Improving Pneumonia Classification and Lesion Detection Using Spatial Attention Superposition and Multilayer Feature Fusion

Pneumonia is a severe inflammation of the lung that could cause serious complications. Chest X-rays (CXRs) are commonly used to make diagnosis pneumonia. In this paper, we propose deep-learning-based method with spatial attention superposition (SAS) and multilayer feature fusion (MFF) facilitate pneumonia based on CXRs. Specifically, an SAS module, which takes advantage channel mechanisms, was designed identify intrinsic imaging features pneumonia-related lesions their locations, MFF module harmonize disparate from different channels emphasize important information. These two modules were concatenated extract critical image serving as basis for diagnosis. We further embedded proposed into baseline neural network developed model called SAS-MFF-YOLO diagnose To validate effectiveness our model, extensive experiments conducted CXR datasets provided by Radiological Society North America (RSNA) AI Research Institute. achieved precision 88.1%, recall 98.2% classification AP50 99% lesion detection Institute dataset. The visualization intermediate maps showed uncovering in Our results demonstrated approach be enhance performance overall imaging.