Predicting lung cancer bone metastasis using CT and pathological imaging with a Swin Transformer model

Bone metastasis is a common and serious complication in lung cancer patients, leading to severe pain, pathological fractures, and reduced quality of life. Early prediction of bone metastasis can enable timely interventions and improve patient outcomes. In this study, we developed a multimodal Swin T...

Full description

Saved in:
Bibliographic Details
Main Authors: Wanling Li, Xin Zou, Jie Zhang, Minghong Hu, Guanfeng Chen, Shanshan Su
Format: Article
Language:English
Published: Elsevier 2025-06-01
Series:Journal of Bone Oncology
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2212137425000223
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Bone metastasis is a common and serious complication in lung cancer patients, leading to severe pain, pathological fractures, and reduced quality of life. Early prediction of bone metastasis can enable timely interventions and improve patient outcomes. In this study, we developed a multimodal Swin Transformer-based deep learning model for predicting bone metastasis risk in lung cancer patients by integrating CT imaging and pathological data. A total of 215 patients with confirmed lung cancer diagnoses, including those with and without bone metastasis, were included. The model was designed to process high-resolution CT images and digitized histopathological images, with the features extracted independently by two Swin Transformer networks. These features were then fused using decision-level fusion techniques to improve classification accuracy. The Swin-Dual Fusion Model achieved superior performance compared to single-modality models and conventional architectures such as ResNet50, with an AUC of 0.966 on the test data and 0.967 on the training data. This integrated model demonstrated high accuracy, sensitivity, and specificity, making it a promising tool for clinical application in predicting bone metastasis risk. The study emphasizes the potential of transformer-based models to revolutionize bone oncology through advanced multimodal analysis and early prediction of metastasis, ultimately improving patient care and treatment outcomes.
ISSN:2212-1374