Skip to main content

CleanNote: preprocessing and harmonization of medical notes

Project description

CleanNote

English Français

CI codecov PyPI version License: MIT


English

Back to top

CleanNote analyzes raw medical notes and transforms them into concise, structured documents focused on symptoms, medical conclusions, and treatments, enabling easier analysis and clinical research.

This solution was developed during a research internship at the **** laboratory, and is currently demonstrated using the AGBonnet/Augmented-clinical-notes dataset together with the mistralai/Mistral-7B-Instruct-v0.3 model.

This work was supervised by three people:

  • Ms. ****, company supervisor and associate professor, who proposed the topic of the preprocessing pipeline,

  • Ms. ****, school supervisor and associate professor, who guided me on frugality aspects and formatting,

  • Mr. ****, company supervisor and PhD student, who provided support on identifying adapted solutions and domain knowledge.


Installation

Back to top

First, make sure you are inside your Python virtual environment (e.g. venv).
To install the latest available version (see the PyPI badge above):

 pip install -U cleanote

If you want to install a specific version (for example 0.2.1):

 pip install -U cleanote==0.2.1

The latest released version is always displayed in the PyPI badge at the top of this README.


Usage

Back to top

After installation, you can using CleanNote with just few lines of code:

from cleanote.dataset import Dataset
from cleanote.model import Model
from cleanote.pipeline import Pipeline

# Load a dataset
data = Dataset(name="AGBonnet/Augmented-clinical-notes", split="train", field="full_note", limit=1)

# Load a model
model = Model(name="mistralai/Mistral-7B-Instruct-v0.3", max_new_tokens=512)

# Create pipeline
pipe = Pipeline(dataset=data, model_h=model)

# Run pipeline
out = pipe.apply()

# Display result
print(out.data.head())