Introduction
1/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Cram´ e-Rao Lower Bounds
2/41
Definition A Cram´ er-Rao Lower Bound (CRB) is the lower bound on the variance of estimators of a deterministic parameter. In other words, CRB represents the lowest estimation error of a parameter in the case of an unbiased estimator. Bounds Calculation In theory, CRB are calculated from the ”true” parameters values of the signal cm , T2m , △αm , △ωm and φ0 . NEMESIS determine CRB from the estimated values of parameters by : • Building the Fisher Information Matrix (FIM) knowing the second derivative
matrix of the model function xˆ(nt2 , nt1 ) • Inverting the (FIM) • Extracting the diagonal values
Error bars In order to calculate the estimation errors, CRB values should be normalised knowing the noise level of the signal. y GPC GSH Lac mIns N
Introduction
3/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Introduction
4/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Development of new acquisition strategies and quantification methods for in vivo 2D Magnetic Resonance Spectroscopy PhD candidate : Tangi ROUSSEL
11th of July 2012 PhD Committee : Examiners :
Supervisors :
Olivier LAFON Lotfi SENHADJI Serge AKOKA Thomas LANGE
Professeur des Universit´es, Universit´e de Lille I Professeur des Universit´es, Universit´e de Rennes I Professeur des Universit´es, Universit´e de Nantes PhD, University Medical Center Freiburg, Allemagne
Sophie CAVASSILA H´ el` ene RATINEY
Professeur des Universit´es, Universit´e de Lyon I Charg´ee de recherches, Universit´e de Lyon I
Introduction
5/41
State of the art
Goals
Quantitative conventional 2D MRS
1
Introduction Magnetic Resonance Spectroscopy Understanding a MRS signal 1D MRS : Limitations In vivo 2D MRS
2
State of the art
3
Goals
4
Quantitative conventional 2D MRS Aims 2D J-PRESS WISH sequence NEMESIS quantification procedure Irregular sampling Conclusions
5
Ultrafast 2D MRS Aims Introduction ufJPRESS sequence Optimising the sequence Data reconstruction & Post-processing procedures
6
Conclusions & Perspectives
Ultrafast 2D MRS
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Magnetic Resonance Spectroscopy
6/41
Magnetic Resonance Imaging Magnetic Resonance Imaging (MRI) is a medical imaging technique used in radiology to visualize internal structures of the body. MRI makes use of the property of NMR to image hydrogen nuclei (protons) inside the body tissue.
Fig. 1: 1.5 T Fast Spin Echo T2 -weighted image from the brain of a patient suffering with glioma
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Magnetic Resonance Spectroscopy
6/41
Magnetic Resonance Imaging Magnetic Resonance Imaging (MRI) is a medical imaging technique used in radiology to visualize internal structures of the body. MRI makes use of the property of NMR to image hydrogen nuclei (protons) inside the body tissue. Magnetic Resonance Spectroscopy Magnetic Resonance Spectroscopy (MRS) allows non-invasive and in vivo exploration of the molecular composition of tissue. It identifies certain molecular constituents - metabolites - involved in physiological or pathological processes. 12
MRS signal
10
8
6
4
2
Quantification Ala Asp Cho CreGABAGlc Gln Glu Gly GPCGSH Lac mIns NAANAAGPCr PCho PE Tau
Metabolites
Fig. 2: Principle of quantitative MRS of human brain using a PRESS sequence
0
Concentration (mmol/kg)
Fig. 1: 1.5 T Fast Spin Echo T2 -weighted image from the brain of a patient suffering with glioma
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Understanding a MRS signal
Interpreting an in vivo NMR spectrum • 12 detectable metabolites at 4.7 T for short
echo time • Linear combination of metabolic spectral
signatures • Baseline (macromolecular contamination) • Noise
NAA
Cre
Cre PCr Glu Cho Gln mIns Tau mIns
In vivo signal metabolites + background + noise Glu Gln
5
4.5
Lip Lac
NAA Asp
In vivo measured background
4
3.5
3
2.5
2
1.5
Lip
1
Chemical shift (ppm)
Fig. 3: In vivo MR spectrum acquired from a rat brain at 7 T (PRESS sequence, TE =20 ms, 10 min scan time) 7/41
Ultrafast 2D MRS
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Understanding a MRS signal
7/41
Interpreting an in vivo NMR spectrum
Metabolic signature
• 12 detectable metabolites at 4.7 T for short
echo time
Each metabolite has a spectral signature characterised by :
• Linear combination of metabolic spectral
• Chemical shifts (ppm)
signatures
• J-coupling constants (Hz)
• Baseline (macromolecular contamination)
• Concentration (mmol/kg)
• Noise
Cre PCr Glu Cho Gln mIns Tau mIns
In vivo signal metabolites + background + noise Glu Gln
4.5
Lip Lac
NAA Asp
In vivo measured background
5
Ala Asp
NAA
Cre
4
3.5
3
2.5
2
1.5
Lip
1
Chemical shift (ppm)
Cho Cre GABA Gln Glu Lac mIns NAA Tau 5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
Chemical shift (ppm)
Fig. 3: In vivo MR spectrum acquired from a rat brain at 7 T (PRESS sequence, TE =20 ms, 10 min scan time)
Fig. 4: Metabolic spectral signatures
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
1D MRS : Limitations
8/41
NAA
Cre
Cre PCr Glu Cho Gln mIns Tau mIns
In vivo signal metabolites + background + noise
GABA
Glu Gln Lip Lac
NAA Asp
In vivo measured background
Lip
Gln Glu 5
4.5
4
3.5
3
2.5
2
1.5
Chemical shift (ppm) 5
4.5
4
3.5
3
2.5
2
1.5
1
Chemical shift (ppm)
Fig. 6: Spectral overlapping between the metabolites
Fig. 5: In vivo NMR spectrum acquired from a rat brain at 7 T
Limitations 1D MRS used with low B0 fields suffers with : • Spectral overlapping between metabolites (Glu/Gln) • Spectral overlapping between metabolites and macromolecules (baseline) • Low concentrated metabolites (GABA)
1
0.5
0
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
In vivo 2D MRS
Conventional localised 2D J-resolved Magnetic Resonance Spectroscopy
9/41
dime
n
J-resolved MRS = TE
dim
nsio
ens ion
FFT along 20
FFT
Spin-echo based acquisition
8 FID acquired Scan Time = 8 x TR
Chemical shift 5
4
3
2
1
15 10 5 0
(Hz)
FID
J-coupling constants
Voxel selection
−5 −10 −15 −20 0
(ppm)
Fig. 7: Understanding conventional 2D J-resolved MRS
A conventional localised 2D J-resolved MRS experiment consists in : • performing numerous acquisitions of FIDs for various Echo Times • Fourier transforming each FID to obtain F2 dimension • Fourier transforming the data set following the t1 dimension
