2019/02/01 by Christoph Angermann, Angermann, Christoph, Haltmeier Markus +7 · 1 citation
Computer Science · #Advanced Neural Network Applications #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Medical Image Segmentation Techniques #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1902.00347
openalex publication_date 2019/02/01 · openalex created_date 2020/07/16 · openalex updated_date 2026/07/28
Convolutional neural networks are state-of-the-art for various segmentation\ntasks. While for 2D images these networks are also computationally efficient,\n3D convolutions have huge storage requirements and require long training time.\nTo overcome this issue, we introduce a network structure for volumetric data\nwithout 3D convolutional layers. The main idea is to include maximum intensity\nprojections from different directions to transform the volumetric data to a\nsequence of images, where each image contains information of the full data. We\nthen apply 2D convolutions to these projection images and lift them again to\nvolumetric data using a trainable reconstruction algorithm.The proposed network\narchitecture has less storage requirements than network structures using 3D\nconvolutions. For a tested binary segmentation task, it even shows better\nperformance than the 3D U-net and can be trained much faster.\n