Transformers in Medical Domain: Swin Transformer as a Binary Classification Model for Pneumonia

Pneumonia disease is a significant worldwide health problem, where accurate and timely diagnosis is crucial for effective treatment. Recently, transformer-based models have shown increasing interest in various domains including natural language processing and computer vision. In this study, we have proposed to use Swin Transformer model, a state-of-the-art model for developing a binary classification model for pneumonia detection using medical chest x-ray images. The proposed model uses the self-attention approach to understand global and local features in the images which leads to enhanced feature representation. The proposed model is also helpful to learn hierarchical representations which improves the accuracy and robustness of pneumonia classification resulting into more accurate, timely diagnosis and intervention. Furthermore, to evaluate the performance of the proposed model we compared its performance results with the EfficientNetB0 model by using traditional performance evaluation metrics such as precision, recall, Area-Under-the Curve (AUC), etc. The dataset used for this study is publicly available dataset having chest x-ray images labelled as normal or pneumonia. The results from our proposed approach shows the promising ability of capturing efficient features leading to accurate and reliable pneunomia classification.