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
In vivo 2D MRS
Conventional localised 2D J-resolved Magnetic Resonance Spectroscopy
dime
n
J-resolved MRS = TE
dim
nsio
ens ion
FFT along 20
FFT
Spin-echo based acquisition
8 FID acquired Scan Time = 8 x TR
Chemical shift 5
4
3
2
1
15 10 5 0
(Hz)
FID
J-coupling constants
Voxel selection
−5 −10 −15 −20 0
(ppm)
Fig. 8: Understanding conventional 2D J-resolved MRS
Pros/Cons • ⊕ Information following F1 dimension : J-coupling constant • ⊖ Acquisition duration depends on the number of increments in Echo Time (nt1 )
resulting in expensive scan time (up to 1h30) 10/41
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
In vivo 2D MRS since 1995 • More and more studies on human brain1 • Few studies on small animal consisting mainly in metabolite identification2
1
Thomas M A et al, JMRI, 6 :453-459, 1996 Meric P et al, MAGMA, 17 :317-338, 2004 Schulte R F et al, NMR Biomed, 19 :255-263, 2006 4 Provencher S W et al, MRM, 30 :672-679,1993 5 Golub G H et al, SIAM JNA, 10 :413-432, 1973 6 Hiba B et al, MRM, 52 :658-662, 2004 7 Frydman L et al, PNAS USA, 99 :15858-15862, 2002 2 3
11/41
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
In vivo 2D MRS since 1995 • More and more studies on human brain1 • Few studies on small animal consisting mainly in metabolite identification2
Few studies on quantitative 2D MRS • One quantification algorithm dedicated to human brain 2D MRS
at 3 T : ProFit3 based on 1D quantification methods LCModel4 & VARPRO5 • No quantification algorithm for small animal brain 2D MRS
Fig. 9: ProFit algorithm 1
Thomas M A et al, JMRI, 6 :453-459, 1996 Meric P et al, MAGMA, 17 :317-338, 2004 Schulte R F et al, NMR Biomed, 19 :255-263, 2006 4 Provencher S W et al, MRM, 30 :672-679,1993 5 Golub G H et al, SIAM JNA, 10 :413-432, 1973 6 Hiba B et al, MRM, 52 :658-662, 2004 7 Frydman L et al, PNAS USA, 99 :15858-15862, 2002 2 3
11/41
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
In vivo 2D MRS since 1995 • More and more studies on human brain1 • Few studies on small animal consisting mainly in metabolite identification2
Few studies on quantitative 2D MRS • One quantification algorithm dedicated to human brain 2D MRS
at 3 T : ProFit3 based on 1D quantification methods LCModel4 & VARPRO5 • No quantification algorithm for small animal brain 2D MRS
Very few studies on acquisition time reduction of in vivo 2D MRS experiment • Few studies on reducing scan time of 2D J-resolved MR
spectroscopic imaging6 • In 2002, L. Frydman introduces ”ultrafast spectroscopy”7 for Fig. 9: ProFit algorithm 1
Thomas M A et al, JMRI, 6 :453-459, 1996 Meric P et al, MAGMA, 17 :317-338, 2004 Schulte R F et al, NMR Biomed, 19 :255-263, 2006 4 Provencher S W et al, MRM, 30 :672-679,1993 5 Golub G H et al, SIAM JNA, 10 :413-432, 1973 6 Hiba B et al, MRM, 52 :658-662, 2004 7 Frydman L et al, PNAS USA, 99 :15858-15862, 2002 2 3
11/41
high resolution spectroscopy, a new technique allowing the acquisition of 2D MRS spectra in very short time.
Introduction
12/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
This PhD thesis work on 2D MRS can be presented under two main headings : Quantitative 2D MRS for small animal brain • Design a 2D MRS sequence on a 4.7 and a 7 T Bruker Biospec
imaging systems • Perform in vivo 2D MRS localised acquisitions on rat brain • Develop a dedicated quantification algorithm to estimate metabolite Gly GPC GSH Lac mIns
concentrations from the acquired data
Introduction
12/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
This PhD thesis work on 2D MRS can be presented under two main headings : Quantitative 2D MRS for small animal brain • Design a 2D MRS sequence on a 4.7 and a 7 T Bruker Biospec
imaging systems • Perform in vivo 2D MRS localised acquisitions on rat brain • Develop a dedicated quantification algorithm to estimate metabolite Gly GPC GSH Lac mIns
concentrations from the acquired data
Acquisition time reduction of 2D MRS experiment • Design an ultrafast localised 2D MRS sequence on a 4.7 and a 7 T
Bruker Biospec imaging systems spoilers
• Optimise and validating the sequence on in vitro phantoms
Introduction
13/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Quantitative conventional 2D MRS
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Aims
14/41
Design a 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence • Irregular data sampling following t1 dimension • Inversion-recovery excitation to acquire 2D baseline
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Aims
14/41
Design a 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence • Irregular data sampling following t1 dimension • Inversion-recovery excitation to acquire 2D baseline
Develop a dedicated quantification algorithm • Complex time domain quantification & strong prior-knowledge • Irregular data sampling handling y GPC GSH Lac mIns N
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Aims
Design a 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence • Irregular data sampling following t1 dimension • Inversion-recovery excitation to acquire 2D baseline
Develop a dedicated quantification algorithm • Complex time domain quantification & strong prior-knowledge • Irregular data sampling handling y GPC GSH Lac mIns N
Develop an algorithm to optimise t1 sampling • Based on Cram´ er-Rao theory • Calculation of optimised sampling following t1 dimension in order to
increase quantification accuracy
14/41
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Aims
14/41
Design a 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence • Irregular data sampling following t1 dimension • Inversion-recovery excitation to acquire 2D baseline
Develop a dedicated quantification algorithm • Complex time domain quantification & strong prior-knowledge • Irregular data sampling handling y GPC GSH Lac mIns N
Develop an algorithm to optimise t1 sampling • Based on Cram´ er-Rao theory • Calculation of optimised sampling following t1 dimension in order to
increase quantification accuracy Validate the methods on small animal • In vivo acquisition performed on a 4.7 and a 7 T Bruker Biospec
imaging systems • Localised MRS experiments on mouse and rat brain
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
2D J-PRESS WISH sequence
