RSNA 2014 

Abstract Archives of the RSNA, 2014


A Novel Technique for Organ Segmentation Utilizing Dynamic Contrast Enhanced Multiphase Datasets

Scientific Papers

Presented on December 2, 2014
Presented as part of SSG07: Informatics (3D, Quantitative and Advanced Visualization)


Elizabeth Weidman MD, Presenter: Nothing to Disclose
Sadaf Jalili MS, RRA, Abstract Co-Author: Nothing to Disclose
Pascal Spincemaille PhD, Abstract Co-Author: Nothing to Disclose
Krishna Juluru MD, Abstract Co-Author: Nothing to Disclose
Jonathan Paul Dyke PhD, Abstract Co-Author: Nothing to Disclose


Tissue enhancement curves may offer a unique organ-signature. We test the accuracy of an automated algorithm that utilizes contrast enhancement curves from dynamic MRI datasets for the purpose of liver segmentation and volume determination. 


Abdominal DCE MRIs obtained from a previously-concluded IRB approved study were retrospectively included. Scans were obtained following injection of Gd-EOB-DTPA at 0.025 mmol/kg at a rate of 1cc/s followed by saline flush. Utilizing a spiral 3D technique, images of the entire abdomen were obtained in 15 phases within the first 90 seconds, following by a single phase every minute up to 20 minutes. All phases were registered using a commercially-available software tool. The custom tool being assessed in this study (created in IDL) allowed a user to identify an organ of interest by placement of an ROI. All voxels in the entire multi-phase MRI dataset demonstrating high correlation (Pearson's R2) with the time-intensity curve of the voxels in the ROI were determined to belong to the same organ and therefore included in segmentation. Volumes of the liver calculated by this automated technique were compared to volumes generated by manual tracing performed by an experienced 3D technologist.


23 DCE MRI exams were included in the analysis. Average liver volumes obtained manually and by the automated technique were 2053 ± 454 mL and 1703 ± 405 mL, respectively (p<0.0001), and were highly correlated (R2 = 0.76). Areas of discrepancy between automated technique and manual segmentation were most frequent at the periphery of the liver due to signal intensity loss secondary to respiratory motion and receiver coil placement.


Liver volumes obtained by the automated and manual techniques were highly correlated. Unlike manual measurements, automated measurements excluded intra-hepatic vascular structures, likely explaining the systematic reduction in automated volumes as compared to manual, and in fact likely providing a more accurate estimate of liver parenchymal volume. Tissue time-intensity curves do offer a unique organ-signature that may be utilized for liver and other organ segmentation. 


Information in DCE MRI datasets can be utilized for automated organ segmentation and may provide an accurate, reproducible and fast method for obtaining organ volumes in surgical planning, transplant evaluation, and monitoring drug treatment response.

Cite This Abstract

Weidman, E, Jalili, S, Spincemaille, P, Juluru, K, Dyke, J, A Novel Technique for Organ Segmentation Utilizing Dynamic Contrast Enhanced Multiphase Datasets.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL.