Extracting volumetric information from standard two-dimensional radiological annotations within the clinical workflow

In a typical radiological reporting workflow, radiologists make image-based annotations to denote regions of clinical significance or to perform quantitative measurements. Interestingly, virtually all annotation software allow only 2D geometric primitives such as line segments and ellipses; 3D volume annotation is not supported. As a result, when dealing with anatomic entities that have volumetric properties (e.g. tumors, organs), a radiologist must summarize volumetric quantities in a written text-report or use a third party software outside the standard workflow to perform volumetric segmentation. In this paper, we describe an automated method to extract volumes from radiological annotations. Specifically, we describe a clustering method that parses the annotations of unconnected line segments to determine the locations of volumes. We show how this extracted information can be used to bootstrap and accelerate subsequent 3D segmentation while avoiding the need to perform redundant markup or segmentation seeding outside the standard radiological workflow. This 3D data can be utilized to enhance important clinical applications such as radiological reporting, exam summarization and visualization.