Prospective evaluation of AI-based BiCycle autoplanning for advanced cervical cancer brachytherapy
Rule-based AI BiCycle autoplanning compared favorably with manual planning in retrospective planning studies. In this study, we prospectively evaluated BiCycle's 'real life' impact by running it in parallel to current manual planning. Apart from dosimetric evaluations, treating physicians scored plans, and planning times were recorded. Following clinical routine, in each brachytherapy fraction, an RTT manually generated a plan which was then optionally adjusted by the treating physician, yielding the Man_Adj plan that was delivered. After treatment, a plan was automatically generated using BiCycle, followed by optional adjustment by the treating physician in the clinical TPS (Auto_Adj plan). Man_Adj and Auto_Adj plans were mutually compared and all planning time recorded. Physicians scored plan quality differences. For 28 of 37 evaluated fractions, the treating physician scored a higher quality for the Auto_Adj plan, in 7 fractions there was parity, and for 2 fractions Man_Adj was preferred. With comparable CTV For 95% of fractions, the treatment plan based on autoplanning was either preferred by the involved physician (76% of fractions) or considered similar in quality as the clinical plan (19%). Planning time reduced from 44.1 min for the manual workflow to 9.4 min. The system has recently been clinically introduced, following with the EU Medical Device Regulation.
