Acceleration in MRI: Phased array coils, parallel imaging, simultaneous multi-slice imaging
2 Barth et al.
FIG. 1. First ever in vivo SMS images obtained from the leg of a healthy volunteer using a four element coil array. The top row shows the images obtained from each coil, the bottom row shows the disentangled slices using a SENSE reconstruction. Figure taken with kind permission from Larkman et al (12).
RF pulses used, that the power deposition would increase linearly with the number of excited slices, and if the pulses had identical phase then the peak power would increase with N2. The peak power could, however, be reduced by finding an optimum phase variation between the pulses (8).
The revolution in imaging acquisition brought about by the introduction of parallel imaging techniques (9–11) at the turn of the millennium had an immediate impact on simultaneous multislice imaging. The possibility of disentangling simultaneously excited slices by means of the coil sensitivity profiles and, thereby dramatically reducing acquisition times, was initially demonstrated by Larkman et al in 2001 (12) (Fig. 1). However, this idea
planar imaging (EPI) in particular, the benefits of SMS are qualitatively different than for standard parallel imaging. The latter reduces the duration of the EPI read- out; hence, both the distortion and the SNR are corre- spondingly reduced. SMS leads to a reduction in acquisition time by a factor N, with no impact on distor- tion or SNR. Although an earlier abstract had demon- strated the combination of EPI with SMS (14), it was the subsequent publications of highly accelerated EPI images by Moeller et al (15), (initially in abstract form (16), who also introduced the term “multiband”) and Feinberg et al (17) that drew widespread attention to this technique. The development of more powerful and robust recon- struction techniques (15,18), combined with the inven-
was only followed up by a relatively small community, tion of blipped CAIPIRINHA (18), that brought the
Background
} MRI is inherently slow compared to other imaging modalities
} This mainly due to the long relaxation times (T1, T2(*)) and the need to acquire images line-by-line and slice-by-slice
} Approaches to reduce measurement time were Partial Fourier, and the acquisition of multiple echoes per excitation (TSE, EPI)
Multi-channel coils to gain SNR
Parallel Imaging
It’s about doing things in parallel using multiple detectors / coils
4321
8
7 6
5
8 channel RX
We get as many data sets as we have coils
4321
8
7 6
5
How to combine the images?
Sum of Squares:
• near optimal SNR
• signal phase destroyed
I= åi2 j
j
Conjugate Sensitivity: å
• optimal SNR
• phase is maintained • intensity correction
is åj j
I=l
j
s2
SNR gains from image combination
Siemens CP (‘bird cage’) against SNR gain for Siemens 12-channel head coil at 1.5T 8-channel coil at 1.5T
Partial Parallel Imaging
} In partial parallel imaging ‘undersample’ k-space, mainly to save time
Replace some of the spatial encoding conventionally done by imaging gradients in the sequence, by simultaneously acquiring the signal with a number of coils in different places.
k-space undersampling = skipping lines
reduction of sampling density in k-space by factor R
results in an
‘aliasing artifact‘ or
‘fold-over‘
1/FoV
FoV
FoVR
1/FoVR
R = 2, a = 2, rate 2 , …
These coils save time with parallel imaging
• Pixel in reduced FOV is a superposition of multiple signal components
C1
C1
aliased pixel:
V1: 𝜌1 V2: 𝜌2 C2
V1+V2
C2
a1 = sC1,1 × 𝜌1 + sC!,2 × 𝜌2 a2 =s2,1 ×𝜌1 +s2,2 ×𝜌2
Partial Parallel Imaging – methods
} in other words: due to using multiple receivers there is some redundancy in the data that allows you to get away with sampling less
} Two distinctly different families of partial parallel imaging: } GRAPPA (Generalized autocalibrating partially parallel
acqusitions) – Griswold et al MRM 2002, Sodickson et al MRM 1999 (SMASH) } SENSE (Sensitivity Encoding) – Pruessman et al MRM 1999
SENSE
So to get r we have to invert S… a=S×r
Three possible cases
Linear Algebra
R > Ncoils
R = Ncoils
R < Ncoils
underdetermined
L
simple inverse S-1, S-1 a = S-1S × r
image: r = S-1× a K
pseudo inverse, SH a = (SH S)× r
find least squares optimum image: r = (SH S)-1 SH × a
J
SENSE summary
} works in image space
} reconstruct aliased image for each coil
} use sensitivity maps and images to set up linear equations and get one correct image, cannot get individual coil images
GRAPPA
} complete data set for each coil
} reconstruct separately (FT) and then combine the images as described before (SoS)
GRAPPA summary
} GRAPPA works in k-space
} does not require explicit knowledge of coil sensitivities } GRAPPA allows reconstruction of individual coil images
Partial Parallel Imaging - recap
Under- sampled
data
weight maps
sensitivity maps
Finished Image!
