Purpose: Intramuscular metastases are difficult to distinguish from surrounding normal muscle tissues when imaged using CT techniques. This makes it challenging to set up accurately for treatment using conventional linear accelerators, especially when SBRT technique is used. The goal of this retrospective study is to quantify the inter-fractional setup uncertainty and inter-observer variation from CT-based registration for treatment, using the MR-based registration as the ground truth.
Methods: Three patients with intramuscular metastases treated with definitive radiotherapy using 1.5 Tesla MR-linac with daily MR-guidance were evaluated. These patients were treated with 30Gy, 40Gy and 60Gy in 3, 5 and 15 fractions, respectively. To simulate CT-based setup, we used an in-house cycle-consistent generative adversarial network model to generate synthetic-CT images from each treatment fraction’s MR image set. Each synthetic-CT image set was aligned to the simulation-CT by six physicists independently using translations only. The difference between the synthetic-CT-based alignment and the MR-based alignment was calculated to compute the inter-fractional setup uncertainty. Inter-observer variation was determined by taking the average of the standard deviations of each fraction’s registration differences among all six observers.
Results: Each physicist performed 23 synthetic-CT-based image alignment with simulation-CT. The average of absolute inter-fractional setup uncertainties for all the patients were 0.18cm, 0.24cm, 0.34cm, and standard deviations were 0.15cm, 0.15cm, 0.27cm in the LR, AP, and SI directions, respectively. The largest inter-fractional setup uncertainties were 0.69cm, 0.66cm, and 1.32cm in the LR, AP, and SI directions, respectively. The mean inter-observer variations were 0.15cm, 0.09cm, 0.14cm, in the LR, AP, and SI directions, respectively.
Conclusion: Substantial setup uncertainties were observed for CT-based alignment when treating intramuscular metastases with hypofractionation/SBRT techniques. Inter-observer variation was smaller compared to inter-fractional set up uncertainties. Our preliminary data suggest that without daily MR guidance, use of larger margin may be necessary to ensure target coverage.