hyperpolarization -description, overview & method · para-hydrogen induced polarization (phip)...
Post on 20-Oct-2019
4 Views
Preview:
TRANSCRIPT
Hyperpolarization - Description, Overview & MethodEducational Session, ISMRM 2017
April 26, 2017
Peder Larson, Ph.D.Associate Professor, Department of Radiology and Biomedical Imaging, University of California, San Francisco, CA, United States
peder.larson@ucsf.eduhttps://radiology.ucsf.edu/research/labs/larson@pezlarson
Speaker Name: Peder Larson
I have the following financial interest or relationship to disclose with regard to the subject matter of this presentation:
Company Name: GE HealthcareType of Relationship: Research Support
Declaration ofFinancial Interests or Relationships
Outline
https://radiology.ucsf.edu/research/labs/larson/educational-materials
(Google: Peder Larson Lab, Educational Materials link on sidebar)
§ Hyperpolarization
• What does it mean?
• Dissolution Dynamic Nuclear Polarization (dDNP)
• Spin Exchange Optical pumping
• Para-hydrogen induced polarization (PHIP, SABRE)
§ Imaging Methods
• RF pulse strategies
• Acquisition strategies
• Analysis – kinetic models
April 26, 2017Hyperpolarization Methods3
Spin Polarization
4
Spins
µ
B = 0
µ
µ
µ
µ
M0 = 0
Net Magnetization
netic field, it will be found to be either in the spin-upor spin-down state, no matter which mixed state cj iit was in before. Furthermore, it will stay in that newstate until the proton is subject to more interactionswith environment (e.g. another measurement). Thisso-called collapse into an eigenstate is a consequenceof QM. It apparently implies that a measurement ofthe net magnetization (e.g. by MRI), will force eachproton into either the spin-up or the spin-down statein agreement with myth 1. This is wrong, however.The emphasized word individual above is importantin the present context, as we can only infer from QMthat the protons are forced into single-spin eigen-states, if we measure their magnetization one-by-oneas can be done with a Stern-Gerlach apparatus, forexample (11). In contrast, that is never done in MRspectrometers or scanners: to get a measurable MR-signal the total magnetization of many nuclei isalways measured, and myth 1 does not follow. Itcould be true nevertheless, but in fact it is not, whichis shown in appendix (proposition 1) by employingthe QM formalism: A measurement of the net mag-netization causes a perturbation of the system that isinsufficient to affect the individual protons signifi-cantly. In particular, they are not brought into theireigenstates by the measurement process.
It is worth noting that even though the argumentsmentioned above may occur complicated for the non-technical reader, they are what many students of MRmore or less implicitly lay ears to, and for no goodreason, as QM is not needed for understanding basicMR. Moreover, the students often hear the wrongversion of the argument.
The lifetime of myth 1 may have been prolongedby an observation that many working with MR have
made: when subject to a magnetic field, an oblong pi-ece of magnetizable material have a strong tendencyto align itself in one of two opposite directions paral-lel to the field (in contrast to permanently magnetizedmaterial that orient itself in one direction only). De-spite a superficial resemblance, this well-known phe-nomenon has nothing to do with the effect expressedin myth 1. Rather it is a consequence of reorientationof magnetic constituents inside the metal. This givesrise to the existence of two low-energy states for theorientation of the metallic piece, parallel and antipar-allel to the field. The magnetic constituents are in ei-ther case parallel to the field, because they have onlyone low-energy state. Similarly, the proton spin hasonly one low-energy state. Nothing but MR-irrele-vant single-proton measurements give spins a tend-ency to align antiparallel to the field.
Consequently, spins can point in any direction andthe energy eigenstates are not more relevant to MRthan any other state (the eigenstates form a conven-ient basis for computations, but they are irrelevantfor the understanding). Hence Fig. 1 that illustratesthe nature of spin eigenstates, do not contribute muchbut confusion in an MR context. QM is later shownto imply that the spin-evolution of individual protonshappens as expected classically unless perturbed,e.g., by a single-spin measurement.
Finally, replacements for Fig. 1 are discussed.According to both classical and QM, spins areexpected to point in all directions in the absence offield as shown in Fig. 2. Except for precession, thesituation does not change much when the polarizingB0-field used for MR is applied as shown in Fig. 3.The energies associated with the orientation of theindividual spins are much smaller than the thermal
Figure 2 In the absence of magnetic field, the spins are pointing randomly hence giving aspherical distribution of spin orientations. This is illustrated to the right by a large number ofexample spins in an implicit magnetization coordinate space similar to that of Fig. 1. [Color fig-ure can be viewed in the online issue, which is available at www.interscience.wiley.com.]
