MRI Data Processing and Reconstruction via Chirp z-Transform
1. MRI Data Processing
and Reconstruction via
Chirp z-Transform
Author: Huiming Dong
Supervisor: Shouliang Qi, Ph. D
2. Introduction
m-SPRITE Imaging Sequence [Balcom et al.]
Derive from SPI sequences family [Emid et al.]
Pure phase encoding
Acquire multiple FID points after each RF excitation (See Fig. 1)
Particularly useful in fast-relaxation nuclei imaging
Fig. 1 m-SPRITE Imaging Sequence
Department of Biomedical Engineering, Northeastern University
3. Introduction
m-SPRITE Imaging Sequence [Balcom et al.]
Derive from SPI sequences family [Emid et al.]
Pure phase encoding
Acquire multiple FID points after each RF excitation (See Fig. 1)
Particularly useful in fast-relaxation nuclei imaging
Fig. 1 m-SPRITE Imaging Sequence
Department of Biomedical Engineering, Northeastern University
4. Introduction
m-SPRITE Imaging Sequence [Balcom et al.]
Derive from SPI sequences family [Emid et al.]
Pure phase encoding
Acquire multiple FID points after each RF excitation (See Fig. 1)
Particularly useful in fast-relaxation nuclei imaging
Fig. 1 m-SPRITE Imaging Sequence
Department of Biomedical Engineering, Northeastern University
5. Introduction
m-SPRITE Imaging Sequence [Balcom et al.]
Derive from SPI sequences family [Emid et al.]
Pure phase encoding
Acquire multiple FID points after each RF excitation (See Fig. 1)
Particularly useful in fast-relaxation nuclei imaging
Fig. 1 m-SPRITE Imaging Sequence
Department of Biomedical Engineering, Northeastern University
6. Introduction
m-SPRITE Imaging Sequence [Balcom et al.]
Derive from SPI sequences family [Emid et al.]
Pure phase encoding
Acquire multiple FID points after each RF excitation (See Fig. 1)
Particularly useful in fast-relaxation nuclei imaging
Fig. 1 m-SPRITE Imaging Sequence
Department of Biomedical Engineering, Northeastern University
7. Introduction
K-Space Data Acquired Utilizing m-SRITE Technique
Sampled in a non-uniform pattern (See Fig. 2)
Challenge the conventional FFT reconstruction methods
The k-space can be separated into Nt different uniformly sampled k-spaces
Each k-space per se has a different FOV size
Reconstruct respectively gives a low SNR
Fig. 2 Non-Uniformly Sampled Data
Department of Biomedical Engineering, Northeastern University
8. Introduction
K-Space Data Acquired Utilizing m-SRITE Technique
Sampled in a non-uniform pattern (See Fig. 2)
Challenge the conventional FFT reconstruction methods
The k-space can be separated into Nt different uniformly sampled k-spaces
Each k-space per se has a different FOV size
Reconstruct respectively gives a low SNR
Fig. 2 Non-Uniformly Sampled Data
Department of Biomedical Engineering, Northeastern University
9. Introduction
K-Space Data Acquired Utilizing m-SRITE Technique
Sampled in a non-uniform pattern (See Fig. 2)
Challenge the conventional FFT reconstruction methods
The k-space can be separated into Nt different uniformly sampled k-spaces
Each k-space per se has a different FOV size
Reconstruct respectively gives a low SNR
Fig. 2 Non-Uniformly Sampled Data
Department of Biomedical Engineering, Northeastern University
10. Introduction
K-Space Data Acquired Utilizing m-SRITE Technique
Sampled in a non-uniform pattern (See Fig. 2)
Challenge the conventional FFT reconstruction methods
The k-space can be separated into Nt different uniformly sampled k-spaces
Each k-space per se has a different FOV size
Reconstruct respectively gives a low SNR
Fig. 2 Non-Uniformly Sampled Data
Department of Biomedical Engineering, Northeastern University
11. Introduction
Chirp z-Transform (CZT) [Rabiner et al.]
A generalization of DFT
Evaluate signals on arbitrary contour on the z-plane (See Fig. 3)
The length of resultant signal can be set to any value for different practical applications
Computational complexity: Klog2K
N 1
n
X (k ) x ( n) z k zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
12. Introduction
Chirp z-Transform (CZT) [Rabiner et al.]