Designing the sequence 20
15
2D J-resolved MRS
10
0
(Hz)
5
−5
• ⊕ Short acquisition time (in comparison with
COSY MRS) due to the small bandwidth required following F1 dimension
−10
−15
4
3.5
3
2.5
2
1.5
−20 1
• ⊕ Easy to implement on an MRI since the
pulse sequence is close to PRESS scheme
(ppm)
Fig. 10: Simulated 2D J-resolved MR spectrum of ethanol
15/41
8 2D J-Resolved Point Resolved SpectroScopy with Weighted Irregular Sampling Handling
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
2D J-PRESS WISH sequence
Designing the sequence 20
15
2D J-resolved MRS
10
0
(Hz)
5
−5
• ⊕ Short acquisition time (in comparison with
COSY MRS) due to the small bandwidth required following F1 dimension
−10
−15
4
3.5
3
2.5
2
1.5
−20 1
• ⊕ Easy to implement on an MRI since the
pulse sequence is close to PRESS scheme
(ppm)
Fig. 10: Simulated 2D J-resolved MR spectrum of ethanol
2D J-PRESS WISH8 sequence : Preparation period • Water suppression scheme (VAPOR module)
to reduce water signal • Saturation bands (OVS module) to reduce
outer volume signal artefacts
relaxation delay
RF Water Suppression + Outer Volume Suppression + Inversion Modules
• Inversion pulse (InvPulse module) to perform
inversion-recovery excitation 8
15/41
2D J-Resolved Point Resolved SpectroScopy with Weighted Irregular Sampling Handling
Fig. 11: 2D J-PRESS WISH preparation
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
2D J-PRESS WISH sequence
Designing the sequence
16/41
relaxation delay
RF
spoilers
Fig. 12: 2D J-PRESS WISH9 sequence
Excitation & Detection • Strongly based on PRESS excitation scheme • TE calculation mode can be chosen in order to profit from TE1 /TE2
9
2D J-Resolved Point Resolved SpectroScopy with Weighted Irregular Sampling Handling
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
2D J-PRESS WISH sequence
Designing the sequence
16/41
relaxation delay
RF
spoilers
Fig. 12: 2D J-PRESS WISH9 sequence
Excitation & Detection • Strongly based on PRESS excitation scheme • TE calculation mode can be chosen in order to profit from TE1 /TE2
Original functionalities • Regular or irregular sampling following t1 dimension • NA weighting of the increments acquired following t1 dimension 9
2D J-Resolved Point Resolved SpectroScopy with Weighted Irregular Sampling Handling
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
2D J-PRESS WISH sequence
In vivo experiment with 2D J-PRESS WISH sequence
17/41
2D J-PRESS WISH experiment • Performed on a horizontal 7T Bruker Biospec MRI • A 5-month-old rat was anaesthetised by inhalation of
isoflurane • Volume coil for emission and surface receive coil • The signal was collected from a 64 µL voxel • 1st and 2nd order local shim adjustment • TE =20 to 140 ms, N1 =24, TR=2.5 s, NA=96 (1h
36min acquisition time)
Fig. 13: Axial slice image of a rat brain (RARE, TE/TR=15/5166ms)
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
2D J-PRESS WISH sequence
In vivo experiment with 2D J-PRESS WISH sequence
17/41
2D J-PRESS WISH experiment • Performed on a horizontal 7T Bruker Biospec MRI • A 5-month-old rat was anaesthetised by inhalation of
isoflurane • Volume coil for emission and surface receive coil • The signal was collected from a 64 µL voxel • 1st and 2nd order local shim adjustment • TE =20 to 140 ms, N1 =24, TR=2.5 s, NA=96 (1h
36min acquisition time)
Fig. 13: Axial slice image of a rat brain (RARE, TE/TR=15/5166ms) 5
x 10
+20
Glu
+15
(Hz)
+10 +5 0
Macromolecules Cre
Cho
NAA
GABA
1.5 1
Cre
0.5
−5
0
−10 −15 −20
2
mIns 4.5
4
−0.5
Tau 3.5
Glu 3
2.5
−1 2
1.5
1
0.5
(ppm)
Fig. 14: In vivo MRS 2D J-resolved spectrum acquired in a rat brain with the 2D J-PRESS WISH sequence
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Model function
NEMESIS10 properties • 2D complex time-domain model function
consisting in a linear combination of metabolite signals
Asp
Ala
GABA
Cho
Cre
Glc Gln
Glu
Gsh
Lac
NAA
PCho
• Strong prior-knowledge consisting of a set of
2D metabolite signals simulated with GAMMA11 • Macromolecular contamination is handled by
Gly
GPC
mIns NAAG
gaussian modelisation of in vivo acquisitions • Levenberg-Marquardt optimisation algorithm
PCr
PEtha
Tau
Fig. 15: Set of 19 2D metabolite signals
18/41
10 Numeric Estimation Method for 2D Spectroscopy Irregulary Sampled data 11 Smith S A et al, JMR, A106 :75-105, 1994
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Model function
NEMESIS10 properties • 2D complex time-domain model function
consisting in a linear combination of metabolite signals
Asp
Ala
GABA
Cho
Cre
Glc Gln
Glu
Gsh
Lac
NAA
PCho
• Strong prior-knowledge consisting of a set of Gly
2D metabolite signals simulated with GAMMA11 • Macromolecular contamination is handled by
GPC
mIns NAAG
gaussian modelisation of in vivo acquisitions • Levenberg-Marquardt optimisation algorithm
PCr
PEtha
Tau
Fig. 15: Set of 19 2D metabolite signals M X
» – t1 xˆ(t2 , t1 ) = exp(i · φ0 ) cm · xˆ(t2 , t1 ) · exp (− ) + (△αm + i · △ωm ) · t2 T2m m=1 m
• cm : Amplitude/concentration of the metabolite signal x ˆ(t2 , t1 )m • T2m : Transverse relaxation time [s] • △αm = 1∗ : Extra damping factor [Hz] T2m • △ωm : Frequency shift [Hz] • φ0 : Global zero-order phase [rad] 10
18/41
Numeric Estimation Method for 2D Spectroscopy Irregulary Sampled data 11 Smith S A et al, JMR, A106 :75-105, 1994
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
19/41
C =
nt2 =N2 nt1 =N1 X X nt2 =0 nt1 =0
12
[x(nt2 , nt1 ) − xˆ(nt2 , nt1 )]2
• x(nt2 , nt1 ) : data signal • x ˆ(nt2 , nt1 ) : model signal
Asp, Ala, Cho, Cre, GABA, Glc, Gln, Glu, Gly, GPC, Gsh, Lac, m-Ins, NAA, NAAG, PCr, PCho, PE, Tau
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
19/41
C =
nt2 =N2 nt1 =N1 X X nt2 =0 nt1 =0
[x(nt2 , nt1 ) − xˆ(nt2 , nt1 )]2
• x(nt2 , nt1 ) : data signal • x ˆ(nt2 , nt1 ) : model signal
Optimisation • 19 metabolites12 + macromolecular contamination = up to 40 parameters ! • Possible optimisation problems such as local minima
12
Asp, Ala, Cho, Cre, GABA, Glc, Gln, Glu, Gly, GPC, Gsh, Lac, m-Ins, NAA, NAAG, PCr, PCho, PE, Tau
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
19/41
C =
nt2 =N2 nt1 =N1 X X
[x(nt2 , nt1 ) − xˆ(nt2 , nt1 )]2
nt2 =0 nt1 =0
• x(nt2 , nt1 ) : data signal • x ˆ(nt2 , nt1 ) : model signal
Optimisation • 19 metabolites12 + macromolecular contamination = up to 40 parameters ! • Possible optimisation problems such as local minima
