Review article
https://doi.org/10.37797/ig.43.3.3
Artificial Intelligence in Radiologic Diagnosis of Pulmonary Infection: a PRISMA-Guided Systematic Review
Fran Posavec
; Health Center Zagreb – East, Zagreb, Croatia
Dušica Kasun
; Faculty of Medicine, Catholic University of Croatia, Zagreb, Croatia
Alemka Markotić
; Faculty of Medicine, Catholic University of Croatia, Zagreb, Croatia; University Hospital for Infectious Diseases "Dr. Fran Mihaljević", Zagreb, Croatia
Petra Margetić
; Faculty of Medicine, Catholic University of Croatia, Zagreb, Croatia; Clinical Hospital Center „Sestre milosrdnice“, Zagreb, Croatia
Klaudija Višković
; Faculty of Medicine, Catholic University of Croatia, Zagreb, Croatia; University Hospital for Infectious Diseases "Dr. Fran Mihaljević", Zagreb, Croatia
*
* Corresponding author.
Abstract
Aim: Pulmonary infections, including bacterial and viral pneumonia and tuberculosis (TB), remain major global health concerns. Artificial intelligence (AI), particularly deep learning, has shown strong potential for improving radiologic diagnosis using chest X-rays (CXR) and computed tomography (CT). The aim of this review article is to analyze recent literature on the achievements of artificial intelligence (AI) in the radiological interpretation of imaging studies in patients with pneumonia and tuberculosis.
Methods: A review that was conducted according to PRISMA 2020 guidelines, analysed English-language studies published between January 1st 2023 and December 31st 2025 that evaluated AI-based tools for detecting pulmonary infections in adults.
Results: A total of 182 studies were identified, and 20 met inclusion criteria (13 pneumonia and 7 TB studies). Most studies used convolutional neural networks (CNN) trained on large imaging datasets. For pneumonia detection, AI models achieved sensitivities of 80–97% and specificities of 85–95%, with pooled estimates 88% and 90%, respectively. TB models showed high performance, with sensitivities up to 99%.
Conclusion: AI demonstrates high diagnostic accuracy and potential clinical utility in thoracic imaging, but external validation and prospective studies are still needed for routine implementation.
Keywords
Artificial intelligence; Deep Learning; Pneumonia; Tuberculosis; Chest X-ray; Computer Tomography
Hrčak ID:
352445
URI
Publication date:
19.10.2026.
Visits: 0 *