Binary Image Manifold to Overcome Computational Complexities

Abstract

Lately, manifold learning based atlas selection methods emerged as very promising methods. However, because of the complexity of prostate structures in raw images, it is not easy to obtain accurate atlas selection results by only calculating the space between raw images around the manifolds. Multitask based technique is generally utilized in medical image segmentation. In multitask based image segmentation, atlas selection and combination is thought as two important aspects affecting the performance. Even though the distance between your regions to become segmented across images could be readily acquired through the label images, it's infeasible to directly compute the space between your test image (grey) and also the label images (binary). In contrast to various other existing methods, the experimental results on prostate segmentation from T2w MRI demonstrated the selected atlases are nearer to the prospective structure and much more accurate segmentation was acquired by utilizing our suggested method. This paper attempts to address this issue by proposing a label image restricted atlas selection method, which exploits the label images to constrain the manifold projection of raw images. Analyzing the information point distribution from the selected atlases within the manifold subspace, a manuscript weight computation way of atlas combination is suggested.