ROSE: Multi-level super-resolution-oriented semantic embedding for 3D microvasculature segmentation from low-resolution images

Yifan Wang, Haikuan Zhu, Hongbo Li, Guoli Yan, Sagar Buch, Ying Wang, Ewart Mark Haacke, Jing Hua, Zichun Zhong

Abstract: Highlights•A deep learning framework to construct the joint multi-level hybrid embedding spaces, where the computed micro vessel probabilities for the 3D volume processing can be synergistically enhanced from the compound 2D image SR acquisition.•A multi-tasking convolutional neural network (CNN) framework to semantically integratively and SR-orientedly learn the feature vectors of 3D volume and compound 2D SR image, and explore their inter-dependencies.•Segmentation and visualization of high-fidelity 3D microvasculature with complicated geometry and tiny sizes from the raw LR volumetric images.•Demonstrating the potential in a novel approach using MR angiography and venography in the diagnosis of microvascular disease.