j. su 1 , m.saranathan 1 , and b.k .rutt 1
DESCRIPTION
ISMRM 2012 E-Poster #4275. Accelerated Variable Flip Angle T 1 Mapping via View Sharing of Pseudo-Random Sampled Higher Order K-Space . J. Su 1 , M.Saranathan 1 , and B.K .Rutt 1 1 Department of Radiology, Stanford University, Stanford, CA, United States. 2.3x Undersamping. T 1 Maps. - PowerPoint PPT PresentationTRANSCRIPT
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE J.Su1, M.Saranathan1, and B.K.Rutt1
1Department of Radiology, Stanford University, Stanford, CA, United States
ISMRM 2012 E-POSTER #4275
Fully Sampled
2.3x Undersamping
T1 Maps
Accelerated % Difference
Declaration of Conflict of Interest or RelationshipI have no conflicts of interest to disclose with regard to the subject matter of this presentation.
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE J.Su1, M.Saranathan1, and B.K.Rutt1
1Department of Radiology, Stanford University, Stanford, CA, United States
ISMRM 2012 E-POSTER #4275
Background
• Variable flip angle T1 mapping (VFA) is a quantitative image method in which a series of scans at different flip angles are collected to extract whole brain relaxation times
• The collection of many angles for accuracy across the wide range of T1 values in tissue is time consuming1
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
1Deoni et al. Magn Reson Med. 2003 Mar;49(3):515-26.
Purpose
• Accelerate VFA by using a view sharing scheme similar to DISCO2
– A novel pseudo-random sampling pattern is used to greatly reduce the appearance of coherent artifacts
• Assess the accuracy and variation of the accelerated T1 maps compared to the fully sampled source
2Saranathan et al. J Magn Reson Imaging. 2012 Feb 14.
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Scanning Methods• 3T GE Signa MR750, 8-channel head RF coil
• VFA: 2mm isotropic covering whole brain, about 15 min., 110x110x80 matrix, fully sampled ellipse– SPGR: TE/TR = 1.2/3.7ms, α = 14 nonlinearly spaced angles,
shown below for simulated curves– This is necessary for accurate estimation of T1 across all brain
tissues
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
View Sharing Methods: Sampling
• This is represents the fully sampled ellipse of data points in Cartesian k-space
• Define two regions in k-space:– The center region (A)– The remaining outer region (B)
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Phas
e En
code
Slice Encode
B
A
A
View Sharing Methods: Sampling
• The center region (A)– 16% of k-space which is fully sampled– The center data is collected for every flip angle
frame
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Phas
e En
code
Slice Encode
A
B1 B2 B3B
View Sharing Methods: Sampling
• The outer region (B)– Broken down into 3 pseudo-random subsampling
patterns, each undersampled by a factor 3
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Phas
e En
code
Slice Encode
A
B
B1 B2 B3B
View Sharing Methods: Sampling
• The outer region (B)– Broken down into 3 pseudo-random subsampling
patterns, each undersampled by a factor 3– The patterns interlace and can be combined to
form a complete composite outer region
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Phas
e En
code
Slice Encode
A
B B1+B2+B3
View Sharing Methods: Sampling
• For each flip angle frame, a different undersampling pattern is used:– A+B1– A+B2– A+B3
• This results in a 2.3x acceleration of the acquisition
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
A+B1A+B2
A+B3
View Sharing Methods
• The sampling patterns are cycled through with each acquired flip angle frame
• Sequential sets of 3 frames are used to reconstruct the accelerated composite data like a sliding window
. . .. . . A+B1
A+B3
A+B2
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
View Sharing Methods: Reconstruction
• A complete composite of the current flip angle is formed by mixing in outer region samples from the previous and next adjacent angles– The fully sampled center region of the current flip angle is retained in its entirely
• For example, here the 2nd flip angle is created by combining with the 1st and 3rd angle
. . .. . .
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
View Sharing Methods: Reconstruction
• The first angle is a special edge case, outer samples are instead mixed from the following two angles
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
View Sharing Methods: Reconstruction
• The last angle is also a special case, samples are borrowed from the preceding two angles
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
View Sharing Methods: Reconstruction
• Borrowed samples are scaled to compensate for the difference between frames caused by the SPGR signal behavior– The average magnitude in the center region is
computed for each frame, μi
– The scale factor is then μ(target angle)/μ(borrowed angle)
– Fixes large discontinuities in k-space
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Reconstruction
• Matlab was used to synthesize the accelerated composites from the fully sampled acquired data
• The k-space data are then brought into the image domain by standard methods
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Post-Processing
• Linearly coregister and brain extract the images with FSL3
• Extract whole-brain T1 values by linearizing the data according to the SPGR signal equation and performing a fit4
• Comparisons are made to the fully sampled images on a voxel-by-voxel level
3FMRIB Software Library4Fram et al. Magn Reson Imaging. 1987;5(3):201-8.
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Results: Worst Case – First Flip Angle
• The percent difference between the accelerated and fully sampled volumes is shown
• The first and last flip angles are the least faithfully reconstructed since they borrow from non-adjacent angles
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
• Even in the worst case, the reconstructed SPGR images are very similar to the originals– Median percent difference: 0.025%– Interquartile range: 2.235%
Results: T1 Map – Fully SampledACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Results: T1 Map – AcceleratedACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
Results: T1 Map• The percent difference
between the accelerated and fully sampled T1 maps is shown
• Accuracy is excellent, the mean shift in T1 values is << 1%– Median: 0.030%
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275
• The variation in the difference map is very low over the whole brain– Standard deviation < 3% – Interquartile range: 1.492%
Discussion & Conclusions• 2.3x acceleration of VFA is reliably and simply achieved with
negligible loss in accuracy and precision
• Reconstruction is fast and possible on the scanner
• Compatible with parallel imaging methods like GRAPPA and SPIRiT– Simply combine the 3 sampling patterns by the parallel imaging
acquisition pattern– With a modest 2x2 acceleration, a net 6-7x speed up is easily achievable,
reducing this 15-minute 2mm isotropic protocol to 2.5 minutes
• Useful for high-resolution mcDESPOT acquisitions as well
ACCELERATED VARIABLE FLIP ANGLE T1 MAPPING VIA VIEW SHARING OF PSEUDO-RANDOM SAMPLED HIGHER ORDER K-SPACE ISMRM 2012 #4275