Human-computer interaction in segmenting organs at risk for radiotherapy: a pilot study

Human-computer interaction in segmenting organs at risk for radiotherapy: a pilot study

Ramkumar, Anjana and Dolz, Jose and Kirisli, Hortense A and Schimek-Jasch, Tanja and Adebahr, Sonja and Nestle, Ursula and Massoptier, Laurent and Varga, Edit and Stappers, Pieter Jan and Niessen, Wiro J

Multimodal imaging towards individualized radiotherapy treatments 2014

Abstract : An accurate segmentation of organs at risk in CT images is a prerequisite in radiotherapy treatment planning. Although there a re a number of automatic segmentation methods, most of them require user interactions during the pre – and post – processing stages. Those interactions directly influence the effectiveness of the segmentation results and the efficiency of the process. In this paper, we explored the effects of user interactions in using a semi – automatic segmentation method, which is based on an algorithm combining watershed and graph – cut methods. The aims of this study are 1) to identify if the users can are comfortable with th ose interactions, 2) to evaluate the quality of results with respect to manual segmentations, and 3) to explore relations of human – computer interactions and the quality of the results in the use of the method. Based on pre – defined protocols, two physicians contoured lung, heart and spinal cord in several cases using the proposed method. The contouring process was video – taped and the segmentation results were analysed . Comparing the results to manual segmentation, an average Dice similarity coefficient (DSC) of 0.95, 0.7 and 0.8 was obtained by both clinicians for lungs, heart and spinal cord respectively. In the qualitative evaluation, despite a lot of post processing actions, the users were satisfied with the proposed method, as they were able to control th e system to produce sound results in an efficient manner.