GRAPPA
SENSE
Partial Parallel Imaging - SNR
} Parallel Imaging without acceleration: • no time saving
}
• SNR will go up (image combination)
• SNR gains are inhomogeneous
Parallel Parallel Imaging with acceleration:
• time saving by factor R
• SNR will go down
• ‘geometry or g-factor’ noise from reconstruction
• Spatially varying sensitivity losses
SNRSENSE = SNR0
g accel.
Partial Parallel Imaging – anatomical scan
9’20’’
e.g. 3D MPRAGE T1 weighted anatomical
4’00’’
Simultaneous multislice imaging: 2D multi-slice imaging can be slow
with blipped CAIPIRINHA technique
This can significantly accelerate the acquisition!
History
} Multiple slices are excited and sampled simultaneously (Bolinger and Leigh, JMR 1988; Glover, JMRI 1991)
} Disentangled during reconstruction using coil sensitivity 2 Barth et al.
information (Larkman et al, JMRI 2001)
FIG. 1. First ever in vivo SMS images obtained from the leg of a healthy volunteer using a four element coil array. The top row shows the images obtained from each coil, the bottom row shows the disentangled slices using a SENSE reconstruction. Figure taken with kind permission from Larkman et al (12).
} Application to EPI (fMRI and DWI) (Moeller et al, MRM 2010;
RF pulses used, that the power deposition would increase planar imaging (EPI) in particular, the benefits of SMS
linearly with the number of excited slices, and if the are qualitatively different than for standard parallel
pulses had identical phase then the peak power would imaging. The latter reduces the duration of the EPI read-
Feinberg et al, PlosONE 2010)
increase with N2. The peak power could, however, be reduced by finding an optimum phase variation between the pulses (8).
The revolution in imaging acquisition brought about by the introduction of parallel imaging techniques (9–11) at the turn of the millennium had an immediate impact on simultaneous multislice imaging. The possibility of disentangling simultaneously excited slices by means of the coil sensitivity profiles and, thereby dramatically reducing acquisition times, was initially demonstrated
out; hence, both the distortion and the SNR are corre- spondingly reduced. SMS leads to a reduction in acquisition time by a factor N, with no impact on distor- tion or SNR. Although an earlier abstract had demon- strated the combination of EPI with SMS (14), it was the subsequent publications of highly accelerated EPI images by Moeller et al (15), (initially in abstract form (16), who also introduced the term “multiband”) and Feinberg et al (17) that drew widespread attention to this technique. The development of more powerful and robust recon-
by Larkman et al in 2001 (12) (Fig. 1). However, this idea struction techniques (15,18), combined with the inven-
Breuer et al, MRM 2005
Controlled Aliasing in Parallel Imaging Results in Higher Acceleration (CAIPIRINHA)
0 0,0 X 0 0,0
0 0,0
0 0,0
0,0 0,0 0,0 0,0
0,0 0,π 0,0 0,π 0,0
0
OR
0π0π
0 0,π π 0,0 0π 0,π
Breuer et al, MRM 2005
Slice 2 Slice 2 Slice 1
Summary
} Phased array coils increase image SNR and are a prerequisite for data undersampling and image acceleration
} SENSE and GRAPPA are two methods to perform the reconstruction of undersampled data by using coil sensitivity information
} SMS/Multiband imaging can reduce acquisition time significantly by sampling multiple slices simultaneously
} Concepts like VERSE and (multi)PINS help to reduce SAR constraints of multiband pulses