A generalization of DFT
Evaluate signals on arbitrary contour on the z-plane (See Fig. 3)
The length of resultant signal can be set to any value for different practical applications
Computational complexity: Klog2K
N 1
n
X (k ) x ( n) z k zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
13. Introduction
Chirp z-Transform (CZT) [Rabiner et al.]
A generalization of DFT
Evaluate signals on arbitrary contour on the z-plane (See Fig. 3)
The length of resultant signal can be set to any value for different practical applications
Computational complexity: Klog2K
N 1
n
X (k ) x ( n) z k zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
14. Introduction
Chirp z-Transform (CZT) [Rabiner et al.]
A generalization of DFT
Evaluate signals on arbitrary contour on the z-plane (See Fig. 3)
The length of resultant signal can be set to any value for different practical applications
Computational complexity: Klog2K
N 1
n
X (k ) x ( n) z k zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
15. Introduction
Chirp z-Transform (CZT) [Rabiner et al.]
A generalization of DFT
Evaluate signals on arbitrary contour on the z-plane (See Fig. 3)
The length of resultant signal can be set to any value for different practical applications
Computational complexity: Klog2K
N 1
n
X (k ) x ( n) z k zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
16. Method
FOV Scaling
DFT of a signal evaluates a signal on the whole unit circle on the z-plane
CZT can evaluate the signal on a part of the unit circle (See Fig. 4)
FOVdes FOV ( NT 1) t
T act
FOVact FOVact t max
1 ( FOVact FOVdes )
A0 1 0 2 (1 T )
2 FOVact
FOVdes T
W0 1 0 (2 ) / Nc 2
FOVact Nc
Fig. 4 Evaluating Contour
Department of Biomedical Engineering, Northeastern University
17. Method
FOV Scaling
DFT of a signal evaluates a signal on the whole unit circle on the z-plane
CZT can evaluate the signal on a part of the unit circle (See Fig. 4)
FOVdes FOV ( NT 1) t
T act
FOVact FOVact t max
1 ( FOVact FOVdes )
A0 1 0 2 (1 T )
2 FOVact
FOVdes T
W0 1 0 (2 ) / Nc 2
FOVact Nc
Fig. 4 Evaluating Contour
Department of Biomedical Engineering, Northeastern University
18. Method
FOV Scaling
DFT of a signal evaluates a signal on the whole unit circle on the z-plane
CZT can evaluate the signal on a part of the unit circle (See Fig. 4)
FOVdes FOV ( NT 1) t
T act
FOVact FOVact t max
1 ( FOVact FOVdes )
A0 1 0 2 (1 T )
2 FOVact
FOVdes T
W0 1 0 (2 ) / Nc 2
FOVact Nc
Fig. 4 Evaluating Contour
Department of Biomedical Engineering, Northeastern University
19. Method
DFT Reconstruction for m-SPRITE MRI Data
Numerous complex computation
Cannot be efficiently implemented by FFT algorithms
Require revised FFT or interpolation methods for reconstruction
N c 1
s(k ) x e j 2xk
dx x m s (k
n 0
n )e 2jkn xm
NT 1N G 1 xm G (v) t p (u )
xm s(G(v)t (u ))e j NG ( )( )( )
u 0 v 0
p xmax Gmax tmax
m 1 2v t p (u ) m 1 2v
N G ( )( 1)( ) N G ( )( 1)T (u )
NC 2 NG t max NC 2 NG
j 2vmT(u) / NC jvT (u) jmT (u) / NT jNGT (u) / 2
Department of Biomedical Engineering, Northeastern University
20. Method
DFT Reconstruction for m-SPRITE MRI Data
Numerous complex computation
Cannot be efficiently implemented by FFT algorithms
Require revised FFT or interpolation methods for reconstruction
N c 1
s(k ) x e j 2xk
dx x m s (k
n 0
n )e 2jkn xm
NT 1N G 1 xm G (v) t p (u )
xm s(G(v)t (u ))e j NG ( )( )( )
u 0 v 0
p xmax Gmax tmax
m 1 2v t p (u ) m 1 2v
N G ( )( 1)( ) N G ( )( 1)T (u )
NC 2 NG t max NC 2 NG
j 2vmT(u) / NC jvT (u) jmT (u) / NT jNGT (u) / 2
Department of Biomedical Engineering, Northeastern University
21. Method
DFT Reconstruction for m-SPRITE MRI Data
Numerous complex computation
Cannot be efficiently implemented by FFT algorithms
Require revised FFT or interpolation methods for reconstruction
N c 1
s(k ) x e j 2xk
dx x m s (k
n 0
n )e 2jkn xm
NT 1N G 1 xm G (v) t p (u )
xm s(G(v)t (u ))e j NG ( )( )( )
u 0 v 0
p xmax Gmax tmax
m 1 2v t p (u ) m 1 2v
N G ( )( 1)( ) N G ( )( 1)T (u )
