Contour assessment tool for quality assurance (CAT-QA) to speed up online adaptive radiotherapy

Objective.Manual review of daily auto-generated contours remains a challenge for clinical implementation of online adaptive radiotherapy (OART). This study introduces a contour assessment tool for quality assurance (CAT-QA), an automatic workflow designed to flag organs-at-risk (OAR) contours requiring manual revision.Approach.CAT-QA applies sequential geometric and dosimetric tests to each auto-generated OAR contour to flag structures requiring review. The tool was retrospectively applied to ten head and neck (H&N) patients (44 CTs with manual contours) treated with proton therapy, split into training and test sets. For each image, three treatment plans were created: one with manual contours (Gold), one with automatic OAR contours (Auto), and one combining auto-contours with manual ones for flagged OARs (CAT-QA plan). Generalizability was assessed on six abdominal patients (8 CTs) without retuning.Main Results.CAT-QA flagged 21% of OARs in H&N and 27% in abdominal cases. No dose failures (>5% of prescribed dose vs. Gold) were observed in H&N. One abdominal OAR (1.4%) exceeded this threshold. In contrast, auto plans resulted in dose failures in 7.5% H&N and 8.5% (abdomen). The higher flag rate observed in the abdomen was driven by a single failed auto-contouring case; excluding this outlier, the average flag rate was 20%, comparable to H&N. CAT-QA runtime averaged <2 min, supporting feasibility for integration into online workflows.Significance.CAT-QA shows promise for improving the safety and efficiency of auto-contouring in OART by flagging OARs that need manual review, with initial results suggesting generalizability across treatment sites.