An attention-based deep learning network for lung nodule malignancy discrimination

Gang Liu; Fei Liu; Jun Gu; Xu Mao; XiaoTing Xie; Jingyao Sang

doi:10.3389/fnins.2022.1106937

An attention-based deep learning network for lung nodule malignancy discrimination

Front Neurosci. 2023 Jan 9:16:1106937. doi: 10.3389/fnins.2022.1106937. eCollection 2022.

Authors

Gang Liu¹, Fei Liu¹, Jun Gu², Xu Mao¹, XiaoTing Xie¹, Jingyao Sang¹

Affiliations

¹ Department of Interventional Radiology, Qinghai Red Cross Hospital, Xining, Qinghai, China.
² School of Computer Science and Technology, Department of Telecommunications, Xi'an Jiaotong University, Xi'an, China.

Abstract

Introduction: Effective classification of lung cancers plays a vital role in lung tumor diagnosis and subsequent treatments. However, classification of benign and malignant lung nodules remains inaccurate.

Methods: This study proposes a novel multimodal attention-based 3D convolutional neural network (CNN) which combines computed tomography (CT) imaging features and clinical information to classify benign and malignant nodules.

Results: An average diagnostic sensitivity of 96.2% for malignant nodules and an average accuracy of 81.6% for classification of benign and malignant nodules were achieved in our algorithm, exceeding results achieved from traditional ResNet network (sensitivity of 89% and accuracy of 80%) and VGG network (sensitivity of 78% and accuracy of 73.1%).

Discussion: The proposed deep learning (DL) model could effectively distinguish benign and malignant nodules with higher precision.

Keywords: artificial intelligence; attention mechanism gate module; lung nodules; malignancy; multimodal.

Grants and funding

This work was supported by Qinghai Province Basic Research Plan—Applied Basic Research Project (2020-ZJ-781).