A. Patient selection and contouring
A retrospective study was performed involving 10 patients who received radiation therapy after breast-conserving surgery for left breast cancer. This study was approved by our institutional ethics committee. Plans were generated for patients who were treated in our hospital between April 2019 and March 2022. The median age of the patients was 59 years (range: 38–82). The TMN classification of the patients were as follows: T0N1M0 (n = 1), T0N1M0 (n =1), T1N2M0 (n = 1), T1N2M0 (n = 1), T2N1M0 (n = 2), T2N2M0 (n = 1), T3N2M0 (n = 1), T4bN1M0 (n = 1), T4bN3M0 (n = 1).
Computed tomography (CT, Aquilion64, Canon Medical Systems, Tochigi, Japan) scans were acquired with a 2-mm slice thickness, head-first supine position, and free breathing mode. Patients were immobilized using the VacLok system (CIVCO Medical Solutions, Kalona, IA) with the use of a wing board for arm positioning above the head. The CT data were imported into Velocity, version 4.0.1 (Varian Medical Systems, Palo Alto, CA, USA) and used to create a structure set. The clinical target volume (CTV) included the left breast and the axillary and supraclavicular lymph node areas. The planning target volume (PTV) was defined as the CTV with a 5-mm margin in all directions. The ipsilateral lung, contralateral lung, heart, contralateral breast, spinal cord, esophagus, stomach, thyroid gland, and larynx were contoured as organs at risk.
For evaluation, breast skin was defined as the area of the left breast within 5 mm of the skin surface (Fig. 1a). The clipping target was defined as the area where the PTV intersects the area of the body contour reduced inward by 3 mm. (Fig. 1b). The virtual bolus was created by expanding the PTV to 5 mm and subtracting the body contour. The density of the virtual bolus was then replaced by 1.0 g/cm3 in the TomoTherapy planning system (Fig. 1c). The virtual bolus was only used during optimization, and the final calculation was performed without the virtual bolus.
B. Treatment planning
Plans were created for the TomoHD system, version 2.1.6 (Accuray Inc., Sunnyvale, CA, USA), which has a 6 MV photon beam with a dose rate of 850 MU/min modulated with 64 binary MLC. The plans were generated in Planning Station, version 5.1.1.6 (Accuray Inc.) with a collapsed cone convolution/superposition algorithm. Dose constraints are shown in Table 1. For the virtual bolus plans, the prescribed dose was 50 Gy to 50% (D50) of the PTV and was delivered in 25 fractions. For the clipping and skin flash plans, a prescription point was set at D50 of the clipped PTV. The plan parameters included a fixed jaw of 2.5 cm, modulation factor of 2.0–2.3, and calculation grid of 1.91 × 1.91 × 2 mm3. In helical mode, a pitch of 0.287 was used, and a directional block (with only beam ejection allowed) was set on part of the bilateral lungs. In direct mode, the pitch was set to 0.250, and the beam angle was tangential to the breast, while two oblique beam angles were used for the lymph node area. When using skin flash, the target was clipped 3 mm from the skin surface, and the outer three MLC leaves (1.875 cm width) were extended (Fig. 1d).
C. Plan evaluation
TomoTherapy DQA Station version 5.1.1.5 (Accuray Inc.) was used to simulate how patient position variation impacted the delivered dose. To simulate setup error, recalculations were performed with the patients positioned at the origin (no shift) and then with the patients being displaced in the anterior or posterior directions. The DICOM-RT files of the recalculated plans were exported, and the dose-volume histograms (DVH) were analyzed using Velocity. Moreover, to compare the dose distributions and profiles between the plans when shifted 5 mm in the anterior/posterior direction and the origin position, an analysis was performed using in-house software based on Python (version 3.8.10) with NumPy (version 1.21.4), Pydicom (version 2.2.2), and Matplotlib (version 3.5.1) modules.
D. Statistical analysis
Data are presented as median ± interquartile range (IQR). All statistical analyses were performed using R, version 4.0.2 (R Foundation for Statistical Computing) [24]. The differences between the two groups were analyzed using Wilcoxon signed rank test. For three groups comparisons, the Friedman test followed by pairwise post hoc comparisons using the Wilcoxon signed rank test with Bonferroni corrections were used. Data were considered statistically significant if p < 0.05.