Coursework of CIS522 Deep Learning: Identify Pneumonia in Chest X-Rays with Machine Learning methods
Pneumonia is an infection of one or both of the lungs caused by bacteria, viruses, or fungi. In medical image diagnosis, pneumonia often presents itself as increased opacity in chest X-Rays. In order to improve the efficiency and accuracy of identifying pneumonia from chest X-Rays, various models have been proposed for this task. In this work, we examine multiple non-deep and deep learning models for classifying chest X-Ray images. We first feed pre-processed data to a family of traditional machine learning models, including SVM, Random Forest, Gradient Boosting, etc. Then, we compare the results of several deep learning models, which include a self-designed CNN, AlexNet, and ResNet18. While most previous work focuses on CNN-based models, we introduce Swin Transformer, which is a state-of-the-art computer vision model that leverages ideas like patch embedding. Statistical results obtained demonstrate that Swin Transformer can generate significantly better classification results for chest X-Rays than CNN-based models.