332 HANSON
Concepts in Magnetic Resonance Part A (Bridging Education and Research) DOI 10.1002/cmr.a
Danish Research Centre for Magnetic Resonance. Educational Materials: http://www.drcmr.dk/educations/education-material
April 26, 2017Hyperpolarization Methods
Spin Polarization in a Magnetic Field
5
Spins
µB0µ
µ
µ
µ
M0 ≠ 0
Net Magnetization
energies so the spins only have a slight tendency topoint along the direction of the field (and noincreased tendency to point opposite the field, neitherclassically, nor quantum mechanically). The situationcan be compared to the one described earlier involv-ing a hypothetical collection of compasses placed inthe earth magnetic field. All compasses will swing tothe north, if they are noninteracting and not dis-turbed. The situation changes if the compasses areplaced in a running tumble-dryer or similar deviceincreasing the collisional energies above those asso-ciated with changing the direction of the compassneedles. The bouncing and interacting compasseswill no longer all point toward north but there willstill be a slight tendency for them to do so. If the netmagnetization is measured, it will point toward north.The situation is like that of the moving protons in aliquid sample where the magnetic interactionsbetween neighboring nuclei cause reorientation ofthe magnetic moments (relaxation). In the absence ofa magnetic field, the angular distribution of spins isspherical in either case. When a field is added, thedistribution is skewed slightly toward the field direc-tion by relaxation.
It is important to understand that precession of theindividual nuclei starts as soon as the sample isplaced in the field (not only after excitation by RFfields, as frequently stated). The nuclei therefore emitradio waves at the Larmor frequency as soon as theyare placed in the field. Similarly, however, theyabsorb radio waves emitted by their neighbors andsurroundings. Because the distribution of spin direc-tions is even in the transversal plane, the net trans-versal magnetization is zero, and there is no netexchange of energy between the sample and its sur-roundings. The exchange of radio waves within thesample is nothing but magnetic interactions betweenneighboring nuclei. These are responsible forrelaxation.
Myth 2: MR Is a Quantum Effect
A quantum effect is a phenomenon that cannot beadequately described by classical mechanics, i.e., onewhere only QM give predictions in accordance withobservations. In the introduction, it was stated thatatom formation is a quantum effect because atomsare not expected to be stable according to classicalmechanics, whereas experiments have proven thatthey are. This does not imply that all phenomenainvolving atoms are defined as quantum effects, sincesuch a broad definition would be quite useless.Instead, phenomena are hierarchically divided intoclassical and quantum phenomena, so a classical phe-nomenon can involve atoms that are themselves inex-plicable by classical mechanics. Similarly, protonspin is a quantum effect but MR is not since the latteris accurately described by classical mechanics. Thisis the subject of the present section.
It is often and correctly said that spin is a quantumeffect. From a classical perspective it cannot beexplained why protons apparently all rotate with thesame constant angular frequency, which result in anobservable angular momentum (spin) and associatedmagnetic moment. Despite the fact that this is reallymind-boggling, it is usually not perceived so bystudents of MR. Just as atoms are taken for granted,it is typically accepted without argument that protonsappear to be rotating and that they as a result behaveas small magnets with a north and a south pole, i.e.they have angular and magnetic moments. Mostbooks state this correctly and there is no apparentreason to elaborate, as a deeper understanding of spinis typically of little use in the context of MR.
Even though spin is a quantum effect, MR is not,according to the definition given earlier, as it doesnot necessitate a quantum explanation. Classically, amagnetic dipole M with an associated angular mo-
Figure 3 Better alternative to Fig. 1 showing the spindistribution in a magnetic field. The spins will precess asindicated by the circular arrow and a longitudinal equilib-rium magnetization (large vertical arrow) will graduallyform as the distribution is skewed slightly toward mag-netic north by T1-relaxation (uneven density of arrows).The equilibrium magnetization is stationary, so eventhough the individual spins are precessing, there is no netemission of radio waves in equilibrium. [Color figure canbe viewed in the online issue, which is available atwww.interscience.wiley.com.]
