Caps-3WD: a three-way diagnostic CapsNet for COVID-19

Weiming Xiong; Denghao Dong; Cong Guo; Libo Zhang; Huamin Wang

doi:10.1117/12.2636646

10 June 2022 Caps-3WD: a three-way diagnostic CapsNet for COVID-19

Weiming Xiong, Denghao Dong, Cong Guo, Libo Zhang, Huamin Wang

Proceedings Volume 12179, Second International Conference on Medical Imaging and Additive Manufacturing (ICMIAM 2022); 1217907 (2022) https://doi.org/10.1117/12.2636646
Event: Second International Conference on Medical Imaging and Additive Manufacturing (ICMIAM 2022), 2022, Xiamen, China

Abstract

Pulmonary medical image processing is an effective diagnostic method for COVID-19, and CapsNet-based methods have achieved good performance. However, as cost-blind methods, these diagnostic methods only consider immediate and deterministic decisions, which easily lead to misdiagnosis and high costs. Therefore, based on a revised CapsNet, we propose a cost-sensitive three-way decision (3WD) method for COVID-19 diagnosis, named as Caps-3WD. To enhance the feature extraction ability for pneumonia areas, we introduce a Restage module to improve convolution layer of the original CapsNet. Further, to lighten the model, we introduce depth wise separable convolution to reconstruct decoder. Additionally, three options are considered in the decision set: infected, normal, and suspected, which are given different costs, respectively. The lowest-cost decision is chosen for each input. In the experimental analysis, we compare Caps- 3WD with CNN-based and CapsNet-based methods on COVID-CXR dataset, which proves the effectiveness of 3WD and the superiority of Caps-3WD in COVID-19 diagnosis.

Citation Download Citation

Weiming Xiong, Denghao Dong, Cong Guo, Libo Zhang, and Huamin Wang "Caps-3WD: a three-way diagnostic CapsNet for COVID-19", Proc. SPIE 12179, Second International Conference on Medical Imaging and Additive Manufacturing (ICMIAM 2022), 1217907 (10 June 2022); https://doi.org/10.1117/12.2636646

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available