Objective. Radiotherapy of left-sided breast cancer in deep inspiration breath-hold (DIBH) reduces the heart dose. Surface guided radiotherapy (SGRT) can guide the DIBH, but the accuracy is subject to variations in the chest wall position relative to the patient surface. Approach. In this study, ten left-sided breast cancer patients received DIBH radiotherapy with tangential fields in 15–18 fractions. After initial SGRT setup in free breathing an orthogonal MV/kV image pair was acquired during SGRT-guided breath-hold. The couch was corrected to align the chest wall during another breath-hold, and a new SGRT reference surface was acquired for the gating. The chest wall position error during treatment was determined from continuous cine MV images in the imager direction perpendicular to the cranio-caudal direction. A treatment error budget was made with individual contributions from the online registration of the setup MV image, the difference in breath-hold level between setup imaging and SGRT reference surface acquisition, the SGRT level during treatment, and intra-fraction shifts of the chest wall relative to the SGRT reference surface. In addition to the original setup protocol (Scenario A), SGRT was also simulated with better integration of image-guidance by capturing either the new reference surface (Scenario B) or the SGRT positional signal (Scenario C) simultaneously with the setup MV image, and accounting for the image-guided couch correction by shifting the SGRT reference surface digitally. Main results. In general, the external SGRT signal correlated well with the internal chest wall position error (correlation coefficient >0.7 for 75% of field deliveries), but external-to-internal target position offsets above 2 mm occasionally occurred (13% of fractions). The PTV margin required to account for the treatment error was 3.5 mm (Scenario A), 3.4 mm (B), and 3.1 mm (C). Significance. Further integration of SGRT with image-guidance may improve treatment accuracy and workflow although the current study did not show large accuracy improvements of scenario B and C compared to scenario A.
Read full abstract