Hugging Face's logo

Datasets:
HealthDataHub
/
PARHAF-biomarkers-annotated

Tasks:
Token Classification
Modalities:
Tabular
Text
Formats:
parquet
Languages:
French
Size:
1K - 10K
Tags:
inception
uima
annotation
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
PARHAF-biomarkers-annotated / README.md
xtannier's picture
xtannier
Update README.md
dbd6a82 verified
preview code
|
raw
history blame contribute delete
15 kB
metadata 
license: cc-by-4.0
language:
  - fr
tags:
  - inception
  - uima
  - annotation
task_categories:
  - token-classification
configs:
  - config_name: document_metadata
    data_files:
      - split: train
        path: data/document_metadata/document_metadata_train.parquet
      - split: dev
        path: data/document_metadata/document_metadata_dev.parquet
  - config_name: spans
    data_files:
      - split: train
        path: data/spans/spans_train.parquet
      - split: dev
        path: data/spans/spans_dev.parquet
  - config_name: relations
    data_files:
      - split: train
        path: data/relations/relations_train.parquet
      - split: dev
        path: data/relations/relations_dev.parquet

Dataset Card for PARHAF-biomarkers-annotated

Logo

Reporting Issues & Contributing

If you encounter any errors or inconsistencies in this dataset, please report them in the discussion section of the "Community" tab on Hugging Face.

For more substantial contributions or collaboration opportunities, feel free to contact us directly.

Dataset Description

  • Points of Contact: FILORI Quentin, KHALIL Youness

Dataset Summary

PARHAF-biomarkers-annotated is a subpart of the PARHAF corpus, an open French corpus of human-authored clinical reports of fictional patients.

It was created to support the development and evaluation of clinical NLP systems for the extraction and characterization of genomic and tissue biomarkers from unstructured pathology reports in oncology..

This dataset contains training data only. The test set will remain under embargo to enable future evaluations under controlled conditions, limiting the risk of LLM contamination through prior data exposure. Please contact us for access to the test data.

This training dataset is divided into a train split (80%) and a dev split (20%) to facilitate experimental design and reproducibility across teams. Teams are free to use the full training set or define a different split configuration.

Each patient record was:

  • written by a senior medical resident
  • reviewed by another senior medical resident, from the same specialty
  • annotated by a specialist of the use case
  • curated by another specialist of the use case

Data statistics

DATASET SUMMARY

Indicator Value
Complete Dataset
Number of JSON files 152
Total annotations 2609
Average document length 2348 characters
80% Threshold (annotations) 2087

Complete dataset distribution by type

Type Annotations % of total
LayersRelation 911 34.92%
SpanBiomarker 1528 58.57%
SpanResultZone 170 6.52%

Complete dataset distribution by populated field

Type.Field Occurrences % of total annot.
LayersRelation.Relation 911 34.92%
SpanBiomarker.AmplificationState 564 21.62%
SpanBiomarker.Biomarker 994 38.10%
SpanBiomarker.BiomarkerAttribute 757 29.01%
SpanBiomarker.MutationBinary 31 1.19%
SpanBiomarker.MutationType 7 0.27%
SpanBiomarker.Normalized 994 38.10%
SpanBiomarker.Presence 994 38.10%
SpanResultZone.ResultZone 170 6.52%

Train / Dev Split Results

Indicator TRAIN DEV
Number of files 121 31
Number of annotations 2101 508
Percentage of dataset 80.53% 19.47%
Avg. document length (chars) 2268 2657

Train / Dev distribution by type

Type TRAIN DEV % train of type
LayersRelation 733 178 80.46%
SpanBiomarker 1233 295 80.69%
SpanResultZone 135 35 79.41%

Train / Dev distribution by populated field

Type.Field TRAIN DEV % train of field
LayersRelation.Relation 733 178 80.46%
SpanBiomarker.AmplificationState 451 113 79.96%
SpanBiomarker.Biomarker 805 189 80.99%
SpanBiomarker.BiomarkerAttribute 611 146 80.71%
SpanBiomarker.MutationBinary 27 4 87.10%
SpanBiomarker.MutationType 5 2 71.43%
SpanBiomarker.Normalized 805 189 80.99%
SpanBiomarker.Presence 805 189 80.99%
SpanResultZone.ResultZone 135 35 79.41%

