A Lightweight Residual Network for Pneumonia Classification

Pneumonia, a global health disease, continues to threaten human life. In this challenge, accurate and rapid diagnosis not only contributes to treatment effectiveness, but also reduces the burden on the healthcare system. With the development of medical imaging technology, chest X-ray image has gradually become an important tool in the diagnosis of pneumonia. However, the surge in the number of pneumonia patients has challenged traditional diagnostic methods with time and resource constraints. Deep learning, particularly convolutional neural networks (CNNs), shows potential for problem solving in this context. However, the traditional CNN model has large parameters and is not suitable for resource-constrained devices. Therefore, this study proposes a lightweight residual network (LR-Net) that combines residual network (ResNet) and depth-separable convolution (DWConv) for early and automated detection of pneumonia in chest X-ray images. Focusing on the binary classification problem (pneumonia and normal), the model was tested on the ChestX-ray2017 dataset with an accuracy of 86.49%, model parameter of 268.098 K and FLOPs of 732.2184 M. This lightweight model greatly reduces the demand for computing resources and memory, making it ideal for mobile devices and edge computing devices. Taken together, the findings of this study demonstrate the effectiveness of lightweight deep learning models combining ResNet and DWConv in pneumonia diagnosis, providing clinicians with a powerful tool to enhance diagnostic insight into pneumonia. In addition, due to its lightweight nature, the model offers new possibilities for the rapid and accurate diagnosis of pneumonia in resource-constrained settings, helping to reduce the strain on global healthcare systems and provide patients with timely treatment.