Program Information
Portal Venous Perfusion Quantitation From Liver DCE-MRI by Voxel Uptake Curve Clustering and Input Function Normalization
A Johansson*, J Balter , M Feng , Y Cao , University of Michigan, Ann Arbor, MI
Presentations
WE-FG-206-10 (Wednesday, August 3, 2016) 1:45 PM - 3:45 PM Room: 206
Purpose: Portal venous perfusion (PV) quantified from liver DCE-MRI by a dual-input single-compartment model is sensitive to errors in the PV input function (PVIF) due to partial volume effect and respiratory motion. To improve repeatability of the local PV perfusion estimate and reduce computation complexity, we developed and evaluated an approach that combines dynamic curve clustering and parameter estimation with normalization of the PVIF.
Methods: The PVIF and arterial input function (AIF), after initial transient behaviors, should reach a state where the contrast concentrations in both blood pools are the same. The PVIF was therefore rescaled such that its tail matched that of the AIF. Also, to reduce the motion effect on the PV perfusion images, voxels with similar contrast uptake curves were pooled together by a clustering algorithm, and then the perfusion parameters of the dynamic curves of the cluster centers were estimated. Perfusion maps were created based upon probabilities of a voxel belonging to the clusters and the perfusion parameters of the clusters. The proposed method was evaluated for repeatability of local PV perfusion in normal tissue for 8 patients with repeated examinations, and by the Pearson correlation coefficient of mean PV perfusion of the whole liver to an independent global liver function measure using ICG clearance rate analyzed over 89 examinations.
Results: The average PV perfusion RMSE of the repeated measurements was reduced from 52.4 by the conventional voxelwise estimation to 33.7 ml/(100 ml*min) by the proposed method. Similarly, the correlation coefficient of the total perfusion to ICG clearance rate was increased from 0.21 to 0.46 by the new method.
Conclusion: This technique shows promise in improving repeatability of PV perfusion from DCE-MRI of the liver. Future developments have the potential to further reduce systematic and random errors in the PV perfusion quantification, and computational complexity.
Funding Support, Disclosures, and Conflict of Interest: NIH RO1 CA132834, NIH PO1 CA59827
Contact Email: