SZ-HospCXR dataset in YOLO format

The original "Shenzhen Hospital CXR" dataset (abbreviated as SZ-HospCXR, from the article: Yang, F.; Lu, P.X.; Deng, M.; Wáng, Y.X.J.; Rajaraman, S.; Xue, Z.; Folio, L.R.; Antani, S.K.; Jaeger, S. Annotations of Lung Abnormalities in the Shenzhen Chest X-ray Dataset for Computer-Aided Screening of Pulmonary Diseases. Data 2022, 7, 95. DOI: mdpi.com/2306-5729/7/7/95) was jointly constructed by the U.S. National Library of Medicine, the Third People’s Hospital of Shenzhen, China, and Guangdong Medical University.

It is a digital imaging resource focused on tuberculosis diagnosis, released under the CC-BY 4.0 license. The images are stored in PNG format with a resolution of approximately 3000×3000 pixels. The annotation type is mask, covering 19 categories of abnormalities including contraction, calcified nodule, miliary lesion, lymph node enlargement, apical thickening, interlobar fissure thickening, among others.We preprocessed this dataset as an external set to validate our own model: first, the mask annotations were converted into bounding box format and adjusted to the YOLO data format; second, based on medical definitions, the original labels of SZ-HospCXR were mapped and integrated into the label system used in this study.