GlioView automates tumor segmentation and assessment of tumor volumetric change, while providing an estimate of uncertainty in these measurements. In this way, GlioView provides clinicians with an increased level of confidence for interpreting imaging data and improve clinical decision-making.
FIGUREhttp://abstract.rsna.org/uploads/2016/16006734/16006734_1dnb.jpg
BackgroundRadiological characterization of high-grade primary brain tumors plays a vital role in diagnosing, treating, and evaluating treatment response. Accurate measurement of tumor burden can be a challenging, labor intensive, and time consuming task for neuroradiologists. Additionally, measurements may have high variability due to rater interpretation and the heterogeneous nature of high-grade gliomas. Currently, there is a significant need for a radiology workstation that can quickly segment brain tumors, including measures of segmentation uncertainty, and provide the radiologist with clear visualization of changes that occur during clinical follow-up.
EvaluationTo address this need, we created an application that: 1) performs automated volumetric segmentation and analysis of brain tumor MR images for each tumor component; 2) visualizes the results and error bounds in tumor segmentation; and 3) captures change and variability over time that arises on subsequent examinations.Segmentation error was estimated by iterative measurements of the tumor boundary using a novel and accurate knowledge-based approach that was incorporated into an automated pipeline that includes image preprocessing and tumor segmentation for all time points for a patient.This process was built into a graphical user interface (GlioView) that lets the user run the automated pipeline and select different visualization options. Results are visualized as color-coded overlays on top of the original images alongside the volume variability estimates.
DiscussionMeasurement of segmentation uncertainty is important for the evaluation of tumor changes over time. GlioView provides an intuitive interface for image analysis and provides an estimate of tumor extent, which can be used for RANO evaluation. The addition of error bounds alongside volumetric measurements provides a more objective approach for interpreting over time change.