TY - JOUR
T1 - CMFuse
T2 - Correlation-based multi-scale feature fusion network for the detection of COVID-19 from Chest X-ray images
AU - Liang, Zhihao
AU - Lu, Huijuan
AU - Zhou, Rongjing
AU - Yao, Yudong
AU - Zhu, Wenjie
N1 - Publisher Copyright:
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023.
PY - 2024/5
Y1 - 2024/5
N2 - COVID-19 broke out in 2019, seriously affecting people’s health and life. Recent studies have indicated that radiological images carry crucial information about COVID-19. Hence, automatic image classification assisted by artificial intelligence (AI) can be employed as a potential diagnostic tool. Nonetheless, in the task of COVID-19 X-ray image recognition, there are local features, including local vascular dilatation, as well as global features, including large ground glass-like shadows, traditional deep neural networks cannot effectively extract features, and the significance of distinct scale features for the task is also divergent, feature element-wise adding or feature concatenating to fuse features from various branches do not consider the internal correlation between features. In view of the above problems, we propose a Correlation-based Multi-scale Feature Fusion Network (CMFuse), combining the advantages of Convolutional Neural Network (CNN) and Transformer. The model captures local spatial contextual features and global semantic information representation of features at different scales in parallel, and the extracted features are adaptively fused at distinct levels through the feature fusion module, down-sampling and other steps to obtain the final classification results. We evaluated CMFuse on the integrated COVID-19 X-ray image dataset, and the results showed that our model attains 97.36% Accuracy, 99.15% Specificity, 97.27% Recall, 97.17% Precision, and 97.22% F1-score, which outperforms other previous related works.
AB - COVID-19 broke out in 2019, seriously affecting people’s health and life. Recent studies have indicated that radiological images carry crucial information about COVID-19. Hence, automatic image classification assisted by artificial intelligence (AI) can be employed as a potential diagnostic tool. Nonetheless, in the task of COVID-19 X-ray image recognition, there are local features, including local vascular dilatation, as well as global features, including large ground glass-like shadows, traditional deep neural networks cannot effectively extract features, and the significance of distinct scale features for the task is also divergent, feature element-wise adding or feature concatenating to fuse features from various branches do not consider the internal correlation between features. In view of the above problems, we propose a Correlation-based Multi-scale Feature Fusion Network (CMFuse), combining the advantages of Convolutional Neural Network (CNN) and Transformer. The model captures local spatial contextual features and global semantic information representation of features at different scales in parallel, and the extracted features are adaptively fused at distinct levels through the feature fusion module, down-sampling and other steps to obtain the final classification results. We evaluated CMFuse on the integrated COVID-19 X-ray image dataset, and the results showed that our model attains 97.36% Accuracy, 99.15% Specificity, 97.27% Recall, 97.17% Precision, and 97.22% F1-score, which outperforms other previous related works.
KW - COVID-19
KW - Chest X-ray images
KW - Deep learning
KW - Image classification
UR - http://www.scopus.com/inward/record.url?scp=85175305231&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85175305231&partnerID=8YFLogxK
U2 - 10.1007/s11042-023-17431-1
DO - 10.1007/s11042-023-17431-1
M3 - Article
AN - SCOPUS:85175305231
SN - 1380-7501
VL - 83
SP - 49285
EP - 49300
JO - Multimedia Tools and Applications
JF - Multimedia Tools and Applications
IS - 16
ER -