SU‐C‐BRB‐04: Delineation Uncertainty Maps: Proof of Concept Study for Recurrent Gynaecological Cancers

Abstract

Purpose:To create a novel infrastructure which allows clinicians to record their GTV delineation uncertainty; to generate delineation uncertainty maps from this data and compare them to clinical GTVs.Methods:A contouring concept was developed in which clinicians draw up to two GTV boundaries per CT slice corresponding to the inner and outermost possible boundaries the GTV may have.A new uncertainty contouring tool was implemented in our research treatment planning system (TPS, DynaPlan) to accommodate these contour pairs and the derivation of delineation probability maps. 3D probability maps were generated from the contour pairs based on the minimum Chebyshev distance, at each pixel within the contours, from each boundary. The probability associated with each pixel was defined as the ratio of its distance from the inner boundary to the sum of its distances from each boundary.These tools were used to create probability maps using contours produced by a clinician experienced in treating recurrent gynaecological cancers. The corresponding 50% probability volumes (GTVP50) were compared against the observers corresponding clinical GTV (GTVC) contour, created in accordance with local clinical protocols, using the Jaccard Index.Results:This approach was successfully implemented within the research TPS. Comparisons between GTVC and GTVP50 were made for 10 recurrences from 9 patients. GTVC was larger than GTVP50 (p<0.01), with mean±s.d. volumes of 69.6±55.4cm3 and 61.6±53.7cm3 respectively, and a corresponding Jaccard Index of 0.83±0.11.Conclusion:Individual clinician's delineation uncertainty can be recorded and a corresponding uncertainty map derived. This is a crucial step in improving the handling of delineation uncertainty and moving away from population based margins, using techniques such as probabilistic treatment planning. Contours produced for recurrent gynaecological GTVs show the clinical GTV to be consistently larger than volume corresponding to 50% probability. Future work is to investigate non‐linear methods for generating these probability maps.This report is independent research arising from an NIHR/HEE Healthcare Science Doctoral Research Fellowship, David Bernstein, HCS‐DRF‐2014‐05‐005, supported by National Institute for Health Research (NIHR). The views expressed in this publication are those of the author(s); not necessarily those of the NHS, NIHR, Health Education England or Department of Health.