Real-Time Multi-Class Pneumonia Detection in Chest X-Rays Using Optimized Deep Learning Methods

Pneumonia remains a global health crisis, with timely and accurate diagnosis being critical to reducing mortality. Traditional diagnostic methods, reliant on manual interpretation of chest X-rays, are often slow and prone to human error, particularly in resource-limited settings. While deep learning offers promising solutions, existing systems are limited by binary classification, insufficient subtype differentiation, and computational inefficiencies. This study addresses these gaps by developing an advanced deep learning framework for real-time, multi-class pneumonia detection. A balanced dataset of 8,250 chest X-ray images-spanning viral, bacterial, and COVID-19 pneumonia, alongside normal caseswas compiled to train and evaluate five models: VGG16, ResNet50, MobileNetV2, YOLOv4, and YOLOv6. Enhanced preprocessing techniques, including histogram equalization and Gaussian filtering, were employed to improve feature extraction. Among the models, YOLOv6 achieved exceptional performance, with 99.00 % accuracy and a rapid inference time of 18 ms per image, outperforming other architectures in both precision and computational efficiency. The study's key contribution lies in its demonstration of YOLOv6's superior capability for precise, real-time pneumonia subtype classification. By integrating this model into clinical workflows, healthcare systems can significantly enhance diagnostic speed and accuracy, enable targeted treatments and improve patient outcomes. The publicly available dataset and implementation code further support broader advancements in AI-driven medical diagnostics. This research not only bridges critical gaps in automated pneumonia detection but also sets a foundation for future innovations in global healthcare.