PROJECT TITLE :

Deformable MR Prostate Segmentation via Deep Feature Learning and Sparse Patch Matching

ABSTRACT:

Automatic and reliable segmentation of the prostate is a crucial however difficult task for various clinical applications like prostate cancer radiotherapy. The main challenges for correct MR prostate localization lie in 2 aspects: (one) inhomogeneous and inconsistent look around prostate boundary, and (a pair of) the large form variation across completely different patients. To tackle these 2 issues, we have a tendency to propose a brand new deformable MR prostate segmentation technique by unifying deep feature learning with the sparse patch matching. First, rather than directly using handcrafted options, we tend to propose to be told the latent feature illustration from prostate MR pictures by the stacked sparse auto-encoder (SSAE). Since the deep learning algorithm learns the feature hierarchy from the information, the learned options are usually a lot of concise and effective than the handcrafted features in describing the underlying data. To enhance the discriminability of learned options, we any refine the feature illustration in a supervised fashion. Second, based on the learned options, a sparse patch matching technique is proposed to infer a prostate likelihood map by transferring the prostate labels from multiple atlases to the new prostate MR image. Finally, a deformable segmentation is employed to integrate a sparse form model with the prostate chance map for achieving the ultimate segmentation. The proposed technique has been extensively evaluated on the dataset that contains sixty six T2-wighted prostate MR images. Experimental results show that the deep-learned options are more effective than the handcrafted features in guiding MR prostate segmentation. Moreover, our methodology shows superior performance than other state-of-the-art segmentation methods.