The Radiation Oncology Feed For radiation oncologists · physicists · RTTs · dosimetrists
technical-relevance lung auto-segmentation AI

High geometric agreement did not guarantee physician acceptance of AI contours

Thoracic AI contours showed strong geometric and dosimetric agreement, but physicians still rejected some heart and oesophageal contours.

RadOncToday · ·1 min read

KEY POINTS

  • Manual contours and two commercial artificial intelligence systems were compared in 23 lung cancer cases for the lungs, heart, oesophagus and spinal canal.
  • Dice similarity coefficients exceeded 0.90 in more than 70% of comparisons, and differences in evaluated dose metrics were generally negligible.
  • Ten thoracic radiation oncologists blindly reviewed ten cases and judged whether each contour could be used without modification.
  • RayStation 2023B oesophageal contours and Ethos-2 heart contours were more frequently considered unacceptable.
  • The commonest reason for rejection was insufficient anatomical accuracy rather than an immediate dose-safety concern.

CLINICAL TAKEAWAY

Dice scores and dose differences cannot replace expert review of auto-segmented structures. Vendor- and organ-specific failure patterns should be assessed locally before automation is allowed to reduce, rather than simply redistribute, contouring workload.

SOURCE

Physics and Imaging in Radiation Oncology