A Chest X-ray Pneumonia Recognition Model Based on Quality Perception

Abstract

. Variations in chest X-ray quality commonly lead to model performance degradation in clinical settings. To address this problem, this paper introduces the Quality-Aware Pneumonia Recognition (QAPN) model. Built on a ConvNeXt backbone, QAPN incorporates a lightweight quality-assessment branch to characterize image degradation and employs a Quality-Guided Attention (QGA) mechanism to adaptively reweight lesion features. Experimental results show that QAPN outperforms all baseline models by 2.7 – 15.0 percentage points in terms of F1 score under various low-quality conditions, including image noise and improper exposure. More importantly, the model automatically reduces its prediction confidence for poor-quality images, thereby improving the robustness and safety of computer-aided diagnosis in complex clinical environments.