Pathological Pulmonary Lobe Segmentation from CT Images using\n Progressive Holistically Nested Neural Networks and Random Walker

2017/08/15 by Kevin George, George, Kevin, Adam P. Harrison +7
Medicine · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Lung Cancer Diagnosis and Treatment

paper · pdf · doi:10.48550/arxiv.1708.04503

openalex publication_date 2017/08/15 · openalex created_date 2022/10/01 · openalex updated_date 2026/07/28

Abstract

Automatic pathological pulmonary lobe segmentation(PPLS) enables regional\nanalyses of lung disease, a clinically important capability. Due to often\nincomplete lobe boundaries, PPLS is difficult even for experts, and most prior\nart requires inference from contextual information. To address this, we propose\na novel PPLS method that couples deep learning with the random walker (RW)\nalgorithm. We first employ the recent progressive holistically-nested network\n(P-HNN) model to identify potential lobar boundaries, then generate final\nsegmentations using a RW that is seeded and weighted by the P-HNN output. We\nare the first to apply deep learning to PPLS. The advantages are independence\nfrom prior airway/vessel segmentations, increased robustness in diseased lungs,\nand methodological simplicity that does not sacrifice accuracy. Our method\nposts a high mean Jaccard score of 0.888\±0.164 on a held-out set of 154 CT\nscans from lung-disease patients, while also significantly (p < 0.001)\noutperforming a state-of-the-art method.\n

Related

• Automatic Pulmonary Lobe Segmentation Using Deep Learning
• Efficient 3D Fully Convolutional Networks for Pulmonary Lobe Segmentation in CT Images
• Development of a Multi-Task Learning V-Net for Pulmonary Lobar Segmentation on Computed Tomography and Application to Diseased Lungs
• CT-Realistic Lung Nodule Simulation from 3D Conditional Generative Adversarial Networks for Robust Lung Segmentation
• Lung Diseases Image Segmentation using Faster R-CNNs