NC 2 NG t max NC 2 NG
j 2vmT(u) / NC jvT (u) jmT (u) / NT jNGT (u) / 2
Department of Biomedical Engineering, Northeastern University
22. Method
DFT Reconstruction for m-SPRITE MRI Data
Numerous complex computation
Cannot be efficiently implemented by FFT algorithms
Require revised FFT or interpolation methods for reconstruction
N c 1
s(k ) x e j 2xk
dx x m s (k
n 0
n )e 2jkn xm
NT 1N G 1 xm G (v) t p (u )
xm s(G(v)t (u ))e j NG ( )( )( )
u 0 v 0
p xmax Gmax tmax
m 1 2v t p (u ) m 1 2v
N G ( )( 1)( ) N G ( )( 1)T (u )
NC 2 NG t max NC 2 NG
j 2vmT(u) / NC jvT (u) jmT (u) / NT jNGT (u) / 2
Department of Biomedical Engineering, Northeastern University
23. Method
CZT Reconstruction Method for m-SPRITE MRI Data
Separate one non-uniform k-space into Nt uniformly sampled k-space
Reconstruct each k-space obtained in last step and scale FOVs via CZT simultaneously
Sum all results together (i.e., signal averaging)
Spatial resolution improvement
SNR improvement
N G 1
Image (u ) CZT [ s(G(v)t p (u ), T (u )] s(G(v)t p (u))e jvT (u )e j 2vmT (u ) / N c
v 0
NT 1 NT 1N G 1
SingleImag e Image (u) s(G(v)t
u 0 u 0 v 0
p (u ))e jvT (u ) e j 2vmT (u ) / Nc
High similarity can be found, except the phase angle
Department of Biomedical Engineering, Northeastern University
24. Method
CZT Reconstruction Method for m-SPRITE MRI Data
Separate one non-uniform k-space into Nt uniformly sampled k-space
Reconstruct each k-space obtained in last step and scale FOVs via CZT simultaneously
Sum all results together (i.e., signal averaging)
Spatial resolution improvement
SNR improvement
N G 1
Image (u ) CZT [ s(G(v)t p (u ), T (u )] s(G(v)t p (u))e jvT (u )e j 2vmT (u ) / N c
v 0
NT 1 NT 1N G 1
SingleImag e Image (u) s(G(v)t
u 0 u 0 v 0
p (u ))e jvT (u ) e j 2vmT (u ) / Nc
High similarity can be found, except the phase angle
Department of Biomedical Engineering, Northeastern University
25. Method
CZT Reconstruction Method for m-SPRITE MRI Data
Separate one non-uniform k-space into Nt uniformly sampled k-space
Reconstruct each k-space obtained in last step and scale FOVs via CZT simultaneously
Sum all results together (i.e., signal averaging)
Spatial resolution improvement
SNR improvement
N G 1
Image (u ) CZT [ s(G(v)t p (u ), T (u )] s(G(v)t p (u))e jvT (u )e j 2vmT (u ) / N c
v 0
NT 1 NT 1N G 1
SingleImag e Image (u) s(G(v)t
u 0 u 0 v 0
p (u ))e jvT (u ) e j 2vmT (u ) / Nc
High similarity can be found, except the phase angle
Department of Biomedical Engineering, Northeastern University
26. Method
CZT Reconstruction Method for m-SPRITE MRI Data
Separate one non-uniform k-space into Nt uniformly sampled k-space
Reconstruct each k-space obtained in last step and scale FOVs via CZT simultaneously
Sum all results together (i.e., signal averaging)
Spatial resolution improvement
SNR improvement
N G 1
Image (u ) CZT [ s(G(v)t p (u ), T (u )] s(G(v)t p (u))e jvT (u )e j 2vmT (u ) / N c
v 0
NT 1 NT 1N G 1
SingleImag e Image (u) s(G(v)t
u 0 u 0 v 0
p (u ))e jvT (u ) e j 2vmT (u ) / Nc
High similarity can be found, except the phase angle
Department of Biomedical Engineering, Northeastern University
27. Method
CZT Reconstruction Method for m-SPRITE MRI Data
Separate one non-uniform k-space into Nt uniformly sampled k-space
Reconstruct each k-space obtained in last step and scale FOVs via CZT simultaneously
Phase Correction
Sum all results together (i.e., signal averaging)
Spatial resolution improvement
SNR improvement
NT 1 NT 1N G 1
SingleImag e Image (u) s(G(v)t
u 0 u 0 v 0
p (u ))e jvT (u ) e j 2vmT (u ) / Nc
NT 1
xm CZT [ s(G(v)t p (u ), T (u )]e j (m )
u 0
NT 1 N G 1
s(G(v)t p (u))e jvT (u ) e j 2vmT (u ) / N e j (m )
c
u 0 v 0
T (u ) N G T (u )
NT 2
Department of Biomedical Engineering, Northeastern University
28. Method
CZT Reconstruction Method for m-SPRITE MRI Data
Separate one non-uniform k-space into Nt uniformly sampled k-space
Reconstruct each k-space obtained in last step and scale FOVs via CZT simultaneously
