A prospective multicenter trial of deep learning auto-segmentation for organs at risk in thoracic radiotherapy - PubMed
5 hours ago
- #Radiotherapy
- #Deep Learning
- #Clinical Trial
- A prospective, multicenter trial tested the iCurveE deep learning model for AI-assisted delineation of organs at risk in thoracic and breast cancer radiotherapy, involving 500 patients across five centers and 37 physicians.
- The study compared manual, AI-generated, and AI-assisted methods for contouring 11 thoracic OARs, measuring primary endpoints like volumetric Dice similarity coefficient (vDSC) and contouring time, along with secondary metrics such as 95% Hausdorff Distance (HD95).
- AI-assisted delineation outperformed manual methods with a higher mean vDSC (0.902 vs. 0.857) and lower HD95 (5.20 mm vs. 8.01 mm), showing statistically significant improvements (p < 0.0001).
- Time efficiency improved by 81.63% with AI-assisted delineation, reducing median contouring time from 55.0 minutes to 10.0 minutes (p < 0.0001), while also decreasing performance variability across centers and physician expertise levels.
- The research validates the clinical applicability of AI-assisted delineation in enhancing delineation performance and promoting healthcare equity, with potential for widespread implementation, though some authors have competing interests as employees of the involved software company.