A Systematic Literature Review of Deep Learning-Based Disease Detection and Classification for Chest X-Ray Imaging

Abstrak

Conventional chest X-ray (CXR) interpretation is often constrained by inter-observer variability, high workload, and time-consuming diagnostic processes. This study aims to consolidate and analyze recent research trends, methodologies, dataset characteristics, and performance outcomes of deep learning (DL) in CXR-based disease detection published between 2021 and 2025. The methodology employs a Systematic Literature Review (SLR), involving research question formulation, comprehensive database searches, and study selection based on predefined inclusion criteria. Results indicate that CNN-based and transfer learning approaches dominate the field, with a significant shift toward multi-disease screening frameworks and the adoption of hybrid or lightweight architectures. The discussion highlights that while models achieve high accuracy, substantial variability in datasets and evaluation protocols hinders direct comparison and clinical generalizability. In conclusion, deep learning has become the prevailing methodology for CXR analysis, but establishing standardized evaluation frameworks and diverse clinical datasets is essential to bridge the gap between methodological development and real-world clinical implementation, and to provide representative quantitative comparisons that highlight performance variability across model architectures and disease scopes.