Quantification strategy • 4-stage quantification (total calculation time = 3 min) • Gradual increase of the number of estimated parameters to reduce optimisation
problems
12
Asp, Ala, Cho, Cre, GABA, Glc, Gln, Glu, Gly, GPC, Gsh, Lac, m-Ins, NAA, NAAG, PCr, PCho, PE, Tau
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
19/41
C =
nt2 =N2 nt1 =N1 X X
[x(nt2 , nt1 ) − xˆ(nt2 , nt1 )]2
nt2 =0 nt1 =0
• x(nt2 , nt1 ) : data signal • x ˆ(nt2 , nt1 ) : model signal
Optimisation • 19 metabolites12 + macromolecular contamination = up to 40 parameters ! • Possible optimisation problems such as local minima
Quantification strategy • 4-stage quantification (total calculation time = 3 min) • Gradual increase of the number of estimated parameters to reduce optimisation
problems Multistart Optimisation • For each quantification stage, multiple gaussian random starting values are
initialised around the values evaluated at the last stage in order to reduce optimisation problems related to initial parameters value • The estimated parameters are kept for a minimal fit residue 12
Asp, Ala, Cho, Cre, GABA, Glc, Gln, Glu, Gly, GPC, Gsh, Lac, m-Ins, NAA, NAAG, PCr, PCho, PE, Tau
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
20/41
1. Base line estimation
4
x 10
Amplitude
15
• Quantification of inversion-recovery 2D MRS data
10
5
• Linear combination of 20 gaussian components
0 40 35
1 30
(m
s)
2 3
25
4 20
5
)
(ppm
• Integration in NEMESIS prior-knowledge
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
20/41
1. Base line estimation
4
x 10
Amplitude
15
• Quantification of inversion-recovery 2D MRS data
10
5
• Linear combination of 20 gaussian components
0 40 35
1 30
(m
s)
2 3
25
4 20
5
)
(ppm
• Integration in NEMESIS prior-knowledge 7
3
2. Global frequency shift estimation • The parameter △ωm is initialised for all metabolites
Amplitude
• Maximum peak detection of singlets : Cho, Cre, NAA
x 10
2.5 2 1.5
Data signal
1
20 ms 28 ms 36 ms 44 ms
0.5 0 5
4.5
4
3.5
3
2.5
2
(ppm)
1.5
1
0.5
0
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
1. Base line estimation
4
x 10
Amplitude
15
• Quantification of inversion-recovery 2D MRS data
10
5
• Linear combination of 20 gaussian components
0 40 35
1 30
2
(m
3
25
s)
)
(ppm
4 20
5
• Integration in NEMESIS prior-knowledge 7
3
2. Global frequency shift estimation • The parameter △ωm is initialised for all metabolites
Amplitude
• Maximum peak detection of singlets : Cho, Cre, NAA
x 10
2.5 2 1.5
Data signal
1
20 ms 28 ms 36 ms 44 ms
0.5 0 5
4.5
4
3.5
3
2.5
2
1.5
1
(ppm)
3. Singlets estimation
7
x 10 2 1.5
• cm parameter is estimated for Cho, Cre et NAA
1 0.5
• Global estimation of the parameters T2m , △αm , △ωm
0 140 120
0
100
1
80
(ms)
20/41
2
60
3
40 20
4 5
(ppm)
and φ0
0.5
0
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Quantification strategy
20/41
1. Base line estimation
4
x 10
Amplitude
15
• Quantification of inversion-recovery 2D MRS data
10
5
• Linear combination of 20 gaussian components
0 40 35
1 30
2
(m
3
25
s)
)
(ppm
4 20
5
• Integration in NEMESIS prior-knowledge 7
3
2. Global frequency shift estimation • The parameter △ωm is initialised for all metabolites
Amplitude
• Maximum peak detection of singlets : Cho, Cre, NAA
x 10
2.5 2
Data signal
1.5 1
20 ms 28 ms 36 ms 44 ms
0.5 0 5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
(ppm)
3. Singlets estimation
7
x 10 2 1.5
• cm parameter is estimated for Cho, Cre et NAA
1 0.5
• Global estimation of the parameters T2m , △αm , △ωm
0 140 120
0
100
1
80
(ms)
2
60
3
40 20
4 5
(ppm)
and φ0
4. Global estimation • Parameters cm , T2m , △αm and △ωm are estimated for
each metabolite • Global estimation of the phase parameter φ0
7
x 10 2 1.5 1 0.5 0 140 120
0
100
(ms)
1
80
2
60
3
40 20
4 5
(ppm)
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Monte Carlo validation
21/41
Quantification strategy vs. Single quantification : study on simulated data • A 2D simulated 7T MRS signal was generated with typical in vivo parameter
values and macromolecular contamination • 100 repetitions of the quantification procedure were performed for the above
simulated data added to Gaussian noise • Quantification was performed with and without the quantification strategy • Biases and standard deviations were computed for the amplitude estimates 20
(Hz)
10
0
−10
−20 5
4
3
2
1
0
(ppm) Fig. 16: Frequency representation of a simulated 2D J-resolved MR spectrum for B0 =7 T
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
NEMESIS quantification procedure
Monte Carlo validation
22/41
Standard deviation (%)
Bias (%)
100
Single quantification Quantification strategy
75 50 25
0 100 75 50 25 0
Ala
Asp
Cho
Cre GABA Glc
Gln
Glu GSH Lac mIns NAA
Tau
Fig. 17: Bias and standard deviations calculated of the amplitude estimates
Results • Global reduction of standard deviation when using the quantification strategy • Slight reduction for Ala and Lac whose spectral signatures are strongly
overlapped with macromolecular contamination
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Irregular sampling
CRISO method
CRISO13 algorithm • Strongly relies on Cram´ er-Rao Lower Bounds (CRB) : lowest estimation error of
a parameter in the case of an unbiased estimator • Calculates optimised sampling following t1 dimension for each metabolite in order
to minimize CRB • Calculates a rank for each t1 increments according to the CRB reduction induced
23/41
13
Cram´ er-Rao guided Irregular Sampling Optimisation
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Irregular sampling
CRISO method
CRISO13 algorithm • Strongly relies on Cram´ er-Rao Lower Bounds (CRB) : lowest estimation error of
a parameter in the case of an unbiased estimator • Calculates optimised sampling following t1 dimension for each metabolite in order
to minimize CRB • Calculates a rank for each t1 increments according to the CRB reduction induced Ala 1st Asp 4th GABA
Glu 8th GSH 12th
Lac
mIns 16th
20
28
36
44
52
60
68
76
84
92
100
108
116
124
132
140
TE (ms)
Fig. 18: Graphical representation of optimised sampling calculated with CRISO for 7 coupled metabolites 23/41
13
Cram´ er-Rao guided Irregular Sampling Optimisation
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Irregular sampling
Monte Carlo validation
24/41
