Enhancing radiotherapy planning skills through structured deliberate practice and artificial intelligence integration: A pilot study.

Abstract

9019 Background: In many cancer treatment protocols, radiotherapy plays a pivotal role in achieving disease control. A key component of radiotherapy planning is contour delineation, which involves accurately outlining tumors and nearby structures to target radiation effectively and minimize toxicity. However, formal training in this skill often falls short, with surveys indicating up to 80% of learners and practitioners calling for a need for more robust educational methods (Leung et al. J Med Imaging Radiat Oncol 2019). While advances in artificial intelligence (AI) can improve contouring speed and consistency, radiation oncologists must also develop the skills to evaluate AI-generated results to maintain high-quality care critically. This pilot study aimed to determine whether a structured training approach that incorporates deliberate practice, personalized feedback, and AI integration could improve the contour delineation proficiency of radiation oncology trainees and practitioners. Methods: A baseline survey was conducted to identify existing gaps in contour training. Participants practiced with four anonymized thoracic imaging standardized datasets offline. Each participant’s initial contours were compared against expert consensus using the Dice similarity coefficient (DSC), a standard metric for spatial overlap. Slice-by-slice visual comparisons and self-reported confidence ratings provided additional qualitative feedback. Over six months, participants engaged in structured lessons, repeated practice sessions (including real clinical cases), expert mentoring, and AI contour assessments. Results: At baseline, 24 heart contours were consistently accurate (DSC > 0.9), while tumour and oesophagus delineation showed wide variability (DSC 0.4–0.9 and 0.2–0.9, respectively). Over time, learners demonstrated measurable improvement. By the final assessment, variability in tumour and oesophagus contours decreased (DSC 0.6–0.9 and 0.7–0.9, respectively), while heart contours remained consistently accurate. Learners also demonstrated improved confidence in both manual and AI-augmented contour delineation. Conclusions: This study suggests that a structured training approach incorporating a deliberate practice curriculum augmented by personalized feedback and AI integration can improve contour delineation skills. Future studies with larger datasets and diverse learner populations could further validate this approach, ultimately aiming to improve patient safety and treatment outcomes across the broader oncology landscape.

Ähnliche Arbeiten