Automatic Chest X-ray Image Caption for Report Generation

Accurate medical report generation is a time-consuming task. Therefore, automating chest X-ray (CXR) image captions for precise medical report generation is critical to improving diagnostic efficiency and accuracy and reducing report generation time through deep learning-based methods. One major difficulty of this task is to accurately capture the abnormality of the diagnostic region in images and generate professional descriptive language. In this paper, we propose the ResGen model, which leverages the encoder-decoder architecture, using ResNet50 as the image encoder and GPT-2 as the text decoder. The model is trained on the open-access IU-Xray dataset. The extensive experiment results showed that our methods achieved state-of-the-art performance in automatic medical report generation tasks.