Phase Correction
Sum all results together (i.e., signal averaging)
Spatial resolution improvement
SNR improvement
NT 1 NT 1N G 1
SingleImag e Image (u) s(G(v)t
u 0 u 0 v 0
p (u ))e jvT (u ) e j 2vmT (u ) / Nc
NT 1
xm CZT [ s(G(v)t p (u ), T (u )]e j (m )
u 0
NT 1 N G 1
s(G(v)t p (u))e jvT (u ) e j 2vmT (u ) / N e j (m )
c
u 0 v 0
T (u ) N G T (u )
NT 2
Department of Biomedical Engineering, Northeastern University
29. Method
Image Scaling through CZT
The length of resultant signal can be set to any value for different practical applications
Can be implemented by simply set the parameter K in accordance with the scaling factor
N 1
X (k ) x ( n) z k n zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
30. Method
Image Scaling through CZT
The length of resultant signal can be set to any value for different practical applications
Can be implemented by simply set the parameter K in accordance with the scaling factor
N 1
X (k ) x ( n) z k n zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
31. Method
Image Scaling through CZT
The length of resultant signal can be set to any value for different practical applications
Can be implemented by simply set the parameter K in accordance with the scaling factor
N 1
X (k ) x ( n) z k n zk AW k
n 0
A A0 e j0
W W0e j0
Fig. 3 Unit Circle on the z-Plane
Department of Biomedical Engineering, Northeastern University
32. Result
Experiments and Parameters
Original MRI data courtesy of James Rioux, University of New Brunswick, Canada
A fiber-reinforced polyester resin
Nt=25, Ng=64
FOV scaling
m-SPRITE data reconstruction
Image scaling
Department of Biomedical Engineering, Northeastern University
33. Result
Experiments and Parameters
Original MRI data courtesy of James Rioux, University of New Brunswick, Canada
A fiber-reinforced polyester resin
Nt=25, Ng=64
FOV scaling
m-SPRITE data reconstruction
Image scaling
Department of Biomedical Engineering, Northeastern University
34. Result
Experiments and Parameters
Original MRI data courtesy of James Rioux, University of New Brunswick, Canada
A fiber-reinforced polyester resin
Nt=25, Ng=64
FOV scaling
m-SPRITE data reconstruction
Image scaling
Department of Biomedical Engineering, Northeastern University
35. Result
FOV Scaling (CZT Versus Bilinear Interpolation)
NcNclog2Nc
Scaling factor 0.89
Better spatial resolution and accuracy (See Fig. 5)
Fig. 5 FOV Scaling by bilinear interpolation and CZT
Department of Biomedical Engineering, Northeastern University
36. Result
FOV Scaling (CZT Versus Bilinear Interpolation)
NcNclog2Nc
Scaling factor 0.89
Better spatial resolution and accuracy (See Fig. 5)
Fig. 5 FOV Scaling by bilinear interpolation and CZT
Department of Biomedical Engineering, Northeastern University
37. Result
FOV Scaling (CZT Versus Bilinear Interpolation)
NcNclog2Nc
Scaling factor 0.89
Better spatial resolution and accuracy (See Fig. 5)
Fig. 5 FOV Scaling by bilinear interpolation and CZT
Department of Biomedical Engineering, Northeastern University
38. Result
FOV Scaling (CZT Versus Bilinear Interpolation)
NcNclog2Nc
Scaling factor 0.89
Better spatial resolution and accuracy (See Fig. 5)
Fig. 5 FOV Scaling by bilinear interpolation and CZT
Department of Biomedical Engineering, Northeastern University
39. Result
m-SPRITE MRI Data Reconstruction
Higher SNR (See Fig. 6)
Better apparent (spatial) resolution
Higher accuracy and less computational complexity
Fig. 6 Reconstruction Results 1
Department of Biomedical Engineering, Northeastern University
40. Result
m-SPRITE MRI Data Reconstruction
Higher SNR (See Fig. 6)
Better apparent (spatial) resolution
Higher accuracy and less computational complexity
Fig. 6 Reconstruction Results 1
Department of Biomedical Engineering, Northeastern University
41. Result
m-SPRITE MRI Data Reconstruction
Higher SNR
Better apparent (spatial) resolution (See Fig. 7)
Higher accuracy and less computational complexity
Fig. 7 Reconstruction Results 2 [Rioux et al.]
