Shamsiddin Kamoliddinov

Deep Learning-Based Model for Pneumonia Detection: Analysis of Chest X-Ray Images

Pneumonia, an acute inflammation of the lung’s air sacs, remains a leading cause of mortality worldwide, claiming over 700,000 children under 5 annually according to UNICEF and 2.5 million adults as reported by the Global Burden of Disease in 2019. In resource-limited settings like Uzbekistan, early detection via chest X-ray analysis is hindered by a shortage of radiologists, necessitating innovative solutions. This study presents a deep learning-based model for automated pneumonia detection, utilizing the Kaggle Chest X-Ray Pneumonia dataset comprising 5863 images (5216 training, 624 validation) with a 1:2 class imbalance. The proposed approach leverages the pre-trained ResNet50 architecture with transfer learning, avoiding fine-tuning to prevent overfitting, a common challenge in medical imaging. To address class imbalance and enhance generalization, data augmentation techniques—including rotation (20°), width/height shift (0.2), horizontal flip, shear (0.2), and zoom (0.2)—were appl