Work

Semantic quality control of cervical cancer RTSTRUCT data through threshold stress-testing and technical DVH executability assessment

Year: 2026

Type: article

Abstract: Technically readable radiotherapy structure sets do not necessarily contain structures suitable for organ-specific dose analysis. We developed a purpose-aware framework for pre-analytic semantic quality control (semantic QC) of real-world RT Structure Set (RTSTRUCT) data and evaluated threshold-dependent review workload and technical dose-volume histogram (DVH) executability. This single-centre retrospective framework-development study included 134 cervical cancer radiotherapy cases and 4311 regions of interest. Cases were assigned to rule-derived QC Levels 1–3, and four empirical cut-points were varied in a complete factorial design of 81 configurations. Technical DVH executability was evaluated across 670 case-OAR tasks. Essential OARs accounted for 1127 ROI records, whereas auxiliary optimisation and dose-shell/isodose-like objects accounted for 1376. The review set ranged from 16 to 107 cases, with review status changing in 91/134 cases. Moderate unmapped-ROI and suspected-duplicate thresholds accounted for 85.0% of the total review-workload sum of squares. Of 670 case-OAR tasks, 545 had a standardised-name candidate, 520 had non-empty multi-slice contours, and all 520 passed technical checks. Technical DVH executability was 77.6% overall and 95.4% among named candidates. Within this internal framework-development cohort, these findings support traceable pre-analytic triage and failure localisation but do not establish semantic accuracy, expert-confirmed structure identity, TPS agreement, clinical dose validity or external generalisability.

Source: Scientific Reports

DOI: 10.1038/s41598-026-74864-y