Binary-SE-ResNet Model Research for Binary Classification of Chest X-Ray

In 2019, the outbreak of the novel coronavirus swept across the globe. With the increase in the number of infections, the demand for automated pneumonia detection gradually increased. This study mainly focused on the binary classification problem of pneumonia images and proposed a Binary-SE-ResNet that utilizes both the original image features and LBP image features extracted through the SE-ResNet34 model. The extracted features are then concatenated and input into an XGBoost model, which is optimized by a Bayesian optimizer to select the best model for the final classification. Considering the characteristics of the dataset, this study employed CLAHE for image enhancement during the preprocessing stage to improve contrast and noise reduction. Ultimately, the proposed model achieved an accuracy of 97.7% in the binary classification of pneumonia, which is a 0.9% improvement compared to the best single model. This demonstrates that the approach of extracting features from different feature maps and utilizing ensemble learning can better capture features and achieve improved classification performance.