Segmentation of Prostate Magnetic Resonance Images Based on an Improved Distance Regularized Level Set Evolution (DRLSE) Model

Abstract

Segmentation of prostate magnetic resonance images is of great significance in the interventional diagnosis and treatment of prostate diseases. In this work, the conventional distance regularized level set evolution (DRLSE) model is improved and applied to prostate segmentation. In magnetic resonance image, the prostate boundary near the bladder is often blurred, while that near the urethra is clear, resulting in a poor performance for the traditional gradient information indicator function. In this study, two indicator functions were used to control the evolution of boundary in the clear segment and blurred segment, respectively, to achieve better segmentation. In addition, an energy check term was added to the external energy function to prevent evolution from stopping at a false boundary. This modification could drive the level set to move to regions with large gray level fluctuation and stop evolution at a blurred boundary. Experimental results demonstrated that the performance of prostate segmentation was satisfactory, judging from the Dice similarity coefficient (DSC) which reached an average of 96%.

Keywords

• distance regularized level set
• indicator function
• evolution curve
• energy constraint