Pneumothorax is a common medical emergency defined as the abnormal collection of air in the pleural space between the lung and chest wall. Its typical symptoms include chest pain and dyspnea, leading to oxygen deficiency or even life-threatening in severe cases. Therefore, an efficient and automatic pneumothorax diagnosis algorithm would be useful in many clinical scenarios. Recently, deep learning methods have achieved impressive progress in medical image segmentation tasks. However, large-scale dataset is one of the critical components for the success of deep learning. On the other hand, there are few public chest X-ray images with pneumothorax. To stimulate the researchers' interest in pneumothorax diagnosis algorithm, we released a new data set PTX-227 here, which contains 227 chest X-ray images of pneumothorax with pixel-level annotations. All images were resized to 1024×1024. The raw image intensity was clipped within the range from 2.5th to 97.5th percentile and then normalized to 0 to 255. The contours of the pneumothorax area were labelled by two senior radiologists using ITK-SNAP. The dataset was anonymized and every record related to patients' privacy was removed. Only the image data and the corresponding labels were included in PTX-227. The paper is still under review and the citation is temporarily unavailable. The citation will be updated once the paper is published.

See also: https://github.com/wangyunpengbio/DeepSDM