Comparison of deep learning models based on Chest X-ray image classification

Abstract

Pneumonia is a common respiratory disease characterized by inflammation in the lungs, emphasizing the importance of accurate diagnosis and timely treatment. Despite some progress in medical image segmentation, overfitting and low efficiency have been observed in practical applications. This paper aims to leverage image data augmentation methods to mitigate overfitting and achieve lightweight and highly accurate automatic detection of lung infections in X-ray images. We trained three models, namely VGG16, MobileNetV2, and InceptionV3, using both augmented and unaugmented image datasets. Comparative results demonstrate that the augmented VGG16 model (VGG16-Augmentation) achieves an average accuracy of 96.8%. While the accuracy of MobileNetV2-Augmentation is slightly lower than that of VGG16-Augmentation, it still achieves an average prediction accuracy of 94.2% and the number of model parameters is only 1/9 of VGG16-augmentation. This is particularly beneficial for rapid screening of pneumonia patients and more efficient real-time detection scenarios. Through this study, we showcase the potential application of image data augmentation methods in pneumonia detection and provide performance comparisons among different models. These findings offer valuable insights for the rapid diagnosis and screening of pneumonia patients and provide useful guidance for future research and the implementation of efficient real-time monitoring of lung conditions in practical healthcare settings.