UNDERSTANDING MAGNETIC RESONANCE 333
Concepts in Magnetic Resonance Part A (Bridging Education and Research) DOI 10.1002/cmr.a
Polarization fraction:
0.0001-0.0005% at room temperature, depending on nucleus (γ) and field (B0)
tanh −hγB02πkT
"
#$
%
&'
April 26, 2017Hyperpolarization Methods
Hyperpolarization
§ Perturb spins from thermal equilibrium to increase fraction aligned parallel (or anti-parallel) to B0
§ Polarizations of > 50%!!§ Methods:
• Optical pumping (for gasses, ie 3He, 129Xe)
• Parahydrogen-induced Polarization (PHIP)
• Dynamic Nuclear Polarization (DNP)
6
Hyperpolarized MRS
Normal MRS
B0
Net Magnetization
April 26, 2017Hyperpolarization Methods
Spin Exchange Optical Pumping
§ Polarization of noble gases (e.g. 3He, 129Xe)
§ Mixtures of alkali-metal vapors and noble gases irradiated with circularly polarized resonant light
§ Major application is pulmonary imaging for lung disease
April 26, 2017Hyperpolarization Methods7
Hyperpolarized 129Xe MRI Walker et al. Rev Mod Phys 1997, doi:10.1103/RevModPhys.69.629Roos et al. Magn Reson Imaging Clin N Am. 2015, doi: 10.1016/j.mric.2015.01.003
Para-hydrogen Induced Polarization (PHIP)
§ Parahydrogen is the Singlet State of Hydrogen gas, H2
§ MR invisible, can store magnetization
§ Transfer the polarization from the singlet-state to other nuclei
April 26, 2017Hyperpolarization Methods8
Triplets, T+, T0, T-
Singlet, S0
para hydrogen
1H
1Hparahydrogen
1H
Ir
hyperpolarizedsubstrate 1H
reversibleexchange
reversibleexchange
1H1H
J
J SABREAdams, … , Duckett et al.Science 2009 323, 1708
Content Courtesy Thomas Theis
Signal Amplification By Reversible Exchange (SABRE)
§ At the appropriate magnetic field (e.g. 6.5 mT for hyperpolarization of protons), J-coupling interactions across the iridium catalyst drive hyperpolarization from parahydrogen to substrate.
§ Reversible exchange of both, parahydrogen and substrate, leads to continuous hyperpolarization buildup.
§ Heteronuclear SABRE greatly generalizes the substrate scope and enables long hyperpolarization lifetimes.
April 26, 2017Hyperpolarization Methods9
R1H
1H 15N
Ir
15N
parahydrogen
hyperpolarizedsubstrate
reversibleexchange
reversibleexchange
1H1H
15N
R
R
Theis et al. JACS 2015 137 (4), 1404
Content Courtesy Thomas Theis
SABRE DiscussionPrimarily limited to molecules containg sp or sp2 hybridized nitrogens§ Pros
• Relatively Simple and Fast• Uses inexpensive equipment• Long-Lived States can be
hyperpolarized directly§ Cons
• Limitations on polarizable substrates
• Reduced efficiency in water• Removal of the toxic
hyperpolarization catalyst
Colell et al. JPCC 2017 121(12), 6626
Substrates:Primarily limited to molecules containg sp or sp2
hybridized nitrogens
Content Courtesy Thomas Theis
April 26, 2017Hyperpolarization Methods10
Dynamic Nuclear Polarization (DNP)
§ Microwaves at appropriate frequency transfer polarization from electrons to nuclei
§ High magnetic field increases polarization of both nuclear and electron spins
§ Very low temperature also used to increase polarization of both nuclear and electron spins
11
Temperature (K)
Pola
rizat
ion proton
carbon
10-30
0.5
1 electron
10-2 10-1 10-0 101 102
At 1.2K, Pe= 94% & PC= 0.086%
For B0 = 3.35 TSample: Amorphous solid material doped with unpaired electrons at a ratio ~ 1 free electron:1000 13C
April 26, 2017Hyperpolarization Methods
3.35 T and ~1.2oKγelectron B0 = 94 GHzγC-13 B0 = 35 MHz
HyperSense Dissolution DNP Polarizer
12
Microwave Irradiation atγelectron B0 ±γC-13 B0
April 26, 2017Hyperpolarization Methods
• The buffer is heated and pressurized
• The sample space is pressurized
• The sample is raised out of the liquid helium
• The dissolution stick is lowered, docking with the sample holder
• The solvent is injected, dissolving the sample, while preserving the enhanced polarization
Slide courtesy of Jan Henrik Ardenkjaer-Larsen, GE Healthcare
Dissolution DNP Procedure
13 April 26, 2017Hyperpolarization Methods
SpinLab Clinical Polarizer
12 3 4
SpinLabPolarizer
Ardenkjaer-Larsen et al. NMR Biomed 2011; 24:927
5 T and ~0.8oKγelectron B0 = 140 GHzγC-13 B0 = 52 MHz
Automated Quality Control System
April 26, 2017Hyperpolarization Methods14
Dissolution DNP 13C AgentsRequirements
• Long T1 relaxation time (polarization half-life)
• Water-soluble is best
• Mixture with free electron source, aka electron paramagentic agent (EPA), free radical
• Low Toxicity
• In vivo interest
Agents
• [1-13C]-pyruvate: metabolism, Warburg effect
• 13C-urea: inert, perfusion
• bis-1,1-(hydroxymethyl)-[1-13C]cyclopropane-d8 (HMCP, HP001): long T1 perfusion agent
• [1,4-13C2]-fumarate: necrosis
• 13C-bicarbonate: pH measurement
• [2-13C]-fructose: metabolism
• [5-13C]-glutamine: metabolism, cell proliferation
• 13C-dehydroascorbate (DHA): Reduction/oxidation potential
• [1-13C]-α-ketoglutarate: IDH mutation status
• [U-2H, U-13C]-glucose: metabolism
• and more
(Review article : Keshari, Wilson. Chem Soc Rev, 2014.)
