NLST-3D+ Annotation

With the National Lung Screening Trial (NLST), for detection evaluation, we utilized open-access annotations provided by [Mikhael et al.(2023)]. We converted over 9,000 2D slice-level bounding box annotations from more than 900 lung cancer patients into 3D representations, resulting in over 1,100 nodule annotations. To extract 3D annotations from the 2D annotations, we first verified the 2D annotations within the DICOM images. Then, we extracted the `seriesinstanceuid`, `slice_location`, and `slice_number` from the DICOM headers. Subsequently, the image coordinate locations were converted to world coordinates. After verifying these annotations in the corresponding NIFTI images, we concatenated overlapping consecutive 2D annotations of the same lesion across multiple slices into a single 3D annotation. The complete code for generating the 3D annotations, along with a visualization script to display these annotations, will be released soon. A preview of the visualization is shown in this Jupyter Notebook: https://github.com/fitushar/AI-in-Lung-Health-Benchmarking-Detection-and-Diagnostic-Models-Across-Multiple-CT-Scan-Datasets/blob/main/NLST_Data_Annotations/3D_Annotation_Visualizations_NLST.ipynb Code Repository:To support reproducible open-access research and benchmarking, we have shared several pre-trained models and baseline results in a GitHub and GitLab repository. GitLab: https://gitlab.oit.duke.edu/cvit-public/ai_lung_health_benchmarkingGitHub: https://github.com/fitushar/AI-in-Lung-Health-Benchmarking-Detection-and-Diagnostic-Models-Across-Multiple-CT-Scan-Datasets