RadGraph-IT

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

This repository hosts the v1 Italian DyGIE model used by the RadGraph-IT package. It jointly extracts named entities and relations from Italian radiology reports using a shared encoder, span extractor, NER head, and relation head. The encoder is IVN-RIN/medBIT-r3-plus.

Authors

Daniel Rabottini, Edoardo Avenia, and Rocco Angelella.

Usage

from transformers import AutoModel

model = AutoModel.from_pretrained("radgraphIT/Radgraph-IT", trust_remote_code=True)
model.eval()

out = model.predict(
    "Non versamento pleurico. Addensamento parenchimale al lobo inferiore del polmone destro."
)
print(out["words"])
print(out["entities"])   # [start, end, label, raw_score, softmax_score] (inclusive word-index spans)
print(out["relations"])  # [start1, end1, start2, end2, label, raw_score, softmax_score]

model.predict(text) applies the word-level preprocessing used during training (regex cleanup and nltk.wordpunct_tokenize). If you already have a list of words tokenized with that exact scheme, use model.predict_words(words) instead.

The bundled tokenizer (AutoTokenizer.from_pretrained("radgraphIT/Radgraph-IT")) is the encoder's subword tokenizer; it is not the word-level splitter used by predict().

The model requires transformers, torch, numpy, and nltk (for predict()). The radgraphit package provides the supported Python and CLI interface, including RadGraph-XL-compatible serialization.

Architecture

Word-level embeddings are produced by mean-pooling medBIT-r3-plus subwords back to one vector per word. Documents longer than 512 tokens are split into word-aligned chunks. Span representations concatenate start/end word embeddings with a span-width embedding. The NER head scores candidate spans of width up to 12; the relation head prunes to the top spans-per-word and scores each retained pair.

Entity labels are Anatomy and Observation, each with definitely present, definitely absent, or uncertain status. Relation labels are modify, located_at, and suggestive_of.

Training and evaluation

The model was trained on the full non-cross-validation split. The best validation epoch was 8.

Metric	NER	Relation
Precision	0.848	0.700
Recall	0.868	0.680
F1	0.858	0.690

The full per-class breakdown is in model/metrics.json.

Limitations and responsible use

RadGraph-IT is a research software model for automatic information extraction. It is not a medical device; its output is not a diagnosis, report, or clinical opinion. Validate outputs for the intended setting and keep human clinical judgement in the loop.

Citation

Please cite both RadGraph-IT and RadGraph-XL when using this model:

@software{radgraphit,
  author = "Daniel Rabottini and Edoardo Avenia and Rocco Angelella",
  title = "{RadGraph-IT}: {RadGraph} graph inference for {Italian} radiology reports",
  year = "2026",
  version = "1.0.0",
  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",
}