Diagnostic Performance of Artificial Intelligence Models in Differentiating Bacterial and Viral Pneumonia: A Meta-analysis

Background: Accurate differentiation between bacterial and viral pneumonia is crucial for appropriate antimicrobial stewardship and clinical management. While artificial intelligence (AI) models, including deep learning and radiomics, have shown promise in medical imaging, their diagnostic performance varies across studies. Objective: The primary objective of this systematic review and meta-analysis is to evaluate the overall diagnostic accuracy of AI models in distinguishing bacterial from viral pneumonia based on medical imaging (e.g., CXR, CT). Methods: We will conduct a comprehensive search in databases including PubMed, Embase, Web of Science, and the Cochrane Library. Studies reporting diagnostic performance metrics (sensitivity, specificity, AUC) of AI models for this specific differential diagnosis will be included. Quality assessment will be performed using the QUADAS-2 tool. Data synthesis will be conducted using bivariate random-effects models to generate pooled sensitivity, specificity, and summary receiver operating characteristic (SROC) curves. Expected Results ： This study aims to provide high-quality evidence regarding the potential of AI as a non-invasive adjunctive tool for differentiating pneumonia pathogens, potentially reducing the misuse of antibiotics and improving diagnostic efficiency in clinical practice.