Optimized pneumonia detection using deep learning: a ResNet50-based model with multiscale efficient spatial attention

Pneumonia, an acute respiratory infection that inflames the air sacs in one or both lungs, remains a significant global health threat, contributing to millions of hospitalizations and deaths annually. In the process of diagnosing pneumonia, the subtle radiographic changes, especially in the early stages, are often missed by conventional methods, leading to delayed diagnosis and treatment. To address this issue, we propose a deep learning-based approach aimed at improving pneumonia classification accuracy from chest X-ray images. The proposed model leverages the ResNet50 architecture, augmented with the Multi-Scale Efficient Spatial Attention (MESA) module, to enhance feature extraction and classification performance. Moreover, the model is trained using Class Balanced Focal Loss (CBFL) to mitigate the effects of class imbalance in the dataset. Experimental results show that the proposed model outperforms several recently published methods in pneumonia classification. It achieves accuracy, precision, sensitivity, specificity, and F1-score of 0.984, 1.0, 0.9744, 1.0, and 0.987, respectively. By leveraging deep learning techniques, this model could assist radiologists in making more accurate and timely diagnoses, ultimately improving patient outcomes and reducing healthcare costs.