Extending Chest X-Ray with Multi Label Disease Sequence

Doctors need to spend a lot of time and energy on reading the patient’s medical images to write the corresponding diagnosis report. Therefore, it is interesting to use the artificial intelligence technology on the research of automatic generation of medical imaging reports. In order to improve the accuracy of the report generation, we extend the Chest X-Ray dataset with multi label of disease combining the medical characteristics of imaging reports. In this paper, we give the concrete steps of expanding the dataset. The original dataset contains 8121 chest X-ray images from different perspectives and 3996 diagnostic reports. Based on the original dataset, we pick out the complete data pair and use a multi label classification network named CheXNet to expand the dataset with the multi label disease sequence. The multi disease label sequence includes the judgment results of 14 common thoracic diseases. After manual correction, we get a dataset which contains 3119 multimodal data pairs. Each data pair is composed of two chest X-ray images and corresponding diagnostic reports, as well as a binary multi label sequence. The extended dataset is more useful for the task of medical diagnostic reports generation.