The use of probability maps to deal with the uncertainties in prostate cancer delineation.

Abstract

The use of dynamic contrast-enhanced (DCE) imaging for delineation of prostate tumors requires that decisions are made on a voxel wise basis about the presence of tumor. While the sensitivity and specificity of this technique is high, we propose a probabilistic approach to deal with the intrinsic imaging uncertainty. Twenty-nine patients with biopsy-proven prostate cancer underwent a DCE-CT exam prior to radiotherapy. From a logistic regression on K(trans) values from healthy and diseased appearing prostate regions we obtained a probability function for the presence of tumor. K(trans) parameter maps were converted into probability maps and a stratification was applied at the 5% and 95% probability level, to identify low-, intermediate-, and high-risk areas for the presence of tumor. In all patients, regions with high-, intermediate-, and low-risk were identified, with median volume percentages of 7.6%, 40.0%, and 52.1%, respectively. The contiguous areas that resulted from the voxel wise stratification can be interpreted as GTV, high-risk CTV, and CTV. K(trans) parameter maps from a DCE-CT exam can be converted into probability maps for the presence of tumor. In this way, the intrinsic uncertainty that a voxel contains tumor can be incorporated into the treatment planning process.