Defense Date

2026

Document Type

Thesis

Degree Name

Master of Science

Department

Medical Physics

First Advisor

William Song

Second Advisor

Tianjun Ma

Third Advisor

Siyong Kim

Abstract

Problems: Adaptive radiotherapy requires clinically practical methods for responding to patient-anatomy changes that occur between the original planning images and the day of treatment delivery. Registering a previously acquired computed tomography scan to a newly acquired magnetic resonance image can introduce cross-modality uncertainty and may delay replanning. This thesis evaluates whether synthetic computed tomography generated from same-day magnetic resonance imaging can reduce registration-related uncertainty while supporting rapid dose recalculation for fraction-level adaptation.

Procedure: Magnetic resonance imaging DICOM data were exported from the ViewRay MRIdian treatment planning system and uploaded to MVision for synthetic computed tomography generation. The synthetic computed tomography data were downloaded, returned to the treatment planning system, registered within the clinical workflow, and used for dose calculation. Image and dose outputs were then reviewed in MIM for visual comparison and analyzed with a validation application together with MATLAB/CERR, Python, and ROOT pipelines. The validation framework examined dose agreement through gamma analysis, dose-volume histogram comparisons, and structure-based endpoints such as D95, and examined image agreement through Hounsfield-unit mean absolute error, pixel-wise fidelity, structural similarity index measure, difference signal-to-noise ratio, relative electron density drift, and radiomics deltas.

Results: Across the evaluated cases, dose agreement was high and the gamma analysis passed. Hounsfield-unit mean absolute error was below 50 HU in many cases, structural similarity was close to 1, and relative electron density drift was low, with some cases below 0.02. Radiomic comparisons showed close agreement for several features, including density anchors, intensity dispersion, and local correlation. Larger differences were observed for selected features, including edge gradient and GLSZM regional structure, and one patient behaved as a notable outlier relative to the broader cohort.

Conclusions: The results support the use of MVision synthetic computed tomography as a practical substitute for registered planning computed tomography in many adaptive magnetic resonance-guided radiotherapy scenarios, particularly for dose evaluation and organ-at-risk review. This approach has the potential to reduce uncertainty associated with computed tomography to magnetic resonance registration and to shorten replanning from days to minutes. Continued validation is still required for outlier patients and for radiomic features that showed larger deviations.

Rights

© The Author

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission

5-7-2026

Share

COinS