raphael-r/bright-eds-chromosomal

BRIGHT NER: EDS-NLP (CamemBERT + CRF) fine-tuned for chromosomal

Description

This is a EDS-NLP (CamemBERT + CRF) architecture fine-tuned to extract clinical neuro-oncology entities related to the chromosomal semantic group. It was trained on a synthetic dataset generated for the properly de-identified BRIGHT project dataset (see the generated_data folder in the primary repository).

This model repository was specifically designed to fit within the bright_db overarching namespace.

Fields

It extracts the following fields (described in French):

• ch1p: Délétion 1p
• ch19q: Délétion 19q
• ch1p19q_codel: Codélétion 1p/19q
• ch7p: Gain/perte 7p
• ch7q: Gain/perte 7q
• ch10p: Délétion 10p
• ch10q: Délétion 10q
• ch9p: Délétion 9p
• ch9q: Délétion 9q
• ampli_egfr: Amplification EGFR
• ampli_cdk4: Amplification CDK4
• ampli_mdm2: Amplification MDM2
• ampli_mdm4: Amplification MDM4
• ampli_met: Amplification MET
• fusion_fgfr: Fusion FGFR
• fusion_ntrk: Fusion NTRK
• fusion_autre: Autre fusion

Performance on Validation Set

Aggregates:

• Macro F1: 0.4373 (Precision: 0.5188, Recall: 0.3984)
• Micro F1: 0.8042 (Precision: 0.8622, Recall: 0.7535)

Per-Label Breakdowns:

Label	Precision	Recall	F1
ch1p	0.8261	0.7037	0.7600
ch19q	1.0000	0.3077	0.4706
ch1p19q_codel	0.8364	0.6216	0.7132
ch7p	0.8804	0.9529	0.9153
ch7q	0.8889	0.6667	0.7619
ch10p	0.8571	0.4000	0.5455
ch10q	0.8636	0.9870	0.9212
ch9p	1.0000	0.7333	0.8462
ch9q	0.0000	0.0000	0.0000
ampli_egfr	0.6667	0.4000	0.5000
ampli_cdk4	0.0000	0.0000	0.0000
ampli_mdm2	0.0000	0.0000	0.0000
ampli_mdm4	0.0000	0.0000	0.0000
ampli_met	0.0000	0.0000	0.0000
fusion_fgfr	0.0000	0.0000	0.0000
fusion_ntrk	0.0000	0.0000	0.0000
fusion_autre	1.0000	1.0000	1.0000

Usage

# Inference Code
import edsnlp

nlp = edsnlp.load("raphael-r/bright-eds-chromosomal")
doc = nlp("Patient presenting with epileptic seizures...")

for ent in doc.ents:
    print(ent.text, "=>", ent.label_)