A case study highlighting the benefits of a novel three dimensional evidence driven, shape model supported, segmentation algorithm

The use of time-of-treatment imaging is increasingly commonplace in radiotherapy, with cone beam CT (CBCT) the primary modality. To make full use of this technology, the tumour target and critical organs need to be segmented from the image volumes. CBCT images unfortunately often suffer from poor contrast and significant image artefacts compared with the standard CT image sets used in treatment planning, making this task more difficult. Delineations created using the common manual contouring tools suffer, in terms of observer variability, as a result of this. We have recently described a novel segmentation algorithm. The algorithm is fully three dimensional and incorporates prior knowledge but with ultimate control of the delineation morphology remaining with the user. Using multi-observer bladder delineations on a set of serial in-treatment CBCT image volumes, we colourwash three dimensional renderings of the mean surface to visualise how certain features in our algorithm and its implementation can help reduce inter-observer variability.