Session Item

In breast radiotherapy, electron boosts are often manually computed (e.g. by calculating monitor units based on target depth), which limits the analysis of relationships between dose distributions and outcomes. This study investigated the feasibility of reconstructing the dose distributions from manually-planned electron boosts in patients treated with breast-conserving radiotherapy as part of the REQUITE study (www.requite.eu).

In the REQUITE dataset, out of 198 patients treated with sequential electron boosts, only 72 had complete stored dose distributions (primary + boost). For the remaining 126, data available included: beam energy (MeV), boost prescription dose (Gy), CT and dose distribution for primary treatment. Data from 50 patients with complete stored dose distributions were used to develop and validate our dose reconstruction method.
First, 10 cases were used to determine the best parameters for Monte-Carlo based electron dose reconstruction on Raystation (V11B) considering the following: a) the CT calibration curve, b) the number of histories, and c) the resolution of the dose calculation grid. A sensitivity analysis was performed to quantify the impact of varying those parameters using a Friedman test on SPSS v.28. Then, a validation set of 40 cases were used to quantify the accuracy of the reconstructed dose distributions. Figure 1 shows the workflow for electron dose reconstruction. The similarity between the reconstructed and stored dose distributions was evaluated using 3D-gamma index (3%3mm criteria), and dose metrics (D2%, D95%, and Dmean) extracted from breast and tumour bed contours. To evaluate the location of dose difference a dose-location histogram (DLH), as implemented in A Computational Environment for Radiological Research (CERR) was used.

The electron dose distribution was most impacted by the resolution of the dose calculation grid (p<0.01), where the finest grid (0.15 cm) had higher similarity to the reference (stored) doses (see Figure 2(A)). In contrast, varying the CT calibration curve and the number of histories had a negligible influence on dose calculations. The absolute difference between the reconstructed and reference doses was less than 1 Gy for breast and tumour bed considered at D95%, D2%, and Dmean (Figure 2(A)). Results were comparable in the validation set, with > 90% gamma passing rate. However, differences were observed at the build-up region (within 1cm from the skin) and at tissue interface regions as shown in the DLH (Figure 2(B)).

In this cohort, the electron dose distributions could be reconstructed retrospectively with satisfactory accuracy. However, caution is advised when assessing dose close to the skin. Further validation is needed to confirm accuracy outside of the REQUITE dataset. This study will be beneficial for large-scale efforts to recover electron dose distributions and increase the number of complete records for retrospective analysis, even in recent data cohorts.