A multicenter end‐to‐end postal audit of skin markerless surface‐guided radiation therapy
Masahide Saito, Ryohei Yamauchi, Hironori Kojima, Yoshinobu Shimohigashi, Takeshi Ono, Yoshifumi Oku, Shinobu Kumagai, Naoki Tohyama, Masahiko Kurooka, Eriko Saito, Masataka Sakamoto, Yusuke Ueshima, Takayuki Kanai, Naoki Hayashi, Tatsunori Saito, Tomoki Kitagawa, Makoto Sasaki, Hiroki Katayama, Hidekazu Suzuki, Chie Kurokawa, Hiroshi OnishiAbstract
Background
Surface‐guided radiation therapy (SGRT) enables skin markerless positioning, but its accuracy may depend on the reference surface, hardware configuration, commissioning, and institutional workflow. Two reference surfaces are commonly used: a DICOM reference generated from planning CT images and an SGRT reference acquired in the treatment room. Multicenter end‐to‐end data across commercial systems remain limited.
Purpose
To evaluate static irradiation accuracy and setup time with two reference surfaces in a nationwide multicenter skin markerless SGRT postal audit.
Methods
A prospective end‐to‐end postal audit was conducted from October 2023 to July 2025 at 40 Japanese institutions using five commercial SGRT systems (AlignRT, Catalyst, ExacTrac Dynamic, IDENTIFY, and VOXELAN), eight per system. Each institution received a common phantom package: a breast‐shaped surface phantom, two cubic phantoms containing gold markers, and radiochromic films. Markerless setup and irradiation were performed with both the DICOM and SGRT references; ExacTrac Dynamic institutions performed the DICOM reference setup only. Static irradiation accuracy was the difference between the film‐derived radiation center and the intended center in the DICOM RT Plan; the primary metric was the three‐dimensional translational vector error (3D vector error). Paired comparisons used the Wilcoxon signed‐rank test, and multivariable linear regression evaluated factors associated with both endpoints.
Results
Critical setup errors exceeding 5 mm occurred at four institutions initially (institution‐level error rate, 10.0%; three with the DICOM and one with the SGRT reference); one DICOM reference setup was not remeasured and remained out of tolerance, and was excluded as an outlier, leaving 39 DICOM and 31 SGRT reference setups for analysis. Excluding ExacTrac Dynamic, the DICOM reference showed a significantly larger 3D vector error than the SGRT reference (1.54 ± 0.91 vs. 1.20 ± 0.61 mm, p = 0.043). Setup time did not differ between references (86.19 ± 43.57 vs. 81.61 ± 37.09 s, p = 0.804). Axis‐specifically, the difference between references was confined to the anterior–posterior direction (mean absolute difference 0.36 mm, p = 0.002), whereas a caudal superior–inferior offset was present with both (−0.50 ± 1.29 and −0.37 ± 0.87 mm; both p < 0.01) and did not differ between them ( p = 0.854). The 1‐camera condition was independently associated with a larger 3D vector error than the 3‐camera condition (β = 0.90 mm; 95% CI, 0.17–1.63 mm; p = 0.017); no factor was associated with setup time, which showed no correlation with 3D vector error.
Conclusions
Skin markerless SGRT achieved millimeter‐level static irradiation accuracy but carried a non‐negligible risk of critical setup errors. The difference between the two reference workflows was confined to the anterior–posterior direction, whereas a superior–inferior offset common to both indicates a residual uncertainty in the audit chain itself. Reference surface selection and camera configuration influenced accuracy, whereas setup time was not a reliable surrogate. These findings highlight the importance of commissioning, standardized reference surface acquisition, routine QA, and independent third‐party audits.