Ping N Wang1, Julia V Velikina2, Leah C Henze Bancroft2, Alexey A Samsonov2, Frederick Felcz2, Roberta M Strigel1,2,3, and James H Holmes2
1Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, 2Department of Radiology, University of Wisconsin-Madison, Madison, WI, United States, 3Carbone Cancer Center, University of Wisconsin-Madison, Madison, WI, United States
Synopsis
Radial acquisition with MOCCO reconstruction has been
proposed for high spatial and temporal resolution breast DCE imaging. However,
the performance of the method has not been evaluated across a wide range of
temporal enhancement curves. In this work, we use a breast digital reference
object with the ability to provide a wide range of lesion contrast kinetics
using the extended Tofts
model for pharmacokinetic simulation. A voxel-by-voxel quantitative analysis
was applied and compared with the ground truth.
Introduction
Dynamic
contrast-enhanced (DCE) MRI provides the most reliable method for following a
known breast cancer to confirm its response to therapy due to its ability to
reveal the contrast difference between tumors and normal tissues. Because of
this property, DCE-MRI has been investigated as a powerful tool for treatment
planning1, screening2 and treatment response assessment3. The lesion curves measured from conventional
DCE-MRI with high spatial resolution but low temporal resolution imaging
protocols can provide only moderate specificity to characterize different
lesion types4. Quantitative pharmacokinetic (PK) modeling of
DCE-MRI has shown promise to provide additional non-invasive evidence for
differentiating benign and malignant lesions5. However, the current breast DCE-MRI strategies
are known to be insufficient to fully capture rapidly changing contrast
kinetics at high spatial resolution during bi-lateral assessment. Studies have
shown the potential of high spatial-temporal resolution breast MR imaging by
using stack-of-stars radial acquisition with compressed sensing (CS)
reconstruction6. Previous work has shown that using a
data-driven low-rank based CS method (MOCCO) demonstrated improved temporal
fidelity when matching spatial resolution and coverage from routine clinical
protocols as compared with CS with total variation7. However, there is a need to validate the
accuracy and robustness of PK parameter estimation using advanced acquisition
and/or reconstruction if these approaches are to be used in clinical practice. Therefore,
the purpose of this study is to assess the performance of radial acquisition
with MOCCO reconstruction for PK assessment across a wide range of parameter values
in the setting of a digital reference object (DRO) for breast DCE MRI.Methods
Simulations:
Time-resolved radial data was simulated using a DRO to model an acquisition
using a 16-channel breast coil with bilateral axial breast MRI protocol (TR/TE=
5.5/2.4 ms, FOV= 340 x 340 mm, 448x448 in-plane matrix, 142 z-phase encodes)
(Fig. 1). Homogeneously enhancing round lesions were simulated with different PK
parameters to generate slow, intermediate, and fast changing lesion kinetics.
Specific values of Ktrans, Ve and the corresponding curve
shapes are listed in Table 1.
Reconstruction:
The low-rank data-driven temporal model for MOCCO8,9 was obtained from the low
frequency region of the fully-sampled central k-space data using progressive
learning with cubic spline approximation9,10 followed by complex
independent component analysis11. The undersampled data were
reconstructed at 5 s temporal resolution, corresponding to 8 projections/frame
(undersampling factor R=88) using MOCCO.
Analysis:
PK modeling was performed by fitting the curves to the extended Tofts model (ETM)12,13 using an in-house implementation
of the Levenberg-Marquaradt algorithm to assess the ability to fit and recover
specific kinetic parameters.Results
In the case of slow, and intermediate lesion kinetic curves
(Figure 2, A-C), the lesion kinetics can be accurately recovered using MOCCO
reconstruction. In the case of fast lesion kinetics (Figure 2, D-H), a decrease
in the wash-in slope was observed when the Ktrans value
increased. Further, the MOCCO reconstruction showed underestimation of the maximum
peak intensity when Ktrans was
larger than 1 min-1 (Figure 2, F-H). Figure 3 presents both the mean
and standard deviation (SD) of the % difference between fitted parameters and
ground truth. The accuracy and precision of the fitted parameters are well
within ±15% for slow, intermediate, and fast lesion kinetics with Ktrans < 1
min-1. For fast kinetic curves with Ktrans > 1
min-1, the underestimation of Ktrans and Ve were
increased because the larger Ktrans value yields rapidly changing curve shapes. Although MOCCO reconstruction showed high
underestimation at fast lesion kinetic curves, the measurements were still
highly correlated (r = 0.995)
(Figure 3, B). Figure 4 displays color maps of the estimated and the associated
errors expressed in % difference to known truth values, where the greatest error most frequently
occurred for voxels on the edge of the lesion.Discussion and conclusions
In this study, we evaluated the robustness of the MOCCO reconstruction to recover different
DCE curve shapes in the setting of a DRO with the ETM. A 5 second temporal
resolution and 0.75mm x 0.75mm spatial resolution was demonstrated and is
expected to be well suited to the demands of bilateral breast DCE-MRI. The
simulated results indicated that MOCCO reconstruction ensures
less than 15% error in Ktrans and Ve when Ktrans < 1 min-1 and Ve = 0.3, which is within the reasonable physiological ranges. Underestimation of Ktrans was found in fast lesion contrast curves, but MOCCO reconstruction still showed the ability to capture rapidly enhancing lesion characteristics. However, some evidence of spatial-temporal blurring
was observed in the voxels at junctions between high-intensity lesion and
low-contrast tissue, which introduced the estimation error into the PK parameters and
will be the focus of future work. Although the ETM was used in this study due
to its wide usage in the literature, alternative PK models can readily be used
in the simulation pipeline.Acknowledgements
The
authors wish to acknowledge support from the following NIH grants: R21EB018483,
R01EB027087, P30CA014520, and R01CA248192. As well as support from GE
Healthcare, the RSNA Research and Education Foundation, and a Research and
Development Grant from the Departments of Radiology and Medical Physics, University of
Wisconsin-Madison.References
- Garimella V, Qutob O, Fox JN, Long ED, Chaturvedi A, Turnbull LW,
Drew PJ. Recurrence rates after DCE-MRI image guided planning for
breast-conserving surgery following neoadjuvant chemotherapy for locally
advanced breast cancer patients. European Journal of Surgical Oncology (EJSO).
