Isometric Convolutional Neural Networks for Bone Suppression of Multi-Planar Dual Energy Chest Radiograph

Abstract

Multi-planar bone suppression is an extended version of conventional anterior view bone suppression problem. It is expected to provide more information about the chest radiograph for medical diagnosis and for 3D reconstruction. In this study, convolutional neural networks with isometric architecture are explored to solve multi-planar bone suppression problem from dual energy chest radiographs and compared with conventional architecture, pyramidal architecture. As a result, the proposed isometric architecture has generally better performance and could produce medically diagnosable multi-planar bone-suppressed chest radiographs.