Landmark based prostate MRI segmentation via improved level set method

Accurate and efficient segmentation of prostate is a significant but difficult task for numerous clinical applications like image-guided prostate intervention and detection of prostate cancer. In this paper, a novel landmark detection based prostate magnetic resonance imaging (MRI) segmentation method is proposed via a modified level set formulation with shape prior. Firstly, the medium slice of prostate MR data is segmented artificially to offer the prior information. Secondly, a prostate detection is implemented by the similarity estimation of texture presentation. Thirdly, we apply a texture constrain procedure to avoid the delineation of fake contour. Finally, the contour of prostate can be captured by the improved level set model with shape prior. A set of experiments using real prostate magnetic resonance images is carried out to demonstrate the significant improvements of our methods on both segmentation accuracy and noise sensitivity comparing to the state-of-the-art methods. The average dice coefficient is 0.91 with expert-defined prostate segmentation in 72 images of 18 different individual.