Department of Biomedical Engineering, Northeastern University
42. Result
m-SPRITE MRI Data Reconstruction (CZT Versus DRS Method)
Higher SNR
Better apparent (spatial) resolution
Higher accuracy and less computational complexity (See Fig. 8)
Dutt, Rokhlin and Sarty method [Dutt et al. and Sarty et al.]
Table I Table of SNR
Nt SNR
1 10.9049
4 16.2190
9 21.7201
12 25.3492
16 28.2785
25 33.3223
Fig. 8 Running Time and Accuracy
[Rioux et al.]
Department of Biomedical Engineering, Northeastern University
43. Result
Image Scaling (Bilinear Interpolation vs FFT Zero Filling vs CZT)
Non-integer scaling factor
No significant advantages (See Fig. 9)
Table II Table of Rescaled Image Quality
FFT Zero
Standard Interpolation CZT
Filling
d 0.2451 0.4361 0.2519
r 0.0606 0.1674 0.1128
Fig. 9 Rescaled Images
Department of Biomedical Engineering, Northeastern University
44. Result
Image Scaling (Bilinear Interpolation vs FFT Zero Filling vs CZT)
Non-integer scaling factor
No significant advantages (See Fig. 9)
Table II Table of Rescaled Image Quality
FFT Zero
Standard Interpolation CZT
Filling
d 0.2451 0.4361 0.2519
r 0.0606 0.1674 0.1128
Fig. 9 Rescaled Images
Department of Biomedical Engineering, Northeastern University
45. Result
Image Scaling (Bilinear Interpolation vs FFT Zero Filling vs CZT)
Non-integer scaling factor
No significant advantages (See Fig. 9)
Table II Table of Rescaled Image Quality
FFT Zero
Standard Interpolation CZT
Filling
d 0.2451 0.4361 0.2519
r 0.0606 0.1674 0.1128
Fig. 9 Rescaled Images
Department of Biomedical Engineering, Northeastern University
46. Conclusion
Accuracy
Spatial resolution
SNR
Computational level
To be discovered
Department of Biomedical Engineering, Northeastern University
47. Conclusion
Accuracy
Spatial resolution
SNR
Computational level
To be discovered
NT 1 NT 1N G 1
SingleImag e Image (u) s(G(v)t
u 0 u 0 v 0
p (u ))e jvT (u ) e j 2vmT (u ) / Nc
NT 1
xm CZT [ s(G(v)t p (u ), T (u )]e j (m )
u 0
NT 1 N G 1
s(G(v)t p (u))e jvT (u ) e j 2vmT (u ) / N e j (m )
c
u 0 v 0
T (u ) N G T (u )
NT 2
Department of Biomedical Engineering, Northeastern University
48. Conclusion
Accuracy
Spatial resolution
SNR
Computational level
To be discovered
Fig. 7 Reconstruction Results 2 [Rioux et al.]
Department of Biomedical Engineering, Northeastern University
49. Conclusion
Accuracy
Spatial resolution
SNR
Computational level
To be discovered
Fig. 6 Reconstruction Results 1
Department of Biomedical Engineering, Northeastern University
50. Conclusion
Accuracy
Spatial resolution
SNR
Computational level
To be discovered
Fig. 8 Running Time and Accuracy [Rioux et al.]
Department of Biomedical Engineering, Northeastern University
51. Conclusion
Accuracy
Spatial resolution
SNR
Computational level
To be discovered
This study only shines very limited lights on the scenery of CZT applications on MRI research and a
majestic panorama of its applications is expected to be discovered unremittingly.
Department of Biomedical Engineering, Northeastern University