Sampling optimisation on simulated data • 200 repetitions of the quantification procedure • 4 sampling strategies were tested : 3 optimised samplings (Ala, Asp & GABA)
and a regular sampling
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Irregular sampling
Monte Carlo validation
24/41
Sampling optimisation on simulated data • 200 repetitions of the quantification procedure • 4 sampling strategies were tested : 3 optimised samplings (Ala, Asp & GABA)
and a regular sampling Optimised sampling for: Regular sampling:
Ala
Asp
GABA
100
Bias (%)
80 60 40
Standard deviation (%)
20 0 80 60 40 20 0
Ala
Asp
GABA
GSH
Lac
NAA
Fig. 19: Bias and standard deviations calculated of the amplitude estimates
Results • Global reduction of standard deviation when using an optimised sampling • Slight reduction of standard deviation for NAA which is not a strongly coupled
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Irregular sampling
In vivo validation
25/41
2D J-PRESS WISH experiment • Performed on a horizontal 7T Bruker Biospec MRI • A swiss mouse model anesthetised by inhalation of isoflurane • Volume coil for emission and surface receive coil • The signal was collected from a 90 µL voxel • TR=3 s, NA=128, TE sampling was set up in order to cover the 4 previous
tested strategies.
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Irregular sampling
In vivo validation
2D J-PRESS WISH experiment • Performed on a horizontal 7T Bruker Biospec MRI • A swiss mouse model anesthetised by inhalation of isoflurane • Volume coil for emission and surface receive coil • The signal was collected from a 90 µL voxel • TR=3 s, NA=128, TE sampling was set up in order to cover the 4 previous
tested strategies. 20
a.
x 10
7
Macromolecules 3
(Hz)
10 2 1
0
0 −10
+/−20
(Hz)
10
−1
b.
Cre
Glu Cho
x 10
GABA
NAA
7
3 2
0
1
Tau
−10
0
Cre −20 4.5
Glu
mIns 4
3.5
3
−1 2.5
2
1.5
1
0.5
(ppm)
25/41
Fig. 20: A 7T in vivo 2D JPRESS spectrum (a) and its estimated spectrum (b)
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Irregular sampling
In vivo validation
26/41
Concentration (mmol/kg)
Optimised sampling for: Regular sampling:
Ala
Asp
GABA
10
8
6
4
2
0
Asp
GABA
Gln
NAA
Fig. 21: In vivo quantification results : metabolite concentration estimates with CRB error bars
Results • Concentration estimates in agreement with literature were found using the Asp
dedicated optimised sampling • In agreement with previous results, quantification results for NAA are
independent of the sampling following t1 dimension
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Conclusions
2D J-PRESS WISH A new localised 2D MRS sequence handling irregular sampling and inversion-recovery excitation was designed
14 27/41
Roussel T et al, ESMRMB Antalya, 119, 2009
15 Roussel T et al, ISMRM-ESMRMB Stockholm, 904, 2010
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Conclusions
2D J-PRESS WISH A new localised 2D MRS sequence handling irregular sampling and inversion-recovery excitation was designed
NEMESIS14 A novel complex time domain quantification procedure relying on strong prior-knowledge was developed y GPC GSH Lac mIns N
14 27/41
Roussel T et al, ESMRMB Antalya, 119, 2009
15 Roussel T et al, ISMRM-ESMRMB Stockholm, 904, 2010
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Conclusions
2D J-PRESS WISH A new localised 2D MRS sequence handling irregular sampling and inversion-recovery excitation was designed
NEMESIS14 A novel complex time domain quantification procedure relying on strong prior-knowledge was developed y GPC GSH Lac mIns N
CRISO15 An algorithm dedicated to sampling optimisation for 2D J-resolved MRS was developed
14 27/41
Roussel T et al, ESMRMB Antalya, 119, 2009
15 Roussel T et al, ISMRM-ESMRMB Stockholm, 904, 2010
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Conclusions
2D J-PRESS WISH A new localised 2D MRS sequence handling irregular sampling and inversion-recovery excitation was designed
NEMESIS14 A novel complex time domain quantification procedure relying on strong prior-knowledge was developed y GPC GSH Lac mIns N
CRISO15 An algorithm dedicated to sampling optimisation for 2D J-resolved MRS was developed
Limitations The CRISO method, despite promising results, has a limited interest for the reduction of 2D MRS experiment acquisition time for in vivo application. 14 27/41
Roussel T et al, ESMRMB Antalya, 119, 2009
15 Roussel T et al, ISMRM-ESMRMB Stockholm, 904, 2010
Introduction
28/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Aims
29/41
Design an ultrafast 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence spoilers
• Spatial localisation of the ultrafast signal
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Aims
29/41
Design an ultrafast 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence spoilers
• Spatial localisation of the ultrafast signal
Develop a post-processing procedures • 2D MRS spectrum reconstruction from raw data • Spectrum quality enhancement using spatial apodisation
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Aims
29/41
Design an ultrafast 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence spoilers
• Spatial localisation of the ultrafast signal
Develop a post-processing procedures • 2D MRS spectrum reconstruction from raw data • Spectrum quality enhancement using spatial apodisation
Validate the methods on in vitro phantoms • Spatial localisation test on a dedicated GABA/Ethanol in vitro
phantom
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Aims
29/41
Design an ultrafast 2D MRS sequence on small imaging systems • 2D MRS J-resolved spectroscopy sequence spoilers
• Spatial localisation of the ultrafast signal
Develop a post-processing procedures • 2D MRS spectrum reconstruction from raw data • Spectrum quality enhancement using spatial apodisation
Validate the methods on in vitro phantoms • Spatial localisation test on a dedicated GABA/Ethanol in vitro
phantom
This work was supported by CNRS (”SiqMu”, PEPS-INSIS CNRS 2010 funding) and was carried out in close collaboration with P. Giraudeau and S. Akoka (CEISAM laboratory, Universit´ e de Nantes)
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Introduction
Ultrafast localised 2D J-resolved Magnetic Resonance Spectroscopy
dime
ens ion
FFT along
EPI based detection
20
No FFT required along !
