ESTRO 2024 - Abstract Book

S3058

Physics - Autosegmentation

ESTRO 2024

Median DSC scores (clinical versus unedited Limbus) all improved when Limbus was used to produce the initial contour, particularly for the oesophagus and brachial plexus. This indicates that less editing was deemed necessary than predicted by the initial study. The Limbus brachial plexus contours were judged as including some normal tissue structures but sufficiently accurate when considering its PRV in the proximity of the PTV. Three of the five OARs required editing by the clinician, with heart and oesophagus requiring edits on > 4 CT slices. Modifications were however small, indicating the scoring system used painted an overly pessimistic picture.

Table 2. Median DSC for clinical contours versus unedited Limbus contours when clinical contours were created (a) manually and (b) Limbus + expert review.

OAR doses were very similar when the treatment plan was calculated on unedited and edited Limbus contours, with all OAR doses remaining within departmental tolerances. Calculating the treatment plan on the unedited Limbus contours would not have changed the clinical decision making for any of the twenty plans. The Clinical Oncologist reported satisfaction with Limbus and commented that CTV outlining was also more efficient if other contours were included in the Limbus structure template (e.g. aorta, great vessels), from which the CTV could be automatically cropped.

Conclusion:

Our results have shown that Limbus AI can be safely implemented into the treatment planning pathway for the contouring of OARs for radical lung radiotherapy. DSC values showed good agreement between clinical contours and Limbus generated contours. Expert review remains an essential step, to correct errors for patients with atypical anatomy and known differences between Limbus algorithms and our clinical practice. The impact of making minor adjustments to Limbus contours on OAR doses was small. The Clinical Oncologist reported high satisfaction with Limbus, and time savings to the generation of OARs and the CTV. Limbus AI plus expert review has been implemented into routine clinical use.

Keywords: lung radiotherapy, contouring, AI

References:

1. 1 Recommendations | Artificial intelligence technologies to aid contouring for radiotherapy treatment planning: early value assessment | Guidance | NICE [Internet]. NICE; 2023 [cited 2023 Sep 29]. Available from: https://www.nice.org.uk/guidance/hte11/chapter/1-Recommendations