External Validation of a Bayesian Network for Error Detection in Radiotherapy Plans

Artificial intelligence (AI) applications have recently been proposed to detect errors in radiotherapy plans. External validation of such systems is essential assess their performance and safety before applying them clinical practice. We collected data from 5238 patients treated at Maastro Clinic introduced a range common plan for the model detect. estimated model’s discrimination by calculating area under receiver-operating characteristic curve (AUC). also assessed its usefulness as an alert system that could reduce need manual checks percentage values flagged positive predictive value (PPV) high sensitivities (95%–99%) error prevalence. The AUC when considering all variables was 67.8% (95% CI, 65.6%–69.9%). varied widely different types (from 90.4% table angle 54.5% planning tumor volume-PTV dose errors). ranged 84% 90% between 95% 99% PPV only slightly higher than prevalence errors. external significantly worse original setting (AUC 68% versus 89%). Its questionable due low potential achieve sensitivity. analyzed apparent limitations we actions overcome them.