FRACCO dataset - ICD-O standardised annotation for French oncology NLP

This repository contains the FRACCO (FRench Annotated Corpus for Clinical Oncology), a French-language corpus of 1,301 synthetic clinical cases in oncology, derived from the Spanish CANTEMIST dataset and translated as part of the FRASIMED project. The corpus is annotated with 71,127 entity mentions covering morphology, topography, and histologic differentiation, and standardised following the International Classification of Diseases for Oncology, Third Edition (ICD-O-3). In addition, higher-level annotations capture composite clinical expressions (expression_CIM) that combine multiple ICD-O concepts.

Annotations were created through a dual-review process involving domain experts, with both manual curation and automated support to ensure quality. Normalisation to ICD-O codes was performed automatically where possible and manually otherwise, additional translations were applied to correct around 1'000 mistranslated expressions from FRASIMED. representing 399 unique morphology codes (from 2’549 different expressions), 272 topography codes (from 3’143 different expressions), and 2’043 unique composite expressions (from 11’144 different expressions).

The dataset is distributed in BRAT standoff format for processing and training (.txt and .ann files aligned with CANTEMIST/FRASIMED numeration) and in .csv for general review of annotation content and codes.

Corresponding code is available at https://github.com/SimedDataTeam/FRACCO.