Medical imaging-based artificial intelligence in pneumonia: A narrative review

Pneumonia is a pulmonary inflammation caused by various pathogens and remains one of the leading causes of global mortality. In clinic, chest X-ray (CXR), computed tomography (CT), and lung ultrasound (LUS) have been the most commonly used imaging techniques for patients with pneumonia. Over the past decade, advances in artificial intelligence (AI) have sparked increasing interests in computer-aided diagnosis (CAD) system development aimed at improving clinical workflow efficiency and promoting a more comprehensive understanding of medical imaging information. The deep learning-based image analysis for pneumonia flourished during the pandemic of coronavirus disease 2019 (COVID-19), focusing on clinical problems include the development of rapid screening system for COVID-19 pneumonia, differential diagnosis, severity assessment, and predicting prognosis. In this narrative review, we firstly establish the general characteristics of imaging techniques used for pneumonia and outlined the typical manifestations used by clinicians in diagnosis and assessment. Then we provide a state-of-the-art review on the emerging applications in developing AI-based tools for pneumonia lesion detection and segmentation and auxiliary diagnosis. Efforts by researchers in combining multimodal imaging data and investigating correlations between different imaging modalities are also presented. Finally, we discussed the challenges in acquiring high-quality data, utilizing multimodal data, and promoting clinical adoption that must be overcome.