William T.C. Yuh MD, MSEE, Abstract Co-Author: Nothing to Disclose

Jian Z. Wang PhD, Abstract Co-Author: Nothing to Disclose

Steffen Sammet MD, PhD, Abstract Co-Author: Nothing to Disclose

Jun Zhang MD, Abstract Co-Author: Nothing to Disclose

Simon Shek-Man Lo MD, Abstract Co-Author: Nothing to Disclose

Nina A. Mayr MD, Presenter: Nothing to Disclose

Hualin Zhang PhD, Abstract Co-Author: Nothing to Disclose

Michael Vinzenz Knopp MD, PhD, Abstract Co-Author: Nothing to Disclose

00030490-DMT et al, Abstract Co-Author: Nothing to Disclose

Pixel-by-pixel analysis of the poorly enhanced pixel populations within a heterogeneous tumor using dynamic contrast enhanced (DCE) MR has been reported to predict treatment outcome in cervical cancer. The poorly perfused tumor subregions (low-DCE pixels) presumably reflect the areas with poor delivery of oxygen and chemotherapy that contribute to treatment failure. However, the adequate sampling amount of the low-DCE pixels for accurate prediction of treatment outcome has not been defined. This project evaluated the range of thresholds for optimal sampling of the low-DCE pixels populations within the heterogeneous tumor that enable best prediction of outcome early during treatment.

One-hundred-and one patients with advanced cervical cancer underwent DCE MR 2 weeks into the radiation/chemotherapy. The signal intensity (SI) of each pixel was calculated from the time-signal-intensity curve of the DCE MR at the plateau phase. The SI of all pixels within the tumor was plotted as a pixel-SI distribution spectrum. The 2nd through 40th percentiles of the low-DCE tumor pixel population were used to define the low-DCE thresholds, and each percentile threshold was correlated with treatment failure (recurrence and cancer death) using logistic regression analysis.

Stepwise evaluation of the DCE percentile thresholds showed that prediction of tumor recurrence and cancer death was significant from the 2nd to 35th percentile of the low-DCE tumor pixels. The discrimination power progressively decreased for thresholds below or above this range. The most accurate prediction was achieved from the 4th to 10th percentile based on ROC analysis (AUC>0.70) and significance levels of the local tumor control vs. tumor recurrence outcomes (p<0.05).

Our results suggest that a wide range of thresholds for low-DCE pixels provide significance in predicting treatment outcome in cervical cancer, with optimal thresholds in the range the 4th to10th percentile. These preliminary results require future studies to refine the outcome predictor. These findings support our hypothesis that poorly perfused subregions of the heterogeneous tumor contribute to treatment failure, and their quantitative analysis early during the treatment course can predict outcome.

To understand the optimal use of thresholds for low tumor perfusion in the prediction of treatment outcome in cervical cancer.

Yuh, W, Wang, J, Sammet, S, Zhang, J, Lo, S, Mayr, N, Zhang, H, Knopp, M, et al, 0, Tumor Heterogeneity and Outcome Prediction: Sampling Poorly Perfused Tumor Pixels.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8010055.html