A Scalable Solution for Pneumonia Diagnosis: Transfer Learning for Chest X-ray Analysis

The high death rate and rapid spread of pneumonia make it a serious global health concern. The precision of traditional diagnostic techniques is often compromised, particularly in resource-limited settings. Existing studies have predominantly focused on either binary or multi-class classification, often relying on limited datasets, which affects the generalizability and robustness of models. Additionally, the interpretability and transparency of deep learning models remain challenging in clinical environments. Addressing these challenges is crucial for advancing diagnostic methodologies to combat pneumonia. This study explores the use of transfer learning to improve the reliability of pneumonia classification using chest X-ray images. An automated classification system was developed, utilizing a combination of image preprocessing techniques, feature extraction with GLRLM, and various machine learning and transfer learning algorithms. Two datasets were employed: the GWCMCx dataset with 5856 chest X-rays and another dataset comprising 9207 chest X-rays. The system was designed for both binary and multi-class classification scenarios. CapsuleNet emerged as the most effective model, achieving an accuracy of ${9 9. 0 2 \%}$ in binary classification and $90.43 \%$ in multi-class classification, significantly outperforming previous studies. Additionally, a user-friendly web application was developed to facilitate the integration of the classification system into clinical workflows, accelerating clinical decision-making. This study demonstrates the potential of transfer learning to automate and improve pneumonia diagnosis, offering a scalable solution that can alleviate radiologists’ workload and improve patient outcomes.