Abstract

Info

Title:: Chest X-ray image classification algorithm based on interactive attention and feature fusion

Author(s):: FAN Xingjia1; XING Xiongfei1; LI Rui2; 1. School of Medical Information Engineering , Gansu University of Chinese Medicine, Lanzhou 730000, China 2. Information Center, Lanzhou University Second Hospital, Lanzhou 730030, China

Keywords:: chest X-ray image deep learning deformable interactive attention multi-scale feature fusion

Abstract:: An improved ConvNeXt-based model for chest X-ray image classification that integrates interactive attention and feature fusion is proposed to address the issues of insufficient feature extraction and detailed information loss in the existing chest X-ray image classification algorithms. A deformable interactive attention module is introduced to synergistically enhance features with different attributes through dual-path feature decoupling and interactive modulation mechanisms. Meanwhile, a multi-scale feature fusion module is constructed to integrate discriminative information across different abstraction levels, thereby improving the models ability to retain fine-grained details. Additionally, a weighted focal loss function is adopted to compensate for the limitations of traditional cross-entropy loss. Experiments on the ChestX-Ray14 dataset demonstrate that the proposed method achieves an average AUC of 0.851 for the classification of 14 diseases, exhibiting competitive performance compared with existing deep learning models. The proposed approach can improve classification accuracy, and provide robust technical support for the auxiliary diagnosis of chest X-rays.