2007;33(2):157–161. doi:10.1016/j.ejso.2006.09.019
- Mann RM, Mus RD, van Zelst J, Geppert
C, Karssemeijer N, Platel B. A Novel Approach to Contrast-Enhanced Breast
Magnetic Resonance Imaging for Screening: High-Resolution Ultrafast Dynamic
Imaging. Investigative Radiology. 2014;49(9):579–585. doi:10.1097/RLI.0000000000000057
- Wang C-H, Yin F-F, Horton J, Chang Z.
Review of treatment assessment using DCE-MRI in breast cancer radiation
therapy. World Journal of Methodology. 2014;4(2):46–58.
doi:10.5662/wjm.v4.i2.46
- Kuhl CK, Schild HH, Morakkabati N. Dynamic
Bilateral Contrast-enhanced MR Imaging of the Breast: Trade-off between Spatial
and Temporal Resolution. Radiology. 2005;236(3):789–800.
doi:10.1148/radiol.2363040811
- Furman-Haran E, Schechtman E, Kelcz F,
Kirshenbaum K, Degani H. Magnetic resonance imaging reveals functional
diversity of the vasculature in benign and malignant breast lesions. Cancer.
2005;104(4):708–718. doi:10.1002/cncr.21225
- Kim SG, Feng L, Grimm R, Freed M,
Block KT, Sodickson DK, Moy L, Otazo R. Influence of temporal regularization
and radial undersampling factor on compressed sensing reconstruction in dynamic
contrast enhanced MRI of the breast: Temporal Regularization and Radial
Undersampling Effects on DCE-MRI. Journal of Magnetic Resonance Imaging.
2016;43(1):261–269. doi:10.1002/jmri.24961
- Wang PN, Velikina JV, Strigel RM, Henze Bancroft LC, Samsonov AA, Cashen TA, Wang K, Kelcz F, Johnson KM, Korosec FR, Ersoz A, Holmes JH. Comparison of data-driven and general temporal constraints on compressed sensing for breast DCE MRI. Magn Reson Med. 2020 Dec 11. doi: 10.1002/mrm.28628. Epub ahead of print. PMID: 33306217.
- Velikina JV, Samsonov AA.
Reconstruction of dynamic image series from undersampled MRI data using data-driven
model consistency condition (MOCCO). Magnetic Resonance in Medicine.
2015;74(5):1279–1290. doi:10.1002/mrm.25513
- Velikina J, Alexander A, Salmons J,
Raimy E, Purnell T, Kecskemeti S, Samsonov A. Ultrafast Speech Imaging at High
Spatial Resolution using Model-Consistency Condition Reconstruction with
Progressive Temporal Basis Learning. Proceedings of the 26th ISMRM Scientific
Meeting 2018. 2018 Paris France.
https://cds.ismrm.org/protected/18MPresentations/abstracts/0245.html
- Boor C de. A Practical Guide to
Splines. In: Applied Mathematical Sciences. 1978. doi:10.1007/978-1-4612-6333-3
- Novey M, Adali T. On Extending the
Complex FastICA Algorithm to Noncircular Sources. IEEE Transactions on Signal
Processing. 2008;56(5):2148–2154. doi:10.1109/TSP.2007.911278
- Tofts PS. Modeling tracer kinetics in
dynamic Gd-DTPA MR imaging. Journal of Magnetic Resonance Imaging.
1997;7(1):91–101. doi:https://doi.org/10.1002/jmri.1880070113
- Sourbron SP, Buckley DL. On the scope
and interpretation of the Tofts models for DCE-MRI. Magnetic Resonance in
Medicine. 2011;66(3):735–745. doi:10.1002/mrm.22861