Train / Dev distribution by value

Type.Field.Value TRAIN DEV % train
SpanBiomarker.Biomarker.TissueBiomarker 764 185 80.51%
LayersRelation.Relation.RefersTo 733 178 80.46%
SpanBiomarker.Presence.True 532 121 81.47%
SpanBiomarker.BiomarkerAttribute.AmplificationState 451 113 79.96%
SpanBiomarker.AmplificationState.Pos 285 62 82.13%
SpanBiomarker.Presence.False 273 68 80.06%
SpanBiomarker.AmplificationState.Neg 166 51 76.50%
SpanResultZone.ResultZone.ResultZone 135 35 79.41%
SpanBiomarker.BiomarkerAttribute.AmplificationRate 115 23 83.33%
SpanBiomarker.Normalized.NoNormalizedNameFound 58 9 86.57%
SpanBiomarker.Normalized.KI67 41 10 80.39%
SpanBiomarker.Normalized.PDL1 37 10 78.72%
SpanBiomarker.Biomarker.GenomicBiomarker 41 4 91.11%
SpanBiomarker.Normalized.TTF1 36 8 81.82%
SpanBiomarker.Normalized.CK7 38 5 88.37%
SpanBiomarker.Normalized.RE 29 4 87.88%
SpanBiomarker.BiomarkerAttribute.MutationBinary 27 4 87.10%
SpanBiomarker.Normalized.CK20 27 4 87.10%
SpanBiomarker.Normalized.RP 25 4 86.21%
SpanBiomarker.Normalized.P53 25 2 92.59%
SpanBiomarker.Normalized.P40 20 6 76.92%
SpanBiomarker.Normalized.PMS2 19 6 76.00%
SpanBiomarker.Normalized.CD20 19 6 76.00%
SpanBiomarker.Normalized.HER2IHC 23 1 95.83%
SpanBiomarker.Normalized.MSH2 17 6 73.91%
SpanBiomarker.Normalized.MSH6 17 6 73.91%
SpanBiomarker.Normalized.MLH1 16 6 72.73%
SpanBiomarker.MutationBinary.True 18 4 81.82%
SpanBiomarker.Normalized.TPS 15 7 68.18%
SpanBiomarker.Normalized.MSI 17 4 80.95%
SpanBiomarker.Normalized.ALK 16 5 76.19%
SpanBiomarker.Normalized.CD10 12 7 63.16%
SpanBiomarker.Normalized.ROS1 14 5 73.68%
SpanBiomarker.Normalized.GATA3 10 7 58.82%
SpanBiomarker.BiomarkerAttribute.MutationName 13 4 76.47%
SpanBiomarker.Normalized.CD5 9 7 56.25%
SpanBiomarker.Normalized.BCL2 9 7 56.25%
SpanBiomarker.Normalized.P63 14 2 87.50%
SpanBiomarker.Normalized.PAX8 11 3 78.57%
SpanBiomarker.Normalized.CD3 10 3 76.92%
SpanBiomarker.Normalized.P16 12 1 92.31%
SpanBiomarker.Normalized.CYCLINE 9 3 75.00%
SpanBiomarker.Normalized.CKAE1ouAE3 10 2 83.33%
SpanBiomarker.Normalized.BCL6 6 5 54.55%
SpanBiomarker.Normalized.CK5ou6 5 5 50.00%
SpanBiomarker.Normalized.MUM1 6 4 60.00%
SpanBiomarker.Normalized.CDX2 10 0 100.00%
SpanBiomarker.Normalized.CHROMOGRANINE 8 2 80.00%
SpanBiomarker.MutationBinary.False 9 0 100.00%
SpanBiomarker.Normalized.SYNAPTOPHYSINE 8 1 88.89%
SpanBiomarker.Normalized.EMA 8 0 100.00%
SpanBiomarker.Normalized.CPS 7 1 87.50%
SpanBiomarker.Normalized.CD79A 7 0 100.00%
SpanBiomarker.BiomarkerAttribute.MutationType 5 2 71.43%
SpanBiomarker.Normalized.WT1 4 2 66.67%
SpanBiomarker.Normalized.CD23 3 3 50.00%
SpanBiomarker.Normalized.BRAF 6 0 100.00%
SpanBiomarker.Normalized.NKX3ou1 6 0 100.00%
SpanBiomarker.Normalized.EGFR 4 2 66.67%
SpanBiomarker.Normalized.CD56 5 1 83.33%
SpanBiomarker.Normalized.CADHERINE 5 0 100.00%
SpanBiomarker.Normalized.PAX5 5 0 100.00%
SpanBiomarker.Normalized.CD34 5 0 100.00%
SpanBiomarker.Normalized.KL1 5 0 100.00%
SpanBiomarker.Normalized.BEREP4 4 0 100.00%
SpanBiomarker.Normalized.PS100 4 0 100.00%
SpanBiomarker.MutationType.Substitution 4 0 100.00%
SpanBiomarker.Normalized.NAPSINE A 3 1 75.00%
SpanBiomarker.Normalized.TDT 4 0 100.00%
SpanBiomarker.Normalized.CALRETININE 3 0 100.00%
SpanBiomarker.Normalized.CK19 3 0 100.00%
SpanBiomarker.Normalized.CD15 3 0 100.00%
SpanBiomarker.Normalized.SMAD4 2 1 66.67%
SpanBiomarker.Normalized.ActineMuscleLisse 2 1 66.67%
SpanBiomarker.Normalized.CD45 3 0 100.00%
SpanBiomarker.Normalized.CD79 3 0 100.00%
SpanBiomarker.Normalized.MYELOPEROXYDASE 3 0 100.00%
SpanBiomarker.Normalized.VIMENTINE 3 0 100.00%
SpanBiomarker.Normalized.D240 2 0 100.00%
SpanBiomarker.Normalized.AntiHepatocytes 1 1 50.00%
SpanBiomarker.Normalized.PSA 2 0 100.00%
SpanBiomarker.Normalized.ACE 2 0 100.00%
SpanBiomarker.Normalized.CD30 2 0 100.00%
SpanBiomarker.Normalized.HP 2 0 100.00%
SpanBiomarker.Normalized.DBA44 2 0 100.00%
SpanBiomarker.Normalized.HNF1BETA 2 0 100.00%
SpanBiomarker.Normalized.MET 2 0 100.00%
SpanBiomarker.Normalized.HER2FISH 1 1 50.00%
SpanBiomarker.Normalized.SATB2 2 0 100.00%
SpanBiomarker.Normalized.CD138 2 0 100.00%
SpanBiomarker.Normalized.CALDESMONE 2 0 100.00%
SpanBiomarker.MutationType.Deletion 0 2 0.00%
SpanBiomarker.Normalized.CD21 0 2 0.00%
SpanBiomarker.Normalized.BAP1 1 0 100.00%
SpanBiomarker.Normalized.EBNA2 1 0 100.00%
SpanBiomarker.Normalized.HMB45 1 0 100.00%
SpanBiomarker.Normalized.PRAME 1 0 100.00%
SpanBiomarker.Normalized.POLE 1 0 100.00%
SpanBiomarker.Normalized.GLYPICAN 3 1 0 100.00%
SpanBiomarker.Normalized.CD31 1 0 100.00%
SpanBiomarker.Normalized.BRCA1 1 0 100.00%
SpanBiomarker.Normalized.BRCA2 1 0 100.00%
SpanBiomarker.MutationType.OtherType 1 0 100.00%
SpanBiomarker.Normalized.RA 1 0 100.00%
SpanBiomarker.Normalized.INI1 1 0 100.00%
SpanBiomarker.Normalized.GFAP 1 0 100.00%
SpanBiomarker.Normalized.CD117 1 0 100.00%

