Automatic differential segmentation of the prostate in 3-D MRI using Random Forest classification and graph-cuts optimization

In this paper we address the problem of automated differential segmentation of the prostate in three dimensional (3-D) magnetic resonance images (MRI) of patients with benign prostatic hyperplasia (BPH). We suggest a framework that consists of two stages: in the first stage, a Random Forest classifier localizes the anatomy of interest. In the second stage, Graph-Cuts (GC) optimization is utilized for obtaining the final delineation. GC optimization regularizes the hypotheses produced by the classification scheme by imposing contextual constraints via a Markov Random Field model. Our method obtains comparable or better results in a fully automated fashion compared with a previous semi-automatic technique [6]. It also performs well, when small training sets are used. This is particularly useful in on-line interactive segmentation systems, where prior knowledge is limited, or in automated approaches that generate ground truth used for model-building.