ABSTRACT
At face value the assessment of auto-contouring performance in radiotherapy may seem trivial: is the contour correct or not? In this chapter, various approaches to the evaluation of auto-contouring are discussed, broadly categorized into quantitative, subject, and clinical assessment methods. For quantitative assessment, challenges in implementation of common methods are considered – particularly related to the segmentation representation used in the radiotherapy workflow: DICOM RTSS. To assist in a common framework for evaluation, python code is provided to perform quantitative assessment directly on RTSS. For all types of evaluation, the impact of inter-observer variation is highlighted. How can we evaluate contouring performance in the absence of a known correct answer? – contouring in radiotherapy is inherently subjective. The chapter concludes with recommendations as to the type of assessments that should be performed according to the purpose of the evaluation.