8 increments (or more) Scan Time = TR
Chemical shift 5
4
3
2
1
15 10 5 0
(Hz)
Voxel selection
dim
J-coupling constants
Chemical shift information is spatially encoded !
n
nsio
−5 −10 −15 −20 0
(ppm)
Fig. 22: Understanding ultrafast 2D J-resolved MRS
An ultrafast localised 2D J-resolved MRS experiment consists in : • performing an ultrafast excitation that spatially encodes the chemical shift
information along one spatial dimension • performing an EPI16 -based detection to collect the ultrafast spectra for
numerous t1 increments • Fourier transforming the data set following the t1 dimension 16 30/41
Echo Planar Imaging
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
ufJPRESS sequence
Designing the sequence
phase cycles relaxation delay
RF ACQ Water Suppression + Outer Volume Suppression
spoilers
Modules
Fig. 23: ufJPRESS pulse sequence 3D localised 2D J-Resolved ultrafast MRS
17 31/41
UltraFast J-resolved Point Resolved SpectroScopy
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
ufJPRESS sequence
Designing the sequence
phase cycles relaxation delay
RF ACQ Water Suppression + Outer Volume Suppression
spoilers
Modules
Fig. 23: ufJPRESS pulse sequence 3D localised 2D J-Resolved ultrafast MRS
ufJPRESS17 : dedicated sequence to in vivo experiment • 7T Bruker Biospec imaging system (small animal) running with Paravision 5.1 • Signal collected with a quadrature coil (transmit/receive, 32 mm, Rapid Biomed) • Preparation : VAPOR module + OVS module
17 31/41
UltraFast J-resolved Point Resolved SpectroScopy
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
ufJPRESS sequence
Ultrafast excitation scheme
phase cycles relaxation delay
RF ACQ Water Suppression + Outer Volume Suppression
spoilers
Modules
Fig. 24: ufJPRESS pulse sequence 3D localised 2D J-Resolved ultrafast MRS
Ultrafast excitation scheme • PRESS localisation scheme nested in a modified version of the ultrafast
excitation scheme proposed by Pelupessy et al.18 • Original 90◦ and 180◦ PRESS slice pulses applied during G1 and G2 gradients
perform the spatial selection in the first two dimensions • Adiabatic 180◦ chirp pulses applied during bipolar excitation gradients (±Ge )
spatially encode the chemical shift information along the third spatial dimension 18 Pelupessy P et al, JACS, 125 :12345-12350, 2003 32/41
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
ufJPRESS sequence
Detection scheme
phase cycles relaxation delay
RF ACQ Water Suppression + Outer Volume Suppression
spoilers
Modules
Fig. 25: ufJPRESS pulse sequence 3D localised 2D J-Resolved ultrafast MRS
Detection scheme • EPI-based detection scheme • EPI bipolar gradients are replaced by a positive acquisition gradient Ga followed
by a 180◦ refocussing pulse19
• Preceded by a ”shifting” gradient Gc to adjust the position of the spectral window • Spectral resolution following the conventional dimension (F1 ) depends in inverse
proportion on the detection scheme duration Td 19 33/41
Giraudeau P et al, JPBA, 43 :1243-1248, 2007
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
ufJPRESS sequence
Phase cycles
34/41
phase cycles relaxation delay
RF ACQ Water Suppression + Outer Volume Suppression
spoilers
Modules
Fig. 26: ufJPRESS pulse sequence 3D localised 2D J-Resolved ultrafast MRS
Phase cycles • φ1 [+x,+y] phase cycle that reduces constant undesired signals (F1 =0 Hz) • φ2 [+y,+y,-y,-y] phase cycle that compensates imperfections of the 180◦ hard
pulse (spurious stimulated echoes) • φ1 phase cycle requires a minimum number of 2 accumulations
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Optimising the sequence
Excitation scheme optimisation
solution (70% w/w in water) • Signal collected from a 8 mm
x 8 mm x 8 mm voxel • Ultrafast excitation duration
100
Normalized peak intensity Peak linewidth
0.8
80
0.6
60
0.4
40
0.2
20
Peak linewidth in the ultrafast dimension (Hz)
• Highly concentrated ethanol
Normalized peak intensity
1
Method
Te = 2τ π optimisation 0 10
• Chirp pulse calibration using
20
30
40
50
60
70
80
90
0 100
Excitation duration (ms)
a spin-echo based sequence Fig. 27: CH3 peak intensity and peak linewidths in the ultrafast dimension according to excitation duration (Te )
35/41
20
Giraudeau P, PhD Thesis, Universit´ e de Nantes, 2008
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Optimising the sequence
Excitation scheme optimisation
solution (70% w/w in water) • Signal collected from a 8 mm
x 8 mm x 8 mm voxel • Ultrafast excitation duration
100
Normalized peak intensity Peak linewidth
0.8
80
0.6
60
0.4
40
0.2
20
Peak linewidth in the ultrafast dimension (Hz)
• Highly concentrated ethanol
Normalized peak intensity
1
Method
Te = 2τ π optimisation 0 10
• Chirp pulse calibration using
20
30
40
50
60
70
80
90
0 100
Excitation duration (ms)
a spin-echo based sequence Fig. 27: CH3 peak intensity and peak linewidths in the ultrafast dimension according to excitation duration (Te )
Results • Signal-to-Noise (S/N) ratio strongly decreases according to Te 20 • Ultrafast spectral resolution increases according to Te • Good compromise is reached for Te = 30 ms
35/41
20
Giraudeau P, PhD Thesis, Universit´ e de Nantes, 2008
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Optimising the sequence
Excitation scheme optimisation
solution (70% w/w in water) • Signal collected from a 8 mm
x 8 mm x 8 mm voxel • Ultrafast excitation duration
100
Normalized peak intensity Peak linewidth
0.8
80
0.6
60
0.4
40
0.2
20
Peak linewidth in the ultrafast dimension (Hz)
• Highly concentrated ethanol
Normalized peak intensity
1
Method
Te = 2τ π optimisation 0 10
• Chirp pulse calibration using
20
30
40
50
60
70
80
90
0 100
Excitation duration (ms)
a spin-echo based sequence Fig. 27: CH3 peak intensity and peak linewidths in the ultrafast dimension according to excitation duration (Te )
Results • Signal-to-Noise (S/N) ratio strongly decreases according to Te 20 • Ultrafast spectral resolution increases according to Te • Good compromise is reached for Te = 30 ms
Limitations • Short excitation duration Te requires high RF power for chirp pulses • Long excitation duration Te requires high gradient strength for Ga 35/41
20
Giraudeau P, PhD Thesis, Universit´ e de Nantes, 2008
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Optimising the sequence
Ultrafast vs. conventional 2D J-Resolved MRS
Method 20
a.
