RSNA 2007 

Abstract Archives of the RSNA, 2007


SST14-04

Evaluation of CT Perfusion Results from Human Tumors: A Comparison of Four Different Two-compartment Models

Scientific Papers

Presented on November 30, 2007
Presented as part of SST14: Physics (CT: New Methods and Applications)

Participants

Volker Hietschold, Presenter: Nothing to Disclose
Philipp Gohl, Abstract Co-Author: Nothing to Disclose
Andrij Abramyuk, Abstract Co-Author: Nothing to Disclose
Arne Koch, Abstract Co-Author: Nothing to Disclose
Michael Laniado MD, Abstract Co-Author: Nothing to Disclose
Nasreddin Djawad Abolmaali MD, Abstract Co-Author: Nothing to Disclose

PURPOSE

To compare four different two-compartment models describing functional parameters of tumor tissues with respect to bias and reliability by means of computer simulations.

METHOD AND MATERIALS

Six pairs of contrast medium (CM) concentration curves and Arterial Input Functions (AIF) were generated based on measurements in patients with a CT Sensation 16 (Siemens). The Lee-Harvey, generalized Kety, Johnson-Wilson and Patlack approaches were applied to determine KTrans (volume transfer constant), rBV (rel. blood volume), ve (rel. interstitial volume), or E (extraction fraction) resp., after adding increasing amounts of Gaussian noise (σ = 1, 3, 6 HU).

RESULTS

Optimum lengths of data acquisition times are about 40 s (Patlack), 60 s (Lee-Harvey), 140 … 180 s (Kety), and 600 s (Johnson-Wilson). KTrans: Most stable results – though biased by down to -50 % – were obtained with the Patlack approach. Using the Kety model, the mean results tend to be more correct at the cost of large stochastic errors. The Lee-Harvey model appeared not to be suited for determination of KTrans due to stochastic errors regularly > 100 %. rBV: Both the Patlack and the generalized Kety models allow for good differentiation between high and low rBVs. ve: With both the Lee-Harvey and the Kety models, differentiation between high and low interstitial volumes is possible only at noise levels about 1 HU. E and ve determined using the Johnson-Wilson model scatter by 50 % or more even when KTrans and rBV are assumed to be known.

CONCLUSION

In routine imaging, sufficiently stable perfusion parameters can be obtained using models with ≤ 3 parameters (plus time shift of the AIF). At an acceptable level of stochastic errors, the generalized Kety model gives the least biased KTrans and rBV results, whereas the Patlack approach distinguishes itself by best reproducibility at significantly biased KTrans results.

CLINICAL RELEVANCE/APPLICATION

For reasons of reliability, determination of CT perfusion parameters should be performed applying most simple model functions like the Patlack approach, even at the cost of a certain bias.

Cite This Abstract

Hietschold, V, Gohl, P, Abramyuk, A, Koch, A, Laniado, M, Abolmaali, N, Evaluation of CT Perfusion Results from Human Tumors: A Comparison of Four Different Two-compartment Models.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5003088.html