Wilson,etal.JMR2010.
Simultaneouspolarizationofmultipleagents
April 26, 2017Hyperpolarization Methods15
Hyperpolarized Carbon-13 Pyruvate
13C-Pyruvate § Most promising HP agent thus far§ Long T1 (≈ 40-60 s)§ Readily polarizable§ Endogenous§ Rapid uptake and conversion to
lactate, alanine, and bicarbonate§ Directly probes the “Warburg Effect”
in cancer§ (Safety and feasibility established in
prostate cancer patients)
17
Warburg Effect
MGVander Heiden etal.Science2009;324:1029-1033
GlycolysisApril 26, 2017Hyperpolarization Methods
Bloch Equation
18
ddtM =
M ×γ
B+
−1/T2 0 00 −1/T2 00 0 −1/T1
#
$
%%%%
&
'
((((
M +
00
M0 /T1
#
$
%%%
&
'
(((
Net magnetization behavior, hyperpolarized or at thermal equilibrium, is described by Bloch equation:
PrecessionRF Excitation
Relaxation back to thermal equilibrium:
M =
00M0
!
"
###
$
%
&&&
April 26, 2017Hyperpolarization Methods
Relaxation to Equilibrium
19
T1 decay of Mz (~50 s in vivo for [1-13C]pyruvate)
Hyperpolarization
M0
10000 M0
M0
April 26, 2017Hyperpolarization Methods
Relaxation to Equilibrium
20
T2 decay of Mxy(~100ms-2s in vivo for pyruvate)
RF excitation
xy
z
April 26, 2017Hyperpolarization Methods
13C MR of Pyruvate Metabolism
21
Dynamic MR Spectrum in vivo
frequency
Pyruvate
13C
Lactate
13C
Bicarbonate
13C
Pyruvate-hydrate
Following injection of 13C-pyruvate
Alanine
13C
April 26, 2017Hyperpolarization Methods
Hyperpolarized 13C Imaging Procedure
1. Hyperpolarization of 13C-pyruvate (45-90 mins)
2. Rapidly dissolution of frozen compound to create a hyperpolarized liquid agent (10 sec)3. Agent is injected to the subject inside the MRI scanner (10 sec)4. 13C MRI/MRSI is performed immediately (1-2 mins)
22
1,2 3
4
April 26, 2017Hyperpolarization Methods
MR Pulse Sequence Components
1. Excite spins (RF)
2. Readout signal (spectral and/or spatial encoding)
3. Repeat
23
Magnetization Vector
MZ
MXYSignal
RF
DAQ
T2
T1
1.
1.
2.
2.
x NTR
3.
T2
T1
April 26, 2017Hyperpolarization Methods
Hyperpolarized RF Pulses
Two key considerations:
1. Efficient use of hyperpolarization
• Variable flip angles
• “Multiband” excitation
2. Spectral selectivity
• Spectral-spatial RF pulses
April 26, 201724 Hyperpolarization Methods
Constant Flip Angle
§ Received signal varies between excitations (can cause blurring)§ Residual unused hyperpolarization after last excitation
April 26, 201725
S1 S2 S3 S4 S5 S6
θθ
θθ θ θ
MZ
MXY
Hyperpolarization Methods
Variable/Progressive Flip Angle
§ Flip angle is strictly increasing§ Received signal constant when accounting for lost magnetization (Si = C)§ Efficient usage of all polarization
April 26, 201726 Zhao, et, JMR B 113 (2) (1996) 179–183.Nagashima, JMR 190 (2) (2008) 183–188.
θ1
S1
θ2 θ3
S2 S3
θ4
S4
θ5= 45°
S5
θ6 = 90°
S6
MZ
MXY
Hyperpolarization Methods
Dynamic Imaging: Conventional Excitation
April 26, 201727
Lactate
13C1
Pyruvate
13C1
flipangle
2.5°
5°
10°
20°
time
received
signal
ExcessPyruvateSNR
Hyperpolarization Methods
Dynamic Imaging: Multiband Excitation
April 26, 201728
flipangle
2.5°
5°
10°
20°
time
received
signal
MorelactateSNRformoretime
PyruvateSNRisstillsufficient
Smallerpyruvateflipleavesmoremagnetizationthatcanthenbecomelactate
Lactate
13C1
Pyruvate
13C1
Hyperpolarization Methods
Multiband Variable Flip Angle Excitation
Combination of§ Variable flip angle across acquisitions for improved SNR§ Multiband pulse designs to account for metabolic conversion between acquisitions
April 26, 201729
MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON MEDICAL IMAGING 5
acquisition number0 5 10 15 20 25 30
flip
angl
e (d
egre
es)
0
20
40
60
80
100Optimized flip angle sequence
pyruvatelactate
Fig. 3. Optimized flip angle sequence for estimating the metabolic rateparameter kPL using the nominal parameter values in Table I and a samplinginterval TR = 2 s between acquisitions.
time (s)0 10 20 30 40 50 60
x t (au)
�104
0
0.5
1
1.5
2
2.5Simulated state trajectories
pyruvate (x1t)lactate (x2t)
(a) states xt
time (s)0 10 20 30 40 50 60
Yt (a
u)0
1000
2000
3000
4000
5000
6000
7000Simulated measurement trajectories
pyruvate (Y1t)lactate (Y2t)
(b) Rician-distributed measurementsYt
Fig. 4. Simulated trajectories of the model (2) using the optimized flip anglesequence shown in Fig. 3 and the arterial input function shown in Fig. 2a.
We see that the pyruvate flip angles follow a pattern similarto flip angle sequences designed for other objectives, beginningwith small flip angles to preserve magnetization for futureacquisitions but increasing toward the end of the sequence [9],[10]. In contrast, the optimized flip angle sequence is muchmore aggressive with the lactate flip angles at the beginningof the experiment than in other variable flip angle sequences.This provides more reliable information about the leading endof the lactate time series, which contains the most informationabout the metabolic rate.
For the particular model and regularization parameter valuesused, the BFGS optimization algorithm converges to thesame optimal sequence for a wide range of initializations. Toconfirm this, we have initialized the search algorithm usingthree flip angle sequences with angles generated randomlybetween 0 and 90�. For all three initializations, the algorithmconverges to the flip angle sequence shown in Fig. 3. Thisdemonstrates that the flip angle sequence presented is likely aglobal optimum.
IV. VALIDATION USING SIMULATED DATA
In this section, we demonstrate the advantage of the opti-mally designed flip angle sequence using computer-simulateddata. Working with simulated data allows us to collect a largenumber of statistically independent data sets and provides usaccess to a “ground truth” value for the parameter vector.This makes it possible to reliably determine the parameter
estimation error that results from noise in the simulatedmeasurements. It is not feasible to acquire such a large numberof data sets in vivo, and these would also not include groundtruth values. Thus we use simulated data to demonstrate thatour optimized flip angle sequence leads to smaller uncertaintyin estimates of the metabolic rate parameter k
PL
.
A. Two-step parameter estimation procedureWhen fitting the data from in vivo experiments, data from
different voxels will correspond to different values of theparameters k
TRANS
, k
PL
, R1P and R1L as these valueschange with spatial location, but all correspond to the samearterial input u(t). Thus we present a fitting procedure thatproceeds in two steps: first we fit a single input function u(t) tothe entire data set, then we fix this input function and estimatevalues of the remaining parameters individually for each of thevoxels in the slice.
B. Simulation results and discussionWe wish to compare the reliability of estimates of k
PL
between data generated using five competing flip angle se-quences:
1) a T1-effective sequence [9] that aims to keep the mea-sured signal constant despite repeated RF excitation andmagnetization exchange between chemical compounds(Fig. 5a)
2) an RF compensated flip angle sequence [29] that aimsto keep the measured signal constant despite repeatedRF excitation (Fig. 5b),
3) a constant flip angle sequence of 15�,4) a sequence that maximizes the total signal-to-noise ratio
in the observed signal
SNR
total
=
NX
t=0
2X
k=1
x̃
k,t
�
k
.
[10] (Fig. 5c), and5) our flip angle sequence that maximizes the Fisher infor-
mation about kPL
(Fig. 3).For each of the five flip angle sequences, we simulate n = 25
independent data sets from the model (2) using the parametervalues given in Table I. We then perform the two-step pa-rameter estimation procedure described in Section IV-A. Theresulting parameter estimates are shown in Fig. 6. We see thatfor all five flip angle sequences, the parameter estimates con-gregate near the ground truth value of the model parameters.
To demonstrate that our optimized flip angle sequenceprovides more accurate estimates of k
PL
than the competingflip angle sequences, we compare the root mean squared(RMS) estimation error between the sequences. We repeatthis experiment for various values of the noise parameter �
2
ranging from 10
3 to 10
6 to demonstrate that the improvementin the estimates is robust to variation in the noise strength. Avalue of approximately 2 ⇥ 10
4, in the center of this range,is typical for prostate tumor mouse model experiments. Foreach value of �2 we compute the RMS error of the k
PL
andnuisance parameter estimates across the n = 25 trajectories
Xing, Larson, et al, JMR 2014 John Maidens, et al, IEEE-TMI 2016, doi: 10.1109/TMI.2016.2574240
Flat Signal Response Optimal kinetic measurement
Hyperpolarization Methods
Spectral Selectivity
April 26, 201730
frequency
Pyruvate
13C
Lactate
13C
Bicarbonate
13C
Alanine
13C1
UsespectrallyselectiveRFpulsestocontrolflipanglesfordifferentcompounds
Hyperpolarization Methods
Spectral Excitation Profile
§ Design RF pulses for desired spectral profile
§ Fourier Transform relationship between RF pulse shape and Magnetization profile: Valid for small tip angles, < 30° (pretty close up to 60°)
§ Non-linear relationship for large tip angles: Use Shinnar-Le Roux transform or other tools for RF pulse design
April 26, 201731
frequency
|MXY|
Fourier or Shinnar-Le
Roux Transform
�f
[1-13C]pyruvate[1-13C]lactate
Hyperpolarization Methods
Spectral-Spatial RF Pulses
April 26, 201732
§ Add oscillating gradient for additional spatial selectivity§ Additional constraints on spectral and spatial selectivity§ Useful in vivo where spatial selectivity is important
Hyperpolarization Methods
Spectral-Spatial RF Pulse Design
Features:• Simultaneous selectivity in frequency and position• Customizable for different nuclei and MR hardware• Multi-band specification• Aliased bands are allowed• Power minimization• Time minimization• Correction for non-uniform sampling• VERSE for ramp sampling and peak B1 reduction
MATLAB code available at: https://github.com/agentmess/Spectral-Spatial-RF-Pulse-Design
April 26, 201733 Kerr et al, ISMRM 2008 #226. Larson et al, JMR 2008. Hong Shang, et al. JMR 2015.
Find Minimum-time Gradient Sublobe
Identify Consistent Spectral Sampling Frequencies Fs
Find Minimum-Time Spectral Filter
Correct for Nonuniform Sampling
2D SLR Transform
Optimal VERSE
Choose Best Design
Multiband Specification
For each Fs
Next Fs
Hyperpolarization Methods
Hyperpolarized Acquisitions: Need for Speed
§ Rapid signal decay
§ Rapid metabolic conversion
§ Single image SNR decreases with more excitations due to T1 decay
§ Fast data acquisition is important for HP agents
April 26, 201734
Simulated Single-Image SNR with T1 decay and progressive flip angle
Hyperpolarization Methods
Readout Strategies
Can be approximately grouped into three categories (from slowest to fastest)
1. MR spectroscopic imaging (MRSI)
2. Model-based Spectral decomposition with multiple TEs (Dixon/IDEAL)
3. MRI with spectrally-selective excitation (“Metabolite-specific Imaging”)
April 26, 201735 Hyperpolarization Methods
Fast MR Spectroscopic Imaging
§ Methods• Echo-planar spectroscopic
imaging (EPSI)• Spiral spectroscopic imaging• Concentric Rings• Radial EPSI• Rosettes• And More
April 26, 201736
Ky (
cm
-1)
Ky (
cm
-1)
Ky (
cm
-1)
time (ms)
time (ms)time (ms)
Kx (cm-1)
Kx (cm-1)Kx (cm-1)
0 20 40 60 80 100 120−1-0.50
0.5
-1
-0.5
0
0.5
1
0 20 40 60 80 100 120-1
-0.50
0.5-1
-0.5
0
0.5
1
0 20 40 60 80 100 120-1
-0.50
0.5
-1
-0.5
0
1
0.5
-1
0
1
0 20 40 60 80 100 120
0 20 40 60 80 100 120-1
0
1
Symmetric EPSI
Flyback EPSI
Readout
time (ms)
Kx (
cm
-1)
Kx (
cm
-1)
time (ms)
EPSI
Spiral Concentric Rings
Simultaneous acquisition of spectral and spatial k-space data in (kx, ky, kz, t) or (kx, ky, kz, kf) space
Hyperpolarization Methods
MRSI Readouts
April 26, 201737
GZ
DAQ
GZ
kz
kf
kz
kf
DAQ
Phase Encoding
Echo-planar spectroscopic imaging (EPSI)
Hyperpolarization Methods
Accelerated MRSI Strategies
April 26, 201738
Ky (
cm
-1)
Ky (
cm
-1)
Ky (
cm
-1)
time (ms)
time (ms)time (ms)
Kx (cm-1)
Kx (cm-1)Kx (cm-1)
0 20 40 60 80 100 120−1-0.50
0.5
-1
-0.5
0
0.5
1
0 20 40 60 80 100 120-1
-0.50
0.5-1
-0.5
0
0.5
1
0 20 40 60 80 100 120-1
-0.50
0.5
-1
-0.5
0
1
0.5
-1
0
1
0 20 40 60 80 100 120
0 20 40 60 80 100 120-1
0
1
Symmetric EPSI
Flyback EPSI
Readout
time (ms)
Kx (
cm
-1)
Kx (
cm
-1)
time (ms)
EPSI
Spiral Concentric RingsFuruyama J, et al. MRM 2012; 67:1515–1522Jiang, et al, MRM 2016, doi: 10.1002/mrm.25577
Adalsteinsson E, et al. MRM 1998;39:889 – 898.Mayer MRM 2006; 56:932–937.
Yen et al. MRM 2009; 62:1-10.
Ky (
cm
-1)
Ky (
cm
-1)
Ky (
cm
-1)
time (ms)
time (ms)time (ms)
Kx (cm-1)
Kx (cm-1)Kx (cm-1)
0 20 40 60 80 100 120−1-0.50
0.5
-1
-0.5
0
0.5
1
0 20 40 60 80 100 120-1
-0.50
0.5-1
-0.5
0
0.5
1
0 20 40 60 80 100 120-1
-0.50
0.5
-1
-0.5
0
1
0.5
-1
0
1
0 20 40 60 80 100 120
0 20 40 60 80 100 120-1
0
1
Symmetric EPSI
Flyback EPSI
Readout
time (ms)
Kx (
cm
-1)
Kx (
cm
-1)
time (ms)
EPSI
Spiral Concentric Rings
Hyperpolarization Methods
Comparison of Accelerated MRSI StrategiesTradeoffs
§ Speed
§ SNR efficiency
§ Robustness to hardware imperfections
§ Bandwidth
§ Resolution
April 26, 2017390.4 0.5 0.6 0.7 0.8 0.9 1
0
500
1000
1500
2000
2500
3000Spectral Bandwidth
Resolution (cm)
SB
W (
Hz)
Concentric Rings
Symmetric EPSI
Flyback EPSI
Spiral
0.4 0.5 0.6 0.7 0.8 0.9 10
1
2
3
4
5
6
7
8
9Acquisition Time
Resolution (cm)
Acq
uis
ito
n T
ime
(s)
Concentric Rings
Symmetric EPSI
Flyback EPSI
Spiral
0.4 0.5 0.6 0.7 0.8 0.9 10
500
1000
1500
2000
2500
3000Spectral Bandwidth with Interleaves
Resolution (cm)
SB
W (
Hz)
Concentric Rings
Symmetric EPSI
Flyback EPSI
Spiral
0.4 0.5 0.6 0.7 0.8 0.9 10.65
0.7
0.75
0.8
0.85
0.9
0.95
1SNR Efficiency
Resolution (cm)
SN
R E
ffic
ien
cy
Concentric Rings
Symmetric EPSI
Flyback EPSI
Spiral
FlybackEPSI
Symmetric EPSI
Concentric Rings
Spiral
Speed - - + ++
SNR - ++ + ++
Robustness ++ - ++ --
Jiang, et al, MRM 2016, doi: 10.1002/mrm.25577Hyperpolarization Methods
Spectral Decomposition§ Dixon/IDEAL (Iterative decomposition of water and fat with echo asymmetry and least-squares
estimation) methods: Originally developed for fat/water imaging
§ Reconstruct individual metabolite images based on known chemical shifts§ Multiple TEs: minimum # of TEs = Npeaks(+1 if B0 field map estimated required)
April 26, 201740
TE2
x
y
TE1x
y
Fat
Water
x
y
Fat
x
y
Water TE1x
y
TE2x
y
TE3x
y
TE4x
y
Hyperpolarization Methods
Spectral Images
Δf = 0 Hz Δf = 614 Hz
Δf = 433 Hz Δf = 272 Hz
Spectral Decomposition: Spiral CSI
April 26, 201741
Brodsky et al. MRM 59(5): 2008 Wiesinger et al., MRM 68(1): 2012
RF Spiral Readouts
TE1
TE2
TE3
TE4
Matrix Inversion FFT
SpectralK-space
Hyperpolarization Methods
Spectral Decomposition: Oversampled Spirals
April 26, 201742
ΔTE
ΔTECritically
Sampled Spiral (η = 1)
2 echoes
Twice Oversampled Spiral (η = 2)
1 echo
Gordon et al., MRM 2013
Hyperpolarization Methods
Single-shot Spiral MRI
Metabolite-specific Imaging§ Idea: Excite only a single metabolite resonance,
followed by any imaging-based readout§ Methods:
• Single-metabolite Spectral-spatial excitation
• Fast imaging readout (EPI, spiral)
§ Fast!
§ Requires chemical-shift separation of metabolites, sensitive to B0 inhomogeneities
April 26, 201743
Spectral-spatial Excitation
Cunningham JMR 2008. Lau MRM 2010 NMR Biomed 2011
Hyperpolarization Methods
Metabolite-specific Imaging
§ Spectral-spatial excitation of individual metabolites with variable flip angles
§ Ramp sampled, symmetric EPI
§ 16 slices of pyruvate and lactate images in 2 s
April 26, 2017
Cunningham et al. JMR 2008.Gordon et al. MRM 2016. doi: 10.1002/mrm.26123.Gordon et al. Proc ISMRM 2017 #728.
Hyperpolarization Methods44
Clinical Metabolite-specific Imaging
April 26, 2017
Cunningham et al. Circ Res 2016, doi: 10.1161/CIRCRESAHA.116.309769.
a b c10
30
20
20
60
40
5
25
15
80
40
10
50
30
20
60
40
d e f
5 cm 5 cm 5 cm
5 cm 5 cm 5 cm
Pyruvate Bicarbonate Lactate
Pyruvate Bicarbonate Lactate
FIGURE 1
by guest on September 20, 2016
http://circres.ahajournals.org/D
ownloaded from
Cardiac imaging with Spiral Readout
Pyruvate
Lactate
Brain imaging with EPI Readout
Gordon et al. Proc ISMRM 2017 #728.
Hyperpolarization Methods45
Technique Comparison
Technique Pros Cons
MRSI Robust to off-resonanceFlexible spectral content
Slow
Spectral Decomposition (IDEAL/Dixon)
Speed+SNR Peak locations must be knownLimits on sequence parameters (TE)
Metabolite-specific Imaging Speed+SNR (max!)Works well with [1-13C]pyruvate spectrum
Sensitive to off-resonanceRequires spectrally separated metabolites
April 26, 201746 Hyperpolarization Methods
Parametrizations: Kinetic Modeling vs. alternativesNumerous options§ Kinetic modeling (e.g. kPL)§ Lactate/pyruvate§ Area-under-curve (AUCratio, Hill, et al.
PLoS One (2013).)
I advocate for unidirectional kPL model• Insensitive to bolus delivery with any
sampling strategy• Incorporate effects of RF pulses• Compare kPL (1/s) across sites, imaging
protocols, and anatomy• Simple (good for low SNR)
Daniels et al, NMR Biomed 2016, doi: 10.1002/nbm.3468
Hyperpolarization Methods47 April 26, 2017
Choice of Model
§ Models evaluated by Akaike Information Criteria (AIC) which balances fit quality with number of model parameters
a. Pyr-lac (all lumped)
b. Extravascular and intravascular compartments
c. Extravascular/extracellular, intracellular, and intravascular compartments
§ Assumptions
• Neglect kLP, lactate transport. Gamma-variate pyruvate input
• Pyruvate input estimated from heart voxels
• T1P = 45s, T1L = 25s
Bankson et al. Cancer Research 2016. doi: 10.1158/0008-5472.CAN-15-0171
April 26, 2017Hyperpolarization Methods48
“Input-less” Fitting§ Actual pyruvate signal as input, change in
lactate as output
§ No assumptions or fitting of pyruvate input
§ Pros: Simple, insensitive to fitting errors in pyruvate (e.g. incorrect bolus shape), works with any sampling strategy
§ Cons: No estimate of perfusion
April 26, 2017
Inpsired by: Khegai, et al. NMR Biomed 2014, Bahrami, et al. Quant Imaging Med Surg 2014.
Hyperpolarization Methods49
kPL
https://github.com/agentmess/hyperpolarized-mri-toolbox
Hyperpolarized-MRI-Toolbox
https://github.com/agentmess/hyperpolarized-mri-toolbox§ MATLAB tools for designing and analyzing HP MRI§ Open-source, contribute your coolest code!
§ Current Features• EPSI waveforms• Spectral-spatial RF• Variable flip angles
• Kinetic Modeling• Numerical phantom
§ Coming soon: Datasets for standardized comparisons of analysis methods
April 26, 2017Hyperpolarization Methods50
top related