README.md

metadata

title: RADAR Abdominal CT
emoji: 🩻
colorFrom: blue
colorTo: green
sdk: gradio
app_file: app.py
pinned: false
license: cc-by-nc-sa-4.0
python_version: '3.12'
sdk_version: 5.49.1
preload_from_hub:
  - >-
    radar-generalist/RADAR
    checkpoint_radar_pretrain.pth,bert-base-chinese/config.json,bert-base-chinese/config_decoder.json,bert-base-chinese/pytorch_model.bin,bert-base-chinese/tokenizer.json,bert-base-chinese/tokenizer_config.json,bert-base-chinese/vocab.txt

RADAR abdominal-CT findings demo: upload a contrast-enhanced abdominal CT as a single volume (.nii / .nii.gz, .nrrd, .mha) or a DICOM series (extensionless or .dcm files, or a .zip, nested folders OK); get top-finding scores plus a table of all 146 per-finding scores.

• Paper/model: radar-generalist/RADAR (CC BY-NC-SA 4.0, non-commercial research use only)
• Source: vendored inference code from the RADAR release (Zenodo damo-radar.zip, record 21271172); weights load at runtime from the public HuggingFace repo, no token needed
• Intake: everything normalizes to one .nii.gz via SimpleITK (no new deps); header+raw pairs (.mhd/.nhdr) need converting first. Multi-file uploads are treated as one DICOM series (largest series wins, ≥8 slices); a lone .dcm is rejected as one slice, not a volume
• Runtime: ZeroGPU (large); CPU staging/conversion runs quota-free, then one @spaces.GPU(duration=90) scoring call per case. Preprocessing (1x1x5mm resample, 96x256x384 pad/crop) is the upstream MONAI chain, reused verbatim (transformers==4.48.3; v5's loader crashes on this 2022-vintage BERT fork)

Research assistance tool, not a medical diagnosis.