AI segmentation as a quality improvement tool in radiotherapy planning for breast cancer

Abstract

AI segmentation has been recently introduced in the local department for delineation of targets and organs-at-risk (OAR) for a wide range of tumour sites. For breast radiotherapy, AI segmentation can provide target delineation (breast and lymph nodes) and required OAR, and this has enabled a stepwise series of improvements to the local planning technique.

Clinician feedback deemed 67 - 89 % of nodal target volumes required no edits or only minor edits, so AI breast and lymph nodes volumes were first used to guide tangent and supraclavicular field placement, instead of a bony-anatomy based technique.

Next, evolution from anatomical field-placement to true inverse optimised planning was introduced using AI to create the required target volumes. For internal mammary node (IMN) treatments, the previous 3-field technique prohibited Deep Inspiration breath-hold (DIBH), due to the couch rotation used to match field edges. The roll-out of VMAT (volumetric-modulated arc therapy) with DIBH enabled by AI therefore resulted in a dose reduction to ipsi-lateral lung, and in mean heart dose compared to the old 3-field technique. Median time from CT scan to VMAT IMN plan approval reduced from 12 days (with manual contouring) to 7 days using reviewed and edited AI-generated volumes.

Consistent, high-quality contours for 9 OAR and breast PTVs for all patients facilitates comparison with NHS-E scorecards as a benchmark for plan quality. Workflows have been simplified, with significant time-savings. DIBH radiotherapy is now available to more patients, further improving dose sparing for heart and lung.