Applying trajectory mining in medical image data

Abstract

During evaluation of CT or MR images radiologists navigate through a volume in different orientations in order to detect a disease. While doing so, they leave a trail, which might hold valuable information for other clinicians. Unfortunately, current systems do not analyze this trail for certain motions or potential patterns. In this work we developed and implemented different strategies to infer the manifestation of a disease from the trail of inspection. Furthermore we evaluate the effectiveness of these strategies by conducting an experiment in which clinicians had to find a tumor in several cases. The results suggest that inferring suspicious areas from the trail is possible.