Coronary artery calcification assessment in National Lung Screening Trial CT images (DeepCAC2)

Overview The dataset of AI derived CACS for the entire NLST dataset includes the segmentation of the heart and CAC, the computed CACS, and the derived risk for 26,228 subjects across a total of 127,776 CT scans. The segmentations are provided in DICOM SEG format, which contains both the heart and CAC segmentations predicted by our deep learning algorithm. The CACS and risk score are provided as key-value pairs in a complementary JSON file with three keys: “Predicted Agatston Score”, which stores the CACS value; “Predicted Risk Group”, which contains the classified risk; and “SeriesInstanceUID”, which links to the original CT scan series. The data is organized in a folder structure where the first-level folder is the patient ID, and the second-level folder is the image’s series instance UID, which contains the cac.seg.dcm segmentation file and the DeepCAC.report.json file. Files in this record: The dataset is compressed using Zstandard and has an uncompressed size of ~160GB. The sha256 checksum of the compressed file is 54a6da8a7c23ab3fa5f62107278805dd927825e66557839f76c2357848cb5955 An export of the archives files is available in `DeepCAC2_NLST.txt`. The `DeepCAC2_NLST_10_cases_example` folder provides 10 extracted examples, compressed as ZIP archive which some readers may find more convenient. The `DeepCAC2_NLST_scores.csv` file contains the aggreagated scores over all individual `DeepCAC.report.json` files.