Author: Jacqueline M. Andreozzi, Tess Armstrong, Shiva Bhandari, John M Bryant, Jessica M Frakes, David J. Gladstone, Sarah E Hoffe, Kujtim Latifi, Arash Naghavi, Steven Nichols, Ibrahim M. Oraiqat, Russell Palm, Gage H. Redler, Stephen A Rosenberg, Joseph Weygand 👨🔬
Affiliation: Department of Radiation Oncology and Applied Science, Dartmouth Health, Thayer School of Engineering, Dartmouth College, Department of Radiation Oncology, Ohio State University, Moffitt Cancer Center, Department of Radiation Oncology, Moffitt Cancer Center 🌍
Purpose: The MR-linac provides unique opportunities for integrating advanced imaging into radiotherapy workflows, but the lower sensitivity of systems like the 0.35T model can pose challenges for diffusion-weighted imaging (DWI). To mitigate this issue, singular value decomposition (SVD) was applied to enhance signal-to-noise ratio (SNR), enabling the acquisition of higher-quality DWI images for potential clinical use.
Methods: DWI scans were obtained from rectal cancer patients treated on a 0.35 T MR-linac. The scans were acquired at multiple treatment time points using a multi-slice echo-planar readout with b-values of 0, 200, 300, 500, and 800 s/mm². To enhance the quality of these images, singular value decomposition (SVD) was applied as a denoising technique by decomposing the data matrix into singular values. Components representing the dominant signal were retained, while smaller, noise-associated components were discarded, and the images were reconstructed from the retained elements. Signal-to-noise ratio (SNR) measurements were calculated in regions of interest, including a vial with well-defined diffusion properties used for standardization, as well as muscle, bladder, and rectum. SNR values before and after applying SVD were compared across the different b-values, and averages from all treatment fractions were analyzed to assess improvements.
Results: The application of the SVD method resulted in a notable improvement in image SNR, with the enhancement becoming more pronounced at higher b-values. Specifically, the average SNR increased by 1.0%, 20.7%, 45.4%, 57.2%, and 94.9% for b-values of 0, 200, 300, 500, and 800 s/mm², respectively.
Conclusion: The application of SVD successfully enhanced the SNR of DWI images acquired on the 0.35 T MR-linac, with more pronounced improvements at higher b-values due to their lower initial sensitivity. These findings highlight the potential of SVD-based techniques to overcome the challenges of low-field MR-linac imaging, supporting the future integration of DWI into clinical radiotherapy workflows.