Systematic Literature Review of Transfer Learning for Pneumonia Classification in Chest X-Rays

Abstrak

Diagnosis of pneumonia through manual interpretation of Chest X-Ray (CXR) images is often hampered by observer subjectivity and radiologist fatigue, which can potentially lead to misdiagnosis. This study aims to evaluate the effectiveness and development trends of Transfer Learning techniques, particularly the ResNet, VGG, and DenseNet architectures, in pneumonia classification through the Systematic Literature Review (SLR) method. In accordance with the PRISMA protocol, the search was conducted in the Scopus database from 2021 to 2025, yielding 76 articles that met the inclusion criteria. Bibliometric analysis shows that the publication trend, initially triggered by the urgency of the pandemic, has now shifted to a phase of technological maturity, with a focus on integrating Explainable AI (XAI) to address black-box problems. Geographically, research contributions are dominated by institutions in Asia and the Middle East. The main findings confirm that Transfer Learning can significantly improve diagnostic accuracy and initial screening efficiency compared to conventional methods. However, challenges such as data imbalance and the need for clinical validation remain obstacles. This study concludes that the future of computer-assisted diagnosis systems depends on improving model transparency to support precise and reliable Clinical Decision Support Systems (CDSS).