Panagiotis Korfiatis PhD, Presenter: Nothing to Disclose

Leland S. Hu MD, Abstract Co-Author: Nothing to Disclose

Zachary Samuel Kelm BS, Abstract Co-Author: Nothing to Disclose

Bradley J. Erickson MD, PhD, Abstract Co-Author: Stockholder, Evidentia Health, Inc

Perfusion analysis software is widely available in clinical practice, however it is often treated as a black box tool. The values produced are generally accepted, but validation is challenging. The purpose of the current study is to describe an evaluation framework we developed for the accuracy of relative cerebral blood volume (rCBV) measurements from these tools, as well as investigate their robustness to noise, i.e. their ability to perform measurements of data originating from different image acquisition protocols.

Boxplots analysis was performed to provides a visual representation of variation of rCBV values for each noise level considered in this study for tumor with and without leakage respectively. Moderate positive correlations (0.3<r<0.7) were found between measurements with and without leakage correction. For the tumor with no simulated leakage, strong positive (r>0.7) correction was found for 1 package. In most cases, correlation decreased as the noise level increased.

We developed software that would create DSC images simulating a gadolinium bolus into a “brain” with gray matter, white matter, and 4 tumors with varying levels of contrast agent leakage. The software allows us to introduce varying levels of noise, and to alter the appearance of the bolus. For this study, we are only altering noise levels. rCBV quantification was performed using three commercially-available software packages (nordicICE, GE FuncTOOL and IB Neuro) on simulated brains with fifteen different levels of Gaussian noise. Furthermore, leakage correction was applied when it was available in the software. For each noise level, 21 simulations were performed. We computed errors in tumor rCBV and the Pearson correlation coefficient (r) was calculated to determine the correlation between the output of each software

Increasing noise degrades the performance of all software packages, some more than others. Leakage-correction improved the accuracy of the rCBV calculation for tissues with contrast leakage. Further investigation is needed to evaluate the use of preprocessing methods as a means to provide robustness to noise.

Korfiatis, P, Hu, L, Kelm, Z, Erickson, B, Perfusion Analysis Software Accuracy Evaluation: A Digital Phantom Based Study.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13025090.html