Constrained Labeling of 2D Slice Data for Reading Images in Radiology

Abstract

An important and underestimated task to support reading of images in radiology is a proper annotation of findings. In radiology reading, 2D slice images from a given modality (e.g. CT or MRI) need to be analyzed carefully by a radiologist, whereas all clinical relevant findings have to be annotated in the images. This includes information in particular for documentation, follow-up investigations and medical team meetings. The main problem of the automatic placement of labels in a clinical context is to find an arrangement of multiple variable-sized labels which guarantees readability, clearness and unambiguity and avoids occlusion of the image itself. Based on a case study of abdominal CT-Images in an oncologic context we analyze the main constraints for label placement in order to extract candidate label positions, evaluate these and determine valid and good label positions. Based on this preprocessing step, different approaches can be applied for placing multiple labels in a scene. We present a new method called Shifting and compare it to other labeling strategies.