• Concentrated ethanol solution (10% w/w
15
in water)
5 0
(Hz)
10
experiments were performed with n1 =128 following t1 dimension
−10 −15
4
3
2
1
−20 0
(ppm) 20
b.
15
0
(Hz)
10 5
−5 −10 −15
5
4
3
2
1
−20 0
(ppm)
Fig. 28: Tilted conventional (a) and ultrafast (b) localised 2D J-resolved spectra 36/41
mm voxel • Conventional and ultrafast 2D J-resolved
−5
5
• Signal collected from a 8 mm x 8 mm x 8
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Optimising the sequence
Ultrafast vs. conventional 2D J-Resolved MRS
36/41
Method 20
a.
• Concentrated ethanol solution (10% w/w
15
in water)
5 0
(Hz)
10
experiments were performed with n1 =128 following t1 dimension
−10 −15
4
3
2
1
mm voxel • Conventional and ultrafast 2D J-resolved
−5
5
• Signal collected from a 8 mm x 8 mm x 8
−20 0
(ppm) 20
b.
F2 dimension F1 dimension Scan time
15
5 0
(Hz)
10
−5 −10
Ultrafast 17.0 Hz 2.3 Hz 20 s
Conventional 1.9 Hz 1.3 Hz 21 mins
Tab. 1: CH3 peak linewidths and scan time comparison between ultrafast and conventional 2D J-resolved MRS experiments
−15
5
4
3
2
1
−20 0
(ppm)
Fig. 28: Tilted conventional (a) and ultrafast (b) localised 2D J-resolved spectra
Observation The chemical shifts and the J coupling values of both spectra are in good agreement with literature data
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Optimising the sequence
Voxel localisation
37/41
Method • Purpose-built phantom for localisation
Fig. 29: Sagittal and axial images of the γ-Aminobutyric acid (GABA) in vitro phantom (FLASH, TE/TR = 5.4/100 ms)
(OVS) bands with a 0.5 mm gap to voxel
5
• Number of Accumulations (NA) = 16
(Hz)
0 −5 −10 −15
3
2
x 5 mm voxel placed in the GABA solution
10
15
4
• Signal collected from a 5 mm x 5 mm
• 3 mm Outer Volume Saturation
20
5
tests : 1.5 mL tube containing a γ-Aminobutyric acid (GABA) solution (10% w/w in water) placed at the center of a 50 mL tube of pure ethanol
−20 1
(ppm)
Fig. 30: 3D localised 2D ultrafast J-resolved spectrum of GABA
resulting in a 2 min 40 s scan time
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Optimising the sequence
Voxel localisation
37/41
Method • Purpose-built phantom for localisation
Fig. 29: Sagittal and axial images of the γ-Aminobutyric acid (GABA) in vitro phantom (FLASH, TE/TR = 5.4/100 ms)
tests : 1.5 mL tube containing a γ-Aminobutyric acid (GABA) solution (10% w/w in water) placed at the center of a 50 mL tube of pure ethanol • Signal collected from a 5 mm x 5 mm
x 5 mm voxel placed in the GABA solution • 3 mm Outer Volume Saturation
20
10
(OVS) bands with a 0.5 mm gap to voxel
5
• Number of Accumulations (NA) = 16
0
(Hz)
15
resulting in a 2 min 40 s scan time
−5 −10
Efficient 3D localisation
−15
5
4
3
2
−20 1
(ppm)
Fig. 30: 3D localised 2D ultrafast J-resolved spectrum of GABA
Very low intensity ethanol peaks (at 1.19 and 3.67 ppm) were reported
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Data reconstruction & Post-processing procedures
38/41
20
a.
Ultrafast artefacts
15
• Raw ultrafast spectra present
5 0
(Hz)
10
−5
• These wiggles are inherent to
−10
ultrafast MR experiment
−15
5
4
3
2
1
−20 0
(ppm) 20
b.
15
0
(Hz)
10 5
−5 −10 −15
5
4
3
2
1
asymmetric sinc wiggles around peaks of interest
−20 0
(ppm)
Fig. 31: Raw (a) and post-processed (b) ultrafast 2D J-resolved spectra of a highly concentrated ethanol solution (70% w/w in water)
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Data reconstruction & Post-processing procedures
38/41
20
a.
Ultrafast artefacts
15
• Raw ultrafast spectra present
5 0
(Hz)
10
−5
asymmetric sinc wiggles around peaks of interest • These wiggles are inherent to
−10
ultrafast MR experiment
−15
5
4
3
2
1
−20 0
(ppm)
Automatic post-processing procedure 20
b.
• Based on spatial apodization
15
• The optimal apodization window
5 0
(Hz)
10
−5 −10 −15
5
4
3
2
1
−20 0
(ppm)
Fig. 31: Raw (a) and post-processed (b) ultrafast 2D J-resolved spectra of a highly concentrated ethanol solution (70% w/w in water)
width is automatically estimated in order to improve S/N ratio without decreasing spectral resolution • The ”apparent” S/N ratio is usually
2.5 times higher while linewidth increases by only 2 Hz
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Data reconstruction & Post-processing procedures
Ultrafast artefacts
4
Amplitude
20
x 10
a.
Gaussian apodization window
15 10
• Raw ultrafast spectra present
asymmetric sinc wiggles around peaks of interest
5
• These wiggles are inherent to ultrafast
0 −5 900
950
1000
1050
1100
1150
MR experiment21
Spatial dimension (pts)
(ms)
200
b.
