Speaker: Hannah Bayat
We present a novel approach to support the investigation of inter-observer variability in tumor segmentation through a multi-faceted, qualitative visual analysis of delineation processes. Tumor segmentation is a crucial step in cancer diagnosis, prognosis, and treatment. Despite the advancements in auto-segmentation tools, clinical practice widely relies on manual delineations performed by radiologists. However, the outcomes of a delineation are subject to
variability. This work aims at capturing the radiologists’ thought processes during a delineation task and at unveiling potential reasons for inter-observer variability. We first present an efficient workflow to capture delineation processes by multiple radiologists. We then describe the design of a visual analysis tool linking the radiologists’ interaction logs and reasoning to the delineation results. Two case studies show that our approach can effectively reveal regions of high uncertainty and suggest explanations for the potential causes of uncertainty.
