Comparison of Auto-Contouring Tools for Delineation of Normal Organs at Risk in Paediatric Patients Undergoing...

Contouring organs at risk (OARs) manually in paediatric patients undergoing cranial-spinal radiation therapy (CSI) is a time-consuming, labour-intensive task. This study aims to assess the accuracy and clinical acceptability of auto-contours produced by the Siemens DirectORGANS auto-contouring software on paediatric patients receiving CSI treatment.

Auto-contours of OARs were produced using the Siemens DirectORGANS Auto-contouring Software from 20 paediatric CSI patients datasets that had previously been manually contoured by a paediatric specialist radiation therapist (RT) for plan production. Manual and auto-contours were retrospectively analysed using quantitative (Dice Similarity Coefficient, Hausdorff Distance, Mean Distance to Agreement) and qualitative (Likert ratings, Turing test) assessment techniques.

Auto-contoured structures were clinically acceptable for use without edits 72.8% of the time, and manual contours were clinically acceptable for use 91.7% of the time. The liver was the only auto-contoured structure that performed better than the manual equivalent. Poor performance by the auto contouring tool was noted for structures surrounded by low contrast edges, such as the breasts, oesophagus, and brainstem, in both quantitative and qualitative assessment techniques. Brain auto-contours were deemed not suitable for clinical use.

The clinical acceptability of many of the auto-contours favours the implementation of this auto-contouring system for clinical use. However, prior to use, all contours should be critically assessed and edited accordingly. Our study results indicate that whilst auto-contouring tools are designed for adult populations, they are suitable for use on paediatric patients when used with caution.