RadGraph-IT 🇮🇹

RadGraph graph inference for Italian radiology reports, with output in the canonical RadGraph-XL format.

radgraphit takes one or more Italian radiology reports, runs the Italian DyGIE v2 model (PyTorch: shared encoder, span extractor, NER head, and relation head), and returns the graph of entities and relations as the RadGraph-XL-compatible dictionary.

• PyPI distribution: RadGraph-IT · Import / CLI: radgraphit · Python: >= 3.10 · Device: CPU or CUDA
• Model: downloaded automatically from pinned Hugging Face revisions on first use (~445 MB / 424 MiB plus small tokenizer/config files, then cached)

Installation

pip install RadGraph-IT

Usage — Python API

from radgraphit import RadGraphIT

predictor = RadGraphIT()
annotations = predictor.predict(
    "Non si evidenziano segni di pneumotorace dopo la rimozione del drenaggio toracico."
)

predict accepts a string or a sequence of strings and preserves order. predictor(report) is an alias for predictor.predict(report):

annotations = predictor(["referto 1", "referto 2"])  # batch, order preserved

Usage — CLI

# report as an argument
radgraphit predict "Non si evidenzia pneumotorace."

# from a file (one report per line), or from stdin
radgraphit predict --input-file reports.txt
echo "Nessuna evidenza di pneumotorace." | radgraphit predict

The result is written as JSON to stdout.

Output

For the input ["Non si evidenzia pneumotorace ..."] the output has this shape (top-level keys "0", "1", … in input order):

{
    "0": {
        "text": "<normalized tokens, space-separated>",
        "entities": {
            "1": {
                "tokens": "<span tokens>",
                "label": "Observation::definitely absent",
                "start_ix": 3,
                "end_ix": 3,
                "relations": [["located_at", "2"]],
            }
        },
        "data_source": None,
        "data_split": "inference",
    },
    "1": {"...": "..."},
}

• The indices (start_ix, end_ix) are token-level, zero-based, and inclusive.
• Relations are directed source → target lists [label, target_entity_id].
• The output is deterministic: entities and relations are stably ordered.

Citation

The output format and the serializer derive from the official RadGraph / RadGraph-XL package (see Third-Party). If you use this package, please cite both this work and the RadGraph-XL paper:

@software{radgraphit,
    author = "Daniel Rabottini and Edoardo Avenia and Rocco Angelella",
    title = "{RadGraph-IT}: {RadGraph} graph inference for {Italian} radiology reports",
    year = "2026",
    version = "1.0.1",
    url = "https://github.com/therabo/RadGraph-IT",
}

@inproceedings{delbrouck-etal-2024-radgraph,
    title = "{R}ad{G}raph-{XL}: A Large-Scale Expert-Annotated Dataset for Entity and Relation
             Extraction from Radiology Reports",
    author = "Delbrouck, Jean-Benoit and Chambon, Pierre and Chen, Zhihong and Varma, Maya and
              Johnston, Andrew and Blankemeier, Louis and Van Veen, Dave and Bui, Tan and
              Truong, Steven and Langlotz, Curtis",
    booktitle = "Findings of the Association for Computational Linguistics ACL 2024",
    year = "2024",
    url = "https://aclanthology.org/2024.findings-acl.765",
    pages = "12902--12915",
}

License

See the MIT License and Third-Party notices.

Not a medical device. The output is an automatic extraction of entities and relations from text and does not constitute a diagnosis, a report, or a clinical opinion.