README.md

metadata

license: cc-by-nc-sa-4.0
extra_gated_fields:
  Name: text
  Email: text
  Country: country
  Organization or Affiliation: text
  What do you intend to use the model for?:
    type: select
    options:
      - Research
      - Personal use
      - Creative Professional
      - Startup
      - Enterprise

Dataset Card: CT-ScanGaze

Dataset Name: phamtrongthang/CT-ScanGaze
Repository: UARK‑AICV/CTScanGaze
License: CC BY-SA 4.0

---

1. Dataset Summary

CT-ScanGaze is the first publicly available dataset that captures 3D eye-gaze trajectories of expert radiologists while interpreting volumetric CT scans. It includes:

• 1045 abdominal and chest CT scans (.nii.gz format)
• Expert radiologist 3D gaze sequences (x, y, z with timestamp and duration)
• Transcribed radiology reports

This dataset supports research in gaze modeling, scanpath prediction, explainable AI, and multi-modal learning in medical imaging.

🏅 This work was acceppted as a highlight paper at ICCV 2025.

---

2. Dataset Structure

Attribute	Description
Total Volumes	1045 CT scans
Modality	CT (.nii.gz)
Gaze Data	3D coordinates + fixation duration
Reports	Transcribed from audio recordings
Radiologists	Two experts (10+ years of experience)

---

3. Intended Uses

• 3D scanpath prediction
• Gaze-informed diagnosis modeling
• Medical report generation
• Visual search behavior analysis
• Explainability in medical AI

---

4. Tasks and Benchmarks

Primary Task: 3D Scanpath Prediction

• Baseline: CT‑Searcher (transformer-based model)
• Evaluation Metrics:
  • Scanpath: ScanMatch, MultiMatch, SED
  • Saliency: CC, KL Divergence, NSS

Other potential tasks:

• Report–gaze alignment
• Gaze-informed classification
• Multi-modal fusion (gaze + imaging + text)

---

5. Data Availability

The processed dataset (.nii.gz, 3D gaze, and reports) is publicly available via Hugging Face.
Raw data, including original DICOM files and video recordings of the interpretation sessions, can be provided upon request for approved research purposes.

👉 To request access to the raw data, please contact: [email protected]

---

6. Citation

Please cite this dataset using the following BibTeX entry:

@article{pham2025ct,
  title={CT-ScanGaze: A Dataset and Baselines for 3D Volumetric Scanpath Modeling},
  author={Pham, Trong-Thang and Awasthi, Akash and Khan, Saba and Marti, Esteban Duran and Nguyen, Tien-Phat and Vo, Khoa and Tran, Minh and Nguyen, Ngoc Son and Van, Cuong Tran and Ikebe, Yuki and others},
  journal={arXiv preprint arXiv:2507.12591},
  year={2025}
}