Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Semi-supervised lung nodule detection with adversarial learning

Semi-supervised lung nodule detection with adversarial learning Lung cancer has long posed a severe threat to human life and health, and early detection as well as effective treatment can significantly improve the survival rates of patients. In recent years, deep learning has greatly propelled the advancement in image processing, yet its application in pulmonary nodule detection through deep learning is still at an early stage. Therefore, researching automatic pulmonary nodule detection technology based on deep learning, enhancing the performance of lung nodule detection, holds significant importance for improving the survival rates among lung cancer patients. This paper aims to address the above issues and propose a semi-supervised pulmonary nodule detection algorithm based on adversarial learning. A multi-task classification network is designed to distinguish between annotated and unannotated candidate pulmonary nodules while removing false-positive pulmonary nodules. The adversarial loss and maximum-minimum strategy are introduced into the training process, and the unannotated images and adversarial loss are used to update the 3D sliding window semantic segmentation network. To enable the adversarial loss to backpropagate, the algorithm makes some improvements to RoI pooling, connecting the 3D sliding window semantic segmentation network and multi-task classification network for end-to-end learning. In experiments, compared to the supervised lung nodule algorithm using only 50% labeled images, the semi-supervised lung nodule detection algorithm using 50% labeled images and 50% unlabeled images increased the average FROC from 0.677 to 0.717. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Multimedia Tools and Applications Springer Journals

Semi-supervised lung nodule detection with adversarial learning

Download PDF
20 pages

Loading next page...
 
/lp/springer-journals/semi-supervised-lung-nodule-detection-with-adversarial-learning-McntadAAWw

References (26)

Publisher
Springer Journals
Copyright
Copyright © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024
ISSN
1380-7501
eISSN
1573-7721
DOI
10.1007/s11042-024-19074-2
Publisher site
See Article on Publisher Site

Abstract

Lung cancer has long posed a severe threat to human life and health, and early detection as well as effective treatment can significantly improve the survival rates of patients. In recent years, deep learning has greatly propelled the advancement in image processing, yet its application in pulmonary nodule detection through deep learning is still at an early stage. Therefore, researching automatic pulmonary nodule detection technology based on deep learning, enhancing the performance of lung nodule detection, holds significant importance for improving the survival rates among lung cancer patients. This paper aims to address the above issues and propose a semi-supervised pulmonary nodule detection algorithm based on adversarial learning. A multi-task classification network is designed to distinguish between annotated and unannotated candidate pulmonary nodules while removing false-positive pulmonary nodules. The adversarial loss and maximum-minimum strategy are introduced into the training process, and the unannotated images and adversarial loss are used to update the 3D sliding window semantic segmentation network. To enable the adversarial loss to backpropagate, the algorithm makes some improvements to RoI pooling, connecting the 3D sliding window semantic segmentation network and multi-task classification network for end-to-end learning. In experiments, compared to the supervised lung nodule algorithm using only 50% labeled images, the semi-supervised lung nodule detection algorithm using 50% labeled images and 50% unlabeled images increased the average FROC from 0.677 to 0.717.

Journal

Multimedia Tools and Applications – Springer Journals

Published: Jan 1, 2025

Keywords: Lung nodule detection; Deep learning; Adversarial learning; Semi-supervised learning

There are no references for this article.