Reproduce “Improving Clinical Target Volume Segmentation Accuracy using Anatomical Priors and Active Learning for the AGITG TOPGEAR Clinical Trial” on OpenResearch

About this paper

Phillip Chlap, Mark Lee, Trevor Leong, Matthew Field, Jason Dowling, Hang Min, Julie Chu, Jennifer Tan and 6 more

A study integrated anatomical priors and active learning to enhance Clinical Target Volume (CTV) segmentation accuracy for gastric cancer patients in the AGITG TOPGEAR clinical trial. This combined approach yielded a mean Dice Similarity Coefficient of 0.87, improving accuracy and efficiency in automated contour quality assurance for radiotherapy.

Key insights

• Anatomical priors, derived from surrounding organ segmentations, consistently and significantly improve CTV segmentation accuracy, offering a substantial initial boost in performance equivalent to multiple rounds of active learning alone.
• Active learning provides incremental improvements in segmentation accuracy, particularly when applied in conjunction with anatomical priors, demonstrating an efficient method for dataset expansion and model refinement in ongoing trials.
• The ensemble-based confidence score reliably correlates with segmentation performance, providing a valuable metric for guiding active learning case selection and for real-time flagging of potentially inaccurate segmentations in clinical QA workflows.

Summary by alphaXiv.