Data Origin

The clinical reports are extracted from the PARHAF corpus. Please refer to PARHAF documentation for more information about this corpus.

Languages

  • fr_FR

Dataset Structure

We distribute both a Hugging Face dataset and a standalone version of the corpus. The standalone dataset consists of a JSON file per patient report, in UIMA CAS JSON format. This format constitutes the canonical version of the corpus. The Hugging Face dataset (Parquet/Arrow) is a derived representation generated automatically from the JSON files.

Both formats therefore contain identical information and differ only in storage layout.

One dataset instance corresponds to one report.

Hugging Face dataset

This snippet shows how to extract and iterate over medical report information per patient using the datasets library.

import pandas as pd
from datasets import load_dataset

dfs = {cfg: load_dataset("HealthDataHub/PARHAF-biomarkers-annotated", cfg, split="train").to_pandas()
       for cfg in ["document_metadata", "spans", "relations"]}

for patient_raw in dfs["document_metadata"].itertuples():
    report_id = patient_raw.report
    text = patient_raw.full_text
    report_spans = dfs["spans"][dfs["spans"]["report"] == report_id]
    report_relations = dfs["relations"][dfs["relations"]["report"] == report_id]
    ...

Data Fields

Path Type Description Possible values
document_metadata
report string Identifiant unique du rapport
full_text string Texte intégral du rapport
spans
report string Identifiant du rapport
span_id integer Identifiant de l'annotation
span_type string Type de l'entité annotée SpanBiomarker, SpanResultZone
begin integer Offset de début
end integer Offset de fin
span_text string Texte de l'entité
attribute_Biomarker string Biomarker GenomicBiomarker, TissueBiomarker
attribute_Presence string Presence False, True
attribute_Normalized string Normalized ACE, ALK, ActineMuscleLisse, AntiHepatocytes, BAP1, BCL2, BCL6, BEREP4, BRAF, BRCA1
attribute_BiomarkerAttribute string BiomarkerAttribute AmplificationRate, AmplificationState, MutationBinary, MutationName, MutationType
attribute_AmplificationState string AmplificationState Neg, Pos
attribute_ResultZone string ResultZone ResultZone
attribute_MutationBinary string MutationBinary False, True
attribute_MutationType string MutationType OtherType, Substitution
relations
report string Identifiant du rapport
relation_id integer Identifiant de la relation
source_term_id integer ID du terme source (Dependent)
source_text string Texte du terme source
target_term_id integer ID du terme cible (Governor)
target_text string Texte du terme cible
attribute_Relation string Relation RefersTo

Data Splits

Only the training set is released here. The remaining portion of the corpus will be temporarily embargoed to enable future evaluations under controlled conditions, thereby limiting the risk of large language model contamination through prior exposure to the data. You can evaluate your system on the test set through the CodaBench platform.

Annotation Guidelines

You can find the detailed annotation protocol here: annotation_guidelines.pdf

Licensing Information

This dataset is released under licenses:

  • CC BY 4.0
  • Etalab 2.0

Citation Information

[More Information Needed]