Motion Compensation using Principal Component Analysis and Projection onto Dipole Fields for Abdominal Magnetic Resonance Thermometry during High-Intensity Focused Ultrasound
Jeremy Tan1,2,3, Adam C. Waspe1,2, Charles Mougenot4, Kullervo Hynynen1,5, James M. Drake1,2, and Samuel Pichardo3,6

1University of Toronto, Toronto, ON, Canada, 2Hospital for Sick Children, Toronto, ON, Canada, 3Thunder Bay Regional Research Institute, Thunder Bay, ON, Canada, 4Philips Healthcare, Toronto, ON, Canada, 5Sunnybrook Research Institute, Toronto, ON, Canada, 6Electrical Engineering, Lakehead University, Thunder Bay, ON, Canada

Synopsis

Accurate thermometry during abdominal high-intensity focused ultrasound is severely compromised by motion and susceptibility artifacts. A hybrid artifact correction method has been developed using principal component analysis as a multi-baseline method and projection onto dipole fields as a near-referenceless approach. The hybrid algorithm was tested using free-breathing porcine and human subjects and achieved an average temperature stability and precision of 0.31 (±0.22) °C and 1.18 (±0.94) °C, respectively in the kidney.

Purpose

Motion and susceptibility changes, caused by respiration (periodic) and peristalsis (aperiodic), create intense MR thermometry artifacts in the abdomen. This novel hybrid method aims to remove both periodic and aperiodic artifacts by combining principal component analysis (PCA) and projection onto dipole fields (PDF). PCA has previously been used for thermometry motion correction, based on the assumption that organ motion produces an equivalent shift in both magnitude and phase images.1 However, this premise is violated when artifacts appear in stationary tissue, due to susceptibility changes outside of the field of view. In this hybrid method, PCA serves as a multi-baseline method and corrects both motion- and susceptibility-based periodic artifacts. The implementation is similar to previous work in facial recognition2 and does not require any motion tracking tools or navigator echoes. Aperiodic motion is handled by the PDF algorithm, originally created for background phase removal in quantitative susceptibility mapping3. It functions as a near-referenceless method, correcting aperiodic artifacts induced by susceptibility changes at air-tissue interfaces.

Methods

The hybrid method incorporates PCA and PDF. PCA: During a learning step, an atlas of eigenimages is computed based on images acquired in a pre-heating period. Newly acquired images are then projected onto the subspace spanned by the eigenimages and a reference image is generated as a sum of these vectors. This PCA reference image is then employed by the PDF method to correct for aperiodic artifacts. PDF: All phase patterns can be generated as a dipole response to either i) local susceptibility sources in tissue or ii) far-reaching susceptibility differences at air-tissue interfaces. By projecting the complete dipole response (i and ii) onto a subspace which spans only the dipole response from the surroundings (ii), the local susceptibility sources can be isolated from the fluctuating external sources. This allows for accurate thermometry inside the tissue regardless of external influences. Evaluation: Non-heating experiments were performed to acquire thermometry data of in vivo kidneys in free-breathing pigs and human volunteers. Data was collected on a Philips 3T Achieva scanner (FOV: 300 x 300 mm2, voxel size = 1.34 mm, slice thickness = 11 mm, TE/TR = 16/26 ms, flip angle = 20°, acquisition matrix = 200 x 198, reconstruction matrix = 224, ETL = 9, NEX = 1, dynamic time = 0.58 s). Data from an in vivo porcine head and neck hyperthermia experiment4 was also processed (Philips 3T Achieva scanner, FOV: 400 x 400 mm2, voxel size = 2.08 mm, slice thickness = 7 mm, TE/TR = 16/45 ms, flip angle = 18°, acquisition matrix = 192 x 191, reconstruction matrix = 192, ETL = 11, NEX = 1, dynamic time = 0.88 s).

Results

Thermometry data of in vivo kidneys (Figure 1) was processed with the PCA-PDF algorithm and a standard subtraction method for comparison. Figure 2 displays progressive correction of both periodic and aperiodic artifacts. Temperature stability (temporal standard deviation of spatial average) and temperature precision (temporal average of spatial standard deviation), measured in the upper portion of the kidney can be seen in Figure 3 for both methods. The PCA-PDF method improves average temperature stability and precision by roughly an order of magnitude as compared with subtraction. Temperature stability is improved from 3.33 (±1.45) ºC to 0.31 (±0.22) ºC and temperature precision is improved from 11.53 (±13.68) ºC to 1.18 (±0.94) ºC. Figure 4 shows preservation of heat-induced phase change in the hyperthermia data. Also shown is a comparison with a multi-baseline method4 which uses echo-navigator data for classification. PCA-PDF corrects fluctuating artifacts and improves stability beyond the performance of the navigator echo assisted multi-baseline method4, without collecting any supplementary data.

Conclusion

The PCA-PDF hybrid method demonstrates effective motion correction across a variety of artifacts in free-breathing specimens. The low standard deviation of the PCA-PDF temperature stability and precision illustrates the algorithm's robust correction of complex artifacts in each subject. The algorithm also maintains high-fidelity temperature reporting and improved stability compared to methods that rely on navigator echoes. Ongoing efforts are directed at improving the performance of PCA-PDF at tissue borders and improving discrimination between overlapping heat and artifacts.

Acknowledgements

Authors acknowledge financial support from the Canadian Institutes of Health Research, the Discovery Program of the Natural Sciences and Engineering Research Council of Canada, and the Federal Economic Development Agency for Southern Ontario. CM is an employee of Philips.

References

1. B. D. de Senneville et al. MR-guided thermotherapy of abdominal organs using a robust PCA-based motion descriptor. IEEE Trans. Med. Imaging. 2011;30(11):1987–1995.
2. M. Turk and A. Pentland. Eigenfaces for Recognition. Journal of Cognitive Neuroscience. 1991;3(1):71–86.
3. T. Liu et al. A novel background field removal method for MRI using projection onto dipole fields (PDF). NMR Biomed. 2011;24(9):1129–1136.
4. S. Pichardo et al. In vivo optimisation study for multi-baseline MR-based thermometry in the context of hyperthermia using MR-guided high intensity focused ultrasound for head and neck applications. Int. J. Hyperth. 2014;30(8):579–592.

Figures

Figure 1: Magnitude image of volunteer abdomen. Yellow circle indicates region of interest in the kidney used for evaluation of temperature stability and precision.

Figure 2: Thermometry in volunteer abdomen. Standard subtraction (left) yields both periodic (black arrow) and aperiodic (white arrows) temperature artifacts which appear red and blue in these thermal maps. PCA (center) corrects periodic artifacts and subsequent use of PDF (right) handles the aperiodic artifacts.

Figure 3: Temperature stability (left) and precision (right) in the upper portion of the kidney for three pigs and two human volunteers. PCA-PDF achieves substantial improvement and consistently low results across all subjects.

Figure 4: Hyperthermia, in vivo porcine head-and-neck region. Subtraction (blue) exhibits considerable fluctuation due to artifacts. Stability is visibly improved by PCA-PDF (left-orange), even when compared to the navigator echo assisted multi-baseline method (right-red)4. Note that heating is stopped at dynamic 650; PCA-PDF corrects trailing artifact and shows smooth cool-down.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0818