TY - GEN
T1 - Label Semantic Improvement with Graph Convolutional Networks for Multi-Label Chest X-Ray Image Classification
AU - Cai, Dachuan
AU - Lu, Huijuan
AU - Chai, Zhuijun
AU - Wang, Renfeng
AU - Zhu, Wenjie
AU - Yao, Yudong
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Multi-label classification of Chest X-ray images is one of the important tasks in the field of medical image analysis, and current methods only capture co-occurrence relationships between labels, making it difficult to establish complex connections among labels. We face a series of challenges when considering global label correlation, visual correlation, and dealing with unbalanced data. In this study, we introduce a Label Semantic Improvement Graph Convolutional Network (LSI-GCN) framework based on GCN, which consists of an Image Representation Learning module, a GCN module with Label Semantic Improvement, and a Distance Metric module to improve classification performance. The task of multi-label classification of chest X-ray images is approached from different perspectives and levels by combining their functionalities to improve the performance, robustness, and accuracy of the model. The Image Representation Learning module is a feature extractor that learns high-level label-specific features from chest X-ray images. The GCN captures label correlations and models global label relationships and category visual correlations, respectively. With this approach, the correlation between disease labels and the relevance of image features can be considered simultaneously, and the accuracy of multi-label classification can be improved. The image features were further optimized using a distance metric module. Experiments were conducted using Chest X-ray14 and CheXpert images as multi-label datasets, and the mean Area Under Curve (AUC) scores obtained were better than those of comparative models such as CheXGCN [3] and the accuracy and robustness of multi-label classification were improved.
AB - Multi-label classification of Chest X-ray images is one of the important tasks in the field of medical image analysis, and current methods only capture co-occurrence relationships between labels, making it difficult to establish complex connections among labels. We face a series of challenges when considering global label correlation, visual correlation, and dealing with unbalanced data. In this study, we introduce a Label Semantic Improvement Graph Convolutional Network (LSI-GCN) framework based on GCN, which consists of an Image Representation Learning module, a GCN module with Label Semantic Improvement, and a Distance Metric module to improve classification performance. The task of multi-label classification of chest X-ray images is approached from different perspectives and levels by combining their functionalities to improve the performance, robustness, and accuracy of the model. The Image Representation Learning module is a feature extractor that learns high-level label-specific features from chest X-ray images. The GCN captures label correlations and models global label relationships and category visual correlations, respectively. With this approach, the correlation between disease labels and the relevance of image features can be considered simultaneously, and the accuracy of multi-label classification can be improved. The image features were further optimized using a distance metric module. Experiments were conducted using Chest X-ray14 and CheXpert images as multi-label datasets, and the mean Area Under Curve (AUC) scores obtained were better than those of comparative models such as CheXGCN [3] and the accuracy and robustness of multi-label classification were improved.
KW - Chest X-ray
KW - Graph Convolutional Networks
KW - Image classification
KW - Multi-label
UR - http://www.scopus.com/inward/record.url?scp=85192486529&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85192486529&partnerID=8YFLogxK
U2 - 10.1109/ITME60234.2023.00147
DO - 10.1109/ITME60234.2023.00147
M3 - Conference contribution
AN - SCOPUS:85192486529
T3 - Proceedings - 2023 13th International Conference on Information Technology in Medicine and Education, ITME 2023
SP - 711
EP - 717
BT - Proceedings - 2023 13th International Conference on Information Technology in Medicine and Education, ITME 2023
T2 - 13th International Conference on Information Technology in Medicine and Education, ITME 2023
Y2 - 24 November 2023 through 26 November 2023
ER -