raphael-r/bright-eds-treatment

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

Description

This is a EDS-NLP (CamemBERT + CRF) architecture fine-tuned to extract clinical neuro-oncology entities related to the treatment 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):

• —chimios: Agents chimiothérapeutiques
• —chimio_protocole: Protocole chimiothérapie (Stupp, PCV, etc.)
• —chm_cycles: Nombre de cycles de chimiothérapie
• —chm_date_debut: Date début chimiothérapie
• —chm_date_fin: Date fin chimiothérapie
• —type_chirurgie: Type procédure (biopsie, exérèse partielle, exérèse totale)
• —qualite_exerese: Étendue résection (totale, subtotale, partielle)
• —chir_date: Date opération chirurgicale
• —rx_dose: Dose totale en Grays
• —rx_fractionnement: Nombre de fractions
• —rx_date_debut: Date début radiothérapie
• —rx_date_fin: Date fin radiothérapie
• —localisation_radiotherapie: Zone ciblée par la radiothérapie
• —corticoides: Corticoïdes prescrits
• —anti_epileptiques: Anticonvulsivants prescrits
• —optune: Dispositif Optune/TTFields
• —essai_therapeutique: Nom essai clinique

Performance on Validation Set

Aggregates:

• —Macro F1: 0.5515 (Precision: 0.5945, Recall: 0.5450)
• —Micro F1: 0.8201 (Precision: 0.9029, Recall: 0.7511)

Per-Label Breakdowns: | Label | Precision | Recall | F1 | |---|---|---|---| | chimios | 0.8750 | 0.7925 | 0.8317 | | chimioprotocole | 0.9179 | 0.8311 | 0.8723 | | chmcycles | 0.7222 | 0.7324 | 0.7273 | | chmdatedebut | 1.0000 | 1.0000 | 1.0000 | | chmdatefin | 0.0000 | 0.0000 | 0.0000 | | typechirurgie | 0.9651 | 0.9468 | 0.9559 | | qualiteexerese | 0.6667 | 0.0400 | 0.0755 | | chirdate | 0.0000 | 0.0000 | 0.0000 | | rxdose | 0.9586 | 0.9456 | 0.9521 | | rxfractionnement | 0.9745 | 0.9444 | 0.9592 | | rxdatedebut | 0.0000 | 0.0000 | 0.0000 | | rxdatefin | 0.0000 | 0.0000 | 0.0000 | | localisationradiotherapie | 0.5946 | 0.3860 | 0.4681 | | corticoides | 0.8736 | 0.9383 | 0.9048 | | antiepileptiques | 0.8910 | 0.9392 | 0.9145 | | optune | 0.0000 | 0.0000 | 0.0000 | | essaitherapeutique | 0.6667 | 0.7692 | 0.7143 |

Usage

# Inference Code
import edsnlp

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

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