Feasibility of adaptive SBRT of prostate cancer: Investigating uncertainties in AI-driven and CBCT-guided online adaptive radiotherapy

Sammanfattning: Purpose: Recent developments in external radiotherapy and its utilization with artificial in- telligence (AI) enables advanced treatment plans for a range of different disease sites. By adapting to the anatomy of the day, target coverage can be ensured while sparing more healthy tissue. The purpose of this project was to study the feasibility and benefits of daily cone beam computed tomography (CBCT)-based online adaptive radiotherapy (oART) for localised prostate cancer using the Varian Ethos™ treatment planning system (TPS). The possibility of increasing dose per fraction, while sparing organs at risk (OAR), was investi- gated. Materials and method: Online adaptive stereotactic body radiotherapy (SBRT) (5 x 7.25 Gy/fr) was simulated in a pre-clinical release of the Ethos TPS for 10 prostate cancer pa- tients using retrospective data. The system used AI-generated influencers for structure- guided or elastic deformation of targets and OAR from the reference situation on the plan- ning computed tomography (CT) image to the online acquired CBCT images. For a lo- calised prostate cancer treatment, the influencers used were bladder, rectum, and prostate, as defined on CT, while the prostate target was defined on magnetic resonance (MR) images. Based on the propagated target, the system generated two plans: scheduled and adapted. The scheduled plan was re-calculated based on the anatomy of the day, whilst the adapted plan was both re-optimised. The influencer editing required to achieve accurate target prop- agation was evaluated. The absorbed dose to clinical target volume (CTV), planning target volume (PTV), and rectum for the scheduled and adapted plans was compared. A 5 mm isotropic CTV-PTV margin was used. Results: The AI could propagate an MR-defined prostate target based on a CT-defined prostate influencer. However, of all propagated targets, 69.4% were larger in volume than the target on the planning CT (reference target). The propagated target position with re- spect to the reference target was satisfactory, where only 10% of all propagated targets were extending outside the reference PTV. The average absolute difference in position and its standard deviation were: 1.19 1.14 mm in the sagittal plane, 1.90 1.66 mm in frontal, and 1.06 1.26 mm in transversal planes. A statistically significant difference was seen between scheduled and adapted plans (n=49) in the absorbed dose to 99% of CTV volume (p=0.00) and to 99% of PTV volume (p=0.00). There was no significant difference in the maximum absorbed dose to the rectum (p=0.36) , in the rectum volume that received 28 Gy (p=0.67) and the rectum volume that received 32 Gy (p=0.10). Conclusions: The Varian Ethos TPS was observed to deform an MR-defined prostate target when CT-defined prostate was used as influencer. The propagated target volume differed from the reference target, but its position was accurate. The system could be advantageous for daily online adapted SBRT prostate treatments with sufficient CTV and PTV coverage. The reduction of toxicity to rectum needs further investigation.