Exploring Deep Learning Models for Pneumonia Classification in Chest Radiological Images: A Systematic Review

Pneumonia continues to be a significant global health issue, with timely and precise diagnosis playing a vital role in patient care. Traditional diagnostic approaches relying on chest radiological images often encounter limitations such as inconsistent interpretations among observers and delays in analysis. To overcome these challenges, the use of deep learning models has emerged as a promising approach for achieving automated and accurate pneumonia detection. This systematic review seeks to deliver a comprehensive summary of recent progress in deep learning applications for pneumonia classification using chest imaging. The review adds value by examining the evolution of deep learning architectures, summarizing widely used datasets, highlighting current challenges, and suggesting directions for future research. A systematic search was carried out across several scientific databases, including ScienceDirect and IEEE Xplore, covering studies published between 2022 and 2024. The studies were chosen according to established inclusion and exclusion criteria, followed by content-based screening to maintain relevance. This review encompasses 36 studies featuring a range of deep learning models, such as CNN, transfer learning techniques (VGG16, ResNet, DenseNet, MobileNet, EfficientNet), hybrid models, ensemble methods, attention-based mechanisms, domain adaptation frameworks, and federated learning approaches. Diverse publicly available datasets, including ChestXRay2017, Guangzhou Medical Center, RSNA, and Covid-19 Radiography Dataset, were widely utilized. Preprocessing techniques such as resizing, normalization, data augmentation (including GAN-based), and segmentation were frequently applied to enhance model performance. Reported classification accuracies ranged from 78.9% to over 99%, with ensemble and hybrid models often achieving superior performance. Nevertheless, challenges such as class imbalance, domain generalization, computational complexity, and clinical interpretability persist. In conclusion, deep learning demonstrates significant potential in improving pneumonia diagnosis through chest radiological image analysis. However, addressing current limitations and enhancing clinical integration remain critical for future advancements in this field.