i.e. CBF, AAT)

50 ASL Quan/fica/on Typically use the Buxton (1) general kine/c ASL model ΔM(t) constructed as a sum over /me of the delivery of magne/za/on to the /ssue ΔM(t) = difference signal measured by ASL α = inversion efficiency (between 0 and 1) M 0b = equilibrium magne/za/on of blood f = cerebral blood flow (CBF) Integeral = convolu/on of delivery func/on, c(t), with product of residue func/on, r(t), and magne/za/on relaxa/on func/on, m(t) Fit to the solu/on to this equa/on for the type of ASL data you have acquired (i.e. CASL, PCASL or PASL) Get voxel- by- voxel map of CBF (and possibly arterial transit /me (ATT)) (1) Buxton et al, MRM,(1998) 40:

51 ASL Quan/fica/on pcasl data: Labelling dura/on=1.4s, 6 x PLDs Perform voxel- wise fit of CASL solu/on to general kine/c model to data (need to assume / measure values for T1, α, M 0b,λ)

52 ASL Quan/fica/on pcasl data Model fit

53 ASL Quan/fica/on Effect of varying CBF, arterial transit /me, T 1 and labelling dura/on on the general kine/c model: (1) Buxton et al, MRM,(1998) 40:

54 M 0,blood Calibra/on Look again at the the equa/on for ΔM(t): Before CBF can be obtained in absolute units (ml/100g /ssue/min), need to measure M 0,blood and es/mate or measure α

55 M 0,blood Calibra/on Measuring M 0b : Measure equilibrium magne/za/on of CSF and use this to calculate M 0,blood Acquire iden/cal image (e.g. EPI / GRASE) to ASL acquisi/on, without labelling and with long TR Calculate mean CSF signal (i.e. within a CSF mask) M 0,csf Calculate M 0,blood from the following rela/onship: Where λ = ρ blood /ρ csf = 0.87 (ml/ml) and we also correct for T2* (for gradient echo) decay of the signal within the echo /me (1) (1) Hersovitch & Raichle, JCBFM,(1985) 5:65-69

56 Inversion Efficiency Degree of inversion (α) achieved by labelling scheme: CASL pcasl Can vary significantly as a func/on of velocity studies involving hypercapnia challenge (CO 2 )? Aslan et al Combine phase- contrast and pcasl to measure α pcasl decreased from 0.95 to 0.84 during inhala/on of 5% CO 2 Cardiac cycle, intra- subject variability in BP etc need to measure on per- subject basis? (1) Aslan et al, MRM,(2010) 63:

57 Sources of error in General Kine/c Model (1) Assumes plug flow no labelled spins arrive at the imaging region before Δt Range of transit /mes more likely can lead to underesjmate of perfusion (especially with PASL) should account for distribu/on of arrival /mes (2) Single- compartment kine/cs assumes rapid exchange of labelled spins between vessels and /ssue Some labelled spins may remain in vessels before being exchanged (some may even pass through the voxel without undergoing exchange (1) ) This assump/on also leads to an underesjmajon of perfusion (3) Labeled magne/za/on is assumed to decay with /ssue T 1 on arrival at the imaging voxel This assump/on leads to an overesjmajon of perfusion (1) Silva et al, MRM,(1997) 38:

58 Typical PCASL experiment setup (1) Run /me- of- flight angiography to visualise vessels in the neck: (2) Run ASL sequence: (3) Run calibra/on scan(s) for M0 blood calcula/on: = Head coil Body coil Sensi/vity map

59 ASL - Current applica/ons

60 Current ASL applica/ons (1) Okell et al, MRM,(2010) 64:

61 Current ASL applica/ons (1) Okell et al, MRM (2010)

62 Current ASL applica/ons Kelly et al, BRAIN (2011)

63 Current ASL applica/ons ASL + BOLD weighting (TE) + gas challenges Bulte et al, Neuroimage (2012)

64 Current ASL applica/ons Whole- brain PCASL acquisi/on, focal pain s/mulus

65 What about 7T? T1 of blood changes from ~1600ms at 3T to ~2000ms at 7T Labeled signal will decay more slowly Improved CNR and ability to acquire more slices? But B 1 homogeneity in the neck? need specific labeling coil? Specific absorp/on rate more RF power needed?

66 ASL Poten/al for phmri

67 Advantages of ASL for phmri Non- invasive (compared to PET) suitable for longitudinal studies CBF single physiological parameter quan/fied in absolute units Don t need to use s/mulus / task response to drug effects Quan/fy CBF at rest and during s/mulus: Separate effects of pharmacological agents on baseline and task- induced ac/va/on NB where task- based studies may be affected by deteriora/ng performance (clinical popula/ons)

68 Advantages of ASL for phmri Baseline CBF affects the func/onal CBF response (during visual s/mulus) but not the BOLD response (CMRO 2 / CBF coupling) Ability to detect baseline CBF with ASL (versus BOLD) is advantageous

69 Advantages of ASL for phmri Pairwise (control- tag) subtrac/on in ASL acts as a high pass filter Frequency spectrum of ASL /me series is flat ( white noise ) - as opposed to BOLD which has elevated power at low frequency ASL is more suitable for studying slow changes in brain func/on (e.g. drug effects taking hours / days) Power BOLD high power at low frequency Ff Frequency (Hz) Aguirre et al, Int Rev Neurobiol (2005)

70 Advantages of ASL for phmri BOLD EPI suffers from distor/on and dropout ASL can use alterna/ve sequences that are resistant to magne/c field inhomogeneity (suscep/bility) effects Improved visualiza/on of orbitofrontal, temporal and limbic regions (linked to neurotransmi er systems)

71 Advantages of ASL for phmri 3D GRASE readout: Aside: data demonstrates importance of acquiring mul/- TI ASL data MacIntosh et al, JCBFM (2008)

72 Advantages of ASL for phmri Drug ac/on /mes range from seconds (intravenous) and hours (oral) to days and weeks (treatment cycle) Need high test- retest reliability (precision) reliability accuracy Within- subject coefficient of varia/on ~10% for global CBF and ~15% for regional measures To detect a 15% change in CBF with 90% power and 15% varia/on between repeated measures requires ~20 subjects Wu et al, MRM (2007), Wang et al, JPET (2011)

73 Recent important technical advances Increased field strength Parallel imaging (increased SNR and temporal stability) Recent progress in commercialisa/on pcasl increased labelling efficiency and SNR Background suppression to suppress background /ssue reduces physiological noise and sensi/vity to mo/on improves sensi/vity and temporal stability whole brain coverage, 4mm isotropic in a few minutes Fernandez- Seara et al, MRM (2008)

74 Challenges for ASL CBF measured by ASL s/ll an indirect measure of regional brain func/on Vascular confounds - Some agents may directly alter CBF (or mediators or neurovascular coupling) rather than brain func/on itself e.g. Caffeine: But.no change significant change in CMRO 2 (measured by calibrated BOLD/ASL) coupling between CBF and CMRO 2 is altered Wang et al, JPET (2011) and Perthen et al, Neuroimage (2008)

75 Challenges for ASL 4 doses of remifentanil PaCO 2 increases with increased seda/on global CBF increase (global systemic CBF increase - A) Need to normalize regional CBF by global CBF to iden/fy remi- induced regional CBF varia/on (limbic CBF effect B) Ko e et al, Anesth Analg (2007)

76 Future direc/ons Con/nued improvement in hardware and sequence development Further move towards calibrated ASL or ASL/BOLD CBF, AAT, CMRO 2, OEF, CBV Perfusion- based fmri / phmri long behavioural tasks (stress, craving, pain )

77 Ques/ons?