SACFNet: Spatial Attention and Channel Feature Fusion Network for Pulmonary Nodules Detection.
Authors
Affiliations (7)
Affiliations (7)
- The School of Management Science and Engineering, Shandong University of Finance and Economics, Jinan, China.
- School of Computer Science and Technology, Shandong University of Finance and Economics, Jinan, 250014, China. [email protected].
- School of Information Science and Engineering, Qilu Normal University, Jinan, 250200, China. [email protected].
- School of Information Science and Engineering, Qilu Normal University, Jinan, 250200, China.
- Faculty of Information Technology, Beijing University of Technology, Beijing, 100083, China.
- School of Software and BNRist, Tsinghua University, Beijing, 100190, China.
- School of Psychology and Neuroscience, University of Glasgow, Glasgow, UK. [email protected].
Abstract
The fast development of computing and human-machine interaction technologies have enabled the computer-aided diagnosis (CAD) of lung diseases. Though significant progress has been made in recent years, CAD still faces the challenge of eliminating or minimizing the misdiagnosis. To this end, inspired by the diagnostic practices of radiologists, we present a novel method named "SACFNet" for pulmonary nodules detection in this paper, which integrates Spatial Attention and Channel Feature Fusion Network into the 3D convolutional neural network(CNN). Specifically, our approach incorporates a dual-branch spatial enhancement module and a multi-scale semantic feature fusion module to enhance the spatio features of lung nodule data and improve attention to global information. And design a multi-scale feature enhancement module to increase multi-scale semantic information. The experimental results indicate that the devised SACFNet is accurate in identifying lung nodules in CT scans, with an average Free-Response Receiver Operating Characteristic(FROC) score of 90.98% on the LUNA16 dataset.