NSCLCR-HAID

Pre-processed NSCLC-Radiomics Dataset Resampled CT Scans with 3D Bounding Box Annotations and Clinical Metadata Description This dataset provides a pre-processed version of the NSCLC-Radiomics collection, consisting of standardized CT images, 3D bounding box annotations derived from expert tumor segmentations, and harmonized clinical metadata. The original NSCLC-Radiomics dataset includes 422 patients diagnosed with non-small cell lung cancer (NSCLC) and was released through The Cancer Imaging Archive (TCIA). The goal of this release is to lower the barrier for reproducible medical AI research by offering AI-ready imaging data with consistent spatial resolution, unified annotation formats, and predefined train/validation/test splits. What’s Included 1. Resampled CT Images (421 patients) Modality: Computed Tomography (CT) Format: NIfTI (.nii.gz) Standardized voxel spacing: 0.703125 × 0.703125 × 1.25 mm³ File naming convention: NSCLC_{PatientID}_0000.nii.gz 2. 3D Bounding Box Annotations One bounding box per patient corresponding to the primary tumor (GTV-1) Derived from expert manual segmentations Represented in world coordinates (center location and box dimensions in millimeters) Provided in CSV format for direct use in detection and localization tasks 3. Clinical Metadata and Dataset Splits Demographic information: age, sex Clinical variables: TNM staging Histological subtype (adenocarcinoma, squamous cell carcinoma, large cell carcinoma, NOS, unknown) Stratified dataset splits: Training: 80% (337 patients) Validation: 10% (42 patients) Test: 10% (42 patients) 4. Dataset Descriptors JSON descriptor files for: 3D object detection tasks Generative and synthesis-based modeling Annotation-level CSV files for classification and radiomics studies Processing Pipeline All CT scans were resampled from heterogeneous native voxel spacings to a uniform resolution of 0.703125 × 0.703125 × 1.25 mm³ using B-spline interpolation. Primary tumor bounding boxes were extracted from expert-annotated GTV-1 segmentations and converted to world-coordinate representations. Clinical metadata were merged with imaging annotations and used to generate reproducible, stratified train/validation/test splits based on histological subtype. Complete preprocessing scripts and documentation are available in the associated GitHub repository. Intended Use This dataset is intended for research and educational purposes only, including: 3D medical image object detection Lung cancer classification by histological subtype Radiomics feature extraction and analysis CT image synthesis and generative modeling Multi-task and multimodal medical AI research This resource is not intended for clinical diagnosis or decision-making. Dataset Statistics Total patients: 421 Imaging modality: CT Annotation type: 3D bounding boxes + clinical labels Training set: 337 patients (80%) Validation set: 42 patients (10%) Test set: 42 patients (10%) File Structure ├── NSCLC-Radiomics-NIFTI.zip/ # Converted NIfTI images │ ├── LUNG1-001/ │ ├── LUNG1-002/ │ └── ... ├── NSCLC-Radiomics-ct.zip/ # Original DICOM CONVERTED TO NIFTI CT scans ├── NSCLC-Radiomics-mask.zip/ # Original DICOM CONVERTED TO NIFTI segmentations ├── vista3Dauto_seg.zip/ # VISTA3D organ Segmentation NIFTI segmentations├── vista3Dauto_seg_plus_orgGTV.zip/ # VISTA3D organ Segmentation +NSCLC segmentation of abnormalities NIFTI segmentations Citation If you use this dataset, please cite both the original NSCLC-Radiomics dataset and publication, as well as this pre-processed version. Original DatasetAerts, H. J. W. L., et al. (2015).Data From NSCLC-Radiomics.The Cancer Imaging Archive.DOI: 10.7937/K9/TCIA.2015.PF0M9REI Original PublicationAerts, H. J. W. L., et al. (2014).Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach.Nature Communications, 5, 4006.DOI: 10.1038/ncomms5006 This Pre-processed Dataset[Your Name / Lab]. (2026).Pre-processed NSCLC-Radiomics Dataset with 3D Bounding Box Annotations.Zenodo.DOI: [Zenodo-assigned DOI] Related Links Original NSCLC-Radiomics collection:https://www.cancerimagingarchive.net/collection/nsclc-radiomics/ Processing code and documentation:https://github.com/fitushar/HAID Original publication:https://doi.org/10.1038/ncomms5006 License This dataset inherits the Creative Commons Attribution 3.0 Unported (CC BY 3.0) license from the original NSCLC-Radiomics collection. All data are de-identified in accordance with HIPAA standards. Keywords lung cancer, NSCLC, CT imaging, medical imaging, radiomics, object detection, 3D bounding boxes, deep learning, health AI, clinical datasets