UM-IHC-CA2i/AMPLIFAI

@

Annotated Multi-Phase Liver Imaging for AI

Official training data repository for the AMPLIFAI Challenge — a MICCAI 2026 benchmark for clinically meaningful AI in liver lesion assessment on multi-phase CT.

Challenge links

Starter kit	Getting started guide → — register, access data, and submit your model
Challenge homepage	um-ihc-ca2i.github.io/amplifai-challenge
Sample Submission	https://github.com/UM-IHC-CA2i/amplifai-codabench
Competition & leaderboard	Codabench
Terms & Conditions	Challenge rules

⚠️ Access requires challenge registration. Refer to Starter Kit for Registration Instructions. Registration closes September 1, 2026.

---

Task

Classify the LI-RADS score of one pre-defined target lesion per case from multi-phase contrast-enhanced CT.

Participants are provided with:

• Multi-phase CT volumes (arterial, portal venous, delayed, and/or non-contrast)
• Voxel-level lesion segmentations
• Feature-level annotations (non-rim APHE, non-peripheral washout, enhancing capsule, etc.)
• Case-level LI-RADS category labels

Models are evaluated solely on predicted LI-RADS category. Feature annotations may be used freely (end-to-end or feature-guided approaches are both valid).

---

Dataset Structure

Data is released in batches. Download the batch zip(s) and the shared metadata file.

batches/
  batch_001.zip    ← cases (CT + annotations)
  batch_002.zip
  ...
metadata.csv       ← cumulative case-level labels across all batches

Per-case structure (inside zip)

batch_001/
└── cases/
    └── CASE00001/
        ├── ct/
        │   ├── CASE00001_ART.nii.gz   (arterial phase)
        │   ├── CASE00001_VEN.nii.gz   (portal venous phase)
        │   ├── CASE00001_DEL.nii.gz   (delayed phase)
        │   └── CASE00001_DRY.nii.gz   (non-contrast / native)
        └── annotations/
            ├── lesion.nii.gz
            ├── aphe.nii.gz
            ├── venous_capsule.nii.gz
            ├── delayed_capsule.nii.gz
            └── ...

Not all cases have all four phases — availability depends on the original acquisition protocol.

---

Download

from huggingface_hub import hf_hub_download

# Download a batch
hf_hub_download(
    repo_id="UM-IHC-CA2i/AMPLIFAI",
    repo_type="dataset",
    filename="batches/batch_001.zip",
    local_dir="./amplifai_data"
)

# Download the metadata
hf_hub_download(
    repo_id="UM-IHC-CA2i/AMPLIFAI",
    repo_type="dataset",
    filename="metadata.csv",
    local_dir="./amplifai_data"
)

Or via wget:

wget https://huggingface.co/datasets/UM-IHC-CA2i/AMPLIFAI/resolve/main/batches/batch_001.zip
wget https://huggingface.co/datasets/UM-IHC-CA2i/AMPLIFAI/resolve/main/metadata.csv

---

Metadata columns

Column	Description
batch_id	Source batch (e.g. batch_001)
case_id	Unique case identifier (e.g. CASE00001)
lesion	Lesion presence
max_diameter_mm	Maximum lesion diameter in mm
aphe	Arterial phase hyperenhancement type
washout_venous	Washout appearance on venous phase
washout_delayed	Washout appearance on delayed phase
capsule_venous	Capsule appearance on venous phase
capsule_delayed	Capsule appearance on delayed phase
lirads_score	Final LI-RADS score (classification target)

---

Citation

If you use this dataset, please cite the AMPLIFAI challenge and refer to the Terms & Conditions.

---

License

CC BY-NC-SA 4.0 — free for non-commercial research use with attribution. Derivatives must be shared under the same license.