Fully Automated Model-Based Prostate Boundary Segmentation Using Markov Random Field in Ultrasound Images

In this paper, a new fully automated model-based approach for segmenting the prostate boundaries in transrectal ultrasound images is proposed. In the preprocessing step, the position of the initial model is automatically estimated using representative patterns. The Expectation Maximization algorithm (EM) and Markov Random Field (MRF) theory are utilized in the deformation strategy to optimally fit the initial model on the prostate boundaries. For the purpose of real time therapy, we propose a less computational complex EM approach for obtaining the probability distribution parameters. We also propose a new internal force energy that uses 2D geometric transformations for preventing the model fault deformation. Successful experimental results with the average Dice Similarity Coefficient (DSC) value 93.9% validate the algorithm.