Bowel bag sparing in palliative radiotherapy (pRT) of rectal cancer (rCa): Tradeoff between organs at risk (OARs) and use of AI to reduce radiation oncologist (RO) workload.

Abstract

e15615 Background: We retrospectively studied the potential to reduce bowel toxicity after pRT of rCa using a bowel dose (BD) minimization strategy; also its effect on dose to remaining normal tissues (NT) and pelvic bone marrow (PBM). Finally, we studied the use of AI to reduce the RO workload. Methods: 25 consecutive rCa patients (pts) previously treated with pRT were included. Their treatments were planned without consideration to BD. The following OARs were manually segmented retrospectively: Bowel Bag (BB-M) excluding PTV, PBM (PBM-M), and NT defined as the patient contour excluding PTV and BB-M. The AI-Rad Companion software was used to autosegment: Bowel Loops (BL-AI) and PBM (PBM-AI). For the 25 pts, two Pareto optimal treatment plans (5 Gy x 5) were generated with the Monaco treatment planning system. The first plan (Reference) was optimized aiming to limit dose outside the PTV. The second plan (Opt-BB-M) was optimized aiming firstly to limit BB-M dose and secondly to limit NT dose. A third plan (Opt-BL-AI) was generated for the 10 worst-off pts (in terms of BD) aiming instead to limit BL-AI dose. Two volume-at-dose (VD) metrics (V18Gy, V10Gy) were used as endpoints. Endpoint differences (ΔVD-OAR) were tested using one-sided Wilcoxon signed-rank tests. Spearman (rs) and Pearson (r) correlation coefficients were used to study correlation between variables (ΔVD-BB vs. ΔVD-BM, and ΔVD-BB vs. ΔVD-NT). Dice Similarity Coefficient (DCS) was used to compare PBM-M and PBM-AI. Results: After reoptimization, BB-M VD (V18Gy; V10Gy) was lower (p < 0.001; p < 0.001), PBM-M VD was higher (p = 0.002; p < 0.001), and NT VD was higher (p < 0.001; p < 0.001). For BB-M, ΔV18Gy and ΔV10Gy (median, min, max) were (-4, -46, 0) cm3 and (-33, -254, 0) cm3; for PBM-M: (+2, -4, +56) cm3 and (+33, -6, +189) cm3; for NT: (+10, -43, +194) cm3 and (+146, -18, +614) cm3. Table 1 presents results for the 10 worst-off pts. Medium/strong correlations were found for ΔV18Gy-BB vs. ΔV18Gy-PBM (rs = -0.72, p < 0.001; r = -0.74, p < 0.001), ΔV10Gy-BB vs. ΔV10Gy-PBM (rs = -0.66, p < 0.001; r = -0.32, p = 0.11), ΔV18Gy-BB vs. ΔV18Gy-NT (rs = -0.74, p < 0.001; r = -0.74, p < 0.001), ΔV10Gy-BB vs. ΔV10Gy-NT (rs = -0.74, p < 0.001; r = -0.51, p = 0.009). The DSC (median, min, max) for PBL was (92%, 87%, 94%). Conclusions: Bowel dose in pRT of rCa can be reduced at the expense of increasing dose to PBM and other NT. AI tools can be used effectively to limit RO workload. [Table: see text]