Enhanced X-ray image classification for pneumonia detection using deep learning based CBAM and SE mechanisms

Pneumonia, a global health concern, remains a significant cause of morbidity and mortality, particularly in children under five and the elderly. Diagnostic challenges are pronounced in resource-limited settings, where expertise in radiological interpretation is scarce. X-ray imaging, a common diagnostic tool, often fails to provide accurate results without expert analysis. This gap in timely and precise diagnosis leads to delayed treatments and worsening patient outcomes. The emergence of antibiotic-resistant strains further emphasizes the urgency for innovative diagnostic solutions. This research integrates advanced attention mechanisms into convolutional neural networks (CNNs) to enhance pneumonia detection from X-ray images. Utilizing a dataset of 5,816 X-rays, preprocessing steps included normalization and data augmentation to improve robustness. The baseline CNN model was augmented with Convolutional Block Attention Module (CBAM) and Squeeze-and-Excitation (SE) networks, which prioritize critical image regions and recalibrate feature channels. Comparative evaluations were conducted using ResNet50 combined with CBAM. The CBAM-enhanced CNN achieved 98.6% accuracy, improving upon the baseline CNN’s 92.08%, with a sensitivity of 98.3% and specificity of 97.9%. The SE-integrated CNN followed with 96.25% accuracy, demonstrating superior feature recalibration. ResNet50 with CBAM attained 93.32% accuracy. Compared to standard CNN models, these models exhibited reduced overfitting, improved generalization, and enhanced feature extraction. The proposed approach ensures a higher precision rate in detecting pneumonia from X-ray images. The model is designed for real-world clinical applications, particularly in low-resource healthcare settings. A lightweight, user-friendly web application was developed to assist radiologists and general practitioners in automated pneumonia detection, reducing reliance on expert interpretation. • The study achieves 98.6% accuracy in pneumonia detection using advanced attention mechanisms integrated into convolutional neural networks. • Attention modules like the Convolutional Block Attention Module (CBAM) and Squeeze-and-Excitation (SE) enhance feature extraction, significantly improving pneumonia detection by focusing on critical image regions. • Enhanced models outperform baseline CNNs and other methods in pneumonia detection, excelling in accuracy, precision, recall, and F1-score, demonstrating robustness and generalization capabilities. • Computational optimizations ensure lightweight models for pneumonia detection, making them suitable for deployment in low-resource environments and addressing practical challenges in medical imaging applications.