Transformer-Based Chest X-ray Report Generation Model

In radiological testing for treatment and diagnosis, chest X-rays and medical radiographs are most frequently utilized in clinical practice. A skilled radiologists may require some patience to read and interpret a chest X-ray; less experienced radiologist may find it more challenging; and the common person might find it incredibly difficult. A deep learning-based X-ray report generating system can help doctors with reporting tasks and give radiologists judgment support. In order to generate chest X-ray reports, this paper proposes a deep learning-oriented model that makes use of a transformer method and transfer learning technique. We use a chexnet model that has already been pre-trained and will eventually be trained on X-ray images for visual extraction. Then, an encoder-decoder for a modified transformer is used. We train and evaluate our network using the IU X-ray dataset, one of the most extensive publicly available datasets currently utilized for chest X-ray image captioning.