Segmentation of prostate from 3-D ultrasound volumes using shape and intensity priors in level set framework

This paper presents a fully automatic prostate segmentation system in transrectal ultrasound images based on 3-D shape and intensity priors. 2-D manual segmentations from training image data are stacked to create the coarse 3-D shape. Min/Max flow is used to transform each coarse shape into smooth 3-D surface. Principle component analysis method is utilized to extract the 3-D shape mode from the training data sets. In a Bayesian inference, the nonlinear shape model is integrated with a nonparametric intensity prior and define a region based energy function. The energy is minimized in a level set frameworks and the control parameters of the convergence lead to the final segmentation. The developed method was tested on 3-D transrectal ultrasound images and its performance compared with manually-defined ground truth. The correct segmentation rate is 0.82.