Automatic post-processing procedure
Estimation of linewidth in the ultrafast dimension
150 100
• Based on spatial apodization22 • The optimal apodization window
50 0 1.8
1.6
1.4
1.2
1
(ppm)
width is automatically estimated in order to improve S/N ratio without decreasing spectral resolution • The ”apparent” S/N ratio is usually
Fig. 32: Inverse Fourier transformed ultrafast signals apodised in the spatial dimension with an optimised gaussian window (a) and the corresponding CH3 2D peak in the F2 t1 plan where linewidth estimation is performed (b) 21 Shapira B et al, JMR, 166 :152-163, 2004 22
Giraudeau P et al, MRC, 49 :307-313, 2011
39/41
2.5 times higher while linewidth increases by only 2 Hz
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Quantitative conventional 2D MRS • 2D J-PRESS WISH : a new localised 2D MRS sequence • NEMESIS : a novel complex time domain quantification procedure • CRISO : an algorithm for sampling optimisation Gly GPC GSH Lac mIns
23 24 40/41
25
Tk´ ac I et al, MRM, 41 :649-656, 1999 Roussel T et al, JMR, 215 :50-55, 2012 Roussel T et al, ISMRM Melbourne, 2012
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Quantitative conventional 2D MRS • 2D J-PRESS WISH : a new localised 2D MRS sequence • NEMESIS : a novel complex time domain quantification procedure • CRISO : an algorithm for sampling optimisation Gly GPC GSH Lac mIns
Perspectives • 2D J-PRESS WISH : ultra short TE acquisition23 • CRISO : integration of NA accumulation handling • NEMESIS : regularisation terms
23 24 40/41
25
Tk´ ac I et al, MRM, 41 :649-656, 1999 Roussel T et al, JMR, 215 :50-55, 2012 Roussel T et al, ISMRM Melbourne, 2012
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Quantitative conventional 2D MRS • 2D J-PRESS WISH : a new localised 2D MRS sequence • NEMESIS : a novel complex time domain quantification procedure • CRISO : an algorithm for sampling optimisation Gly GPC GSH Lac mIns
Perspectives • 2D J-PRESS WISH : ultra short TE acquisition23 • CRISO : integration of NA accumulation handling • NEMESIS : regularisation terms
Ultrafast 2D MRS • ufJPRESS : First 3D localised 2D J-resolved ultrafast MRS sequence spoilers
23 24 40/41
25
• Feasibility stage shows good quality for in vitro spectra24 25
Tk´ ac I et al, MRM, 41 :649-656, 1999 Roussel T et al, JMR, 215 :50-55, 2012 Roussel T et al, ISMRM Melbourne, 2012
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Quantitative conventional 2D MRS • 2D J-PRESS WISH : a new localised 2D MRS sequence • NEMESIS : a novel complex time domain quantification procedure • CRISO : an algorithm for sampling optimisation Gly GPC GSH Lac mIns
Perspectives • 2D J-PRESS WISH : ultra short TE acquisition23 • CRISO : integration of NA accumulation handling • NEMESIS : regularisation terms
Ultrafast 2D MRS • ufJPRESS : First 3D localised 2D J-resolved ultrafast MRS sequence spoilers
• Feasibility stage shows good quality for in vitro spectra24 25
Perspectives • Great potential as it could be combined to high SNR spectroscopy applications • Optimised trajectory in the plan (k/ν1 , F2 ) during detection • Ultrafast spectroscopic imaging 23 24 40/41
25
Tk´ ac I et al, MRM, 41 :649-656, 1999 Roussel T et al, JMR, 215 :50-55, 2012 Roussel T et al, ISMRM Melbourne, 2012
Introduction
41/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Publications • 1 first-author article : 3D localised 2D ultrafast J-resolved magnetic resonance
spectroscopy : In vitro study on a 7T imaging system, JMR, 215 :50-55, 2012 • 2 first-author oral communications in international conferences • 4 first-author poster communications in international conferences • 3 first-author communications in national conferences
Introduction
41/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Publications • 1 first-author article : 3D localised 2D ultrafast J-resolved magnetic resonance
spectroscopy : In vitro study on a 7T imaging system, JMR, 215 :50-55, 2012 • 2 first-author oral communications in international conferences • 4 first-author poster communications in international conferences • 3 first-author communications in national conferences
Scientific collaborations • ”SiqMu” project (PEPS CNRS 2010 funding) in close collaboration with P.
Giraudeau and S. Akoka (CEISAM laboratory, Universit´ e de Nantes) • Scientific collaboration with Frydman group in October 2012
Introduction
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Publications • 1 first-author article : 3D localised 2D ultrafast J-resolved magnetic resonance
spectroscopy : In vitro study on a 7T imaging system, JMR, 215 :50-55, 2012 • 2 first-author oral communications in international conferences • 4 first-author poster communications in international conferences • 3 first-author communications in national conferences
Scientific collaborations • ”SiqMu” project (PEPS CNRS 2010 funding) in close collaboration with P.
Giraudeau and S. Akoka (CEISAM laboratory, Universit´ e de Nantes) • Scientific collaboration with Frydman group in October 2012
MRI facilities • 4.7 T small animal Bruker Biospec MRI : CREATIS, CPE, Villeurbanne • 7 T small animal Bruker Biospec MRI : ANIMAGE, CERMEP, Bron • Bruker pulse programming course, Ettlingen, Germany
41/41
Introduction
41/41
State of the art
Goals
Quantitative conventional 2D MRS
Ultrafast 2D MRS
Conclusions & Perspectives
Publications • 1 first-author article : 3D localised 2D ultrafast J-resolved magnetic resonance
spectroscopy : In vitro study on a 7T imaging system, JMR, 215 :50-55, 2012 • 2 first-author oral communications in international conferences • 4 first-author poster communications in international conferences • 3 first-author communications in national conferences
Scientific collaborations • ”SiqMu” project (PEPS CNRS 2010 funding) in close collaboration with P.
Giraudeau and S. Akoka (CEISAM laboratory, Universit´ e de Nantes) • Scientific collaboration with Frydman group in October 2012
MRI facilities • 4.7 T small animal Bruker Biospec MRI : CREATIS, CPE, Villeurbanne • 7 T small animal Bruker Biospec MRI : ANIMAGE, CERMEP, Bron • Bruker pulse programming course, Ettlingen, Germany
Computer calculation facilities CREATIS computer cluster for Monte Carlo studies
Introduction
42/41
State of the art
Goals
Quantitative conventional 2D MRS
Thank you for your attention !
Ultrafast 2D MRS
Conclusions & Perspectives