A novel stochastic combination of 3D texture features for automated segmentation of prostatic adenocarcinoma from high resolution MRI

Abstract

In this work, we present a new methodology for fully automated segmentation of prostatic adenocarcinoma from high resolution MR by using a novel feature ensemble of 3D texture features. This work represents the first attempt to solve this difficult problem using high resolution MR. The difficulty of the problem stems from lack of shape and structure in the adenocarcinoma. Hence, in our methodology we compute statistical, gradient and Gabor filter features at multiple scales and orientations in 3D to capture the entire range of shape, size and orientation of the tumor. For an input scene, a classifier module generates Likelihood Scenes for each of the 3D texture features independently. These are then combined using a weighted feature combination scheme. The ground truth for quantitative evaluation was generated by an expert pathologist who manually segmented the tumor on the MR using registered histologic data. Our system was quantitatively compared against the performance of the individual texture features and against an expert's manual segmentation based solely on visual inspection of the 4T MR data. The automated system was found to be superior in terms of Sensitivity and Positive Predictive Value. © Springer-Verlag Berlin Heidelberg 2003.