Lijun Zhang, Debing Zhuo, Xiaolu Wu, Yuchou He, Guodong Kang
Abstract
Accurate classification of pulmonary nodules is essential for early lung cancer diagnosis, yet the limited spatial information provided by a single computed tomography slice may hinder the characterization of complex nodule morphology. Although multi-view strategies can capture complementary anatomical information, conventional fusion methods often fail to account for lesion-specific features and differences in the contribution of individual views. To address these limitations, we propose a tri-view pulmonary nodule classification framework that combines Lesion-Guided Attention (LGA) with View-Weighted Fusion (VWF).
Introduction
Background and motivation
Lung cancer remains one of the leading causes of cancer-related mortality worldwide. According to the 2020 global cancer estimates, approximately 2.2 million new cases and 1.8 million deaths occur annually [1]. Prognosis is strongly associated with the stage at diagnosis, and early detection can substantially improve patient survival [2]. Pulmonary nodules are common early radiological manifestations of lung cancer; therefore, accurate benign–malignant classification is important for screening, treatment planning, and follow-up management.
Materials and methods
Methodology
Overall model architecture.
This study develops a tri-view model for pulmonary nodule benign–malignant classification using MedViT as the encoder backbone, as illustrated in Fig 1. Representative axial, coronal, and sagittal slices are processed by independent encoding branches, refined using LGA, and adaptively aggregated using VWF before classification.
Results
Comparison with baseline models
The test-set performance of the proposed model and baseline models is summarized in Table
Discussion
This study evaluated a tri-view pulmonary nodule classification framework combining lesion-prior-guided soft residual feature modulation with sample-adaptive view-weighted fusion. On the held-out test set, the proposed multi-view model achieved an ACC of 93.08%, an AUC of 97.06%, and an F1-score of 92.68%, obtaining the highest values for these three metrics among the models included in the baseline comparison. In the ablation study, the Full configuration achieved the highest ACC and F1-score, whereas MV + LGA achieved a slightly higher AUC.
Conclusion
We developed a tri-view pulmonary nodule classification framework that integrates lesion-prior-guided soft residual feature modulation with sample-adaptive view-weighted fusion to exploit complementary information from axial, coronal, and sagittal views. On the held-out LIDC-IDRI test set, the proposed framework achieved an ACC of 93.08%, an AUC of 97.06%, and an F1-score of 92.68%.
Acknowledgments
The authors thank The Cancer Imaging Archive (TCIA) for hosting the publicly available LIDC-IDRI dataset used in this study.
Citation: Zhang L, Zhuo D, Wu X, He Y, Kang G (2026) Multi-view attention-based deep learning for benign–malignant classification of pulmonary nodules. PLoS One 21(9): e0357963. https://doi.org/10.1371/journal.pone.0357963
Editor: Yongjie Li, University of Electronic Science and Technology of China, CHINA
Received: March 14, 2026; Accepted: August 25, 2026; Published: September 15, 2026
Copyright: © 2026 Zhang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting information files.
Funding: This work was supported by the National Natural Science Foundation of China (Grant Nos. 62565010, 11464015, and 12064012 to GK; Grant No. 52268049 to DZ; https://www.nsfc.gov.cn/), the Natural Science Foundation of Hunan Province (Grant No. 2020JJ4496 to GK; https://kjt.hunan.gov.cn/), and the Science Research Foundation of the Education Department of Hunan Province (Grant No. 20A0585 to GK; https://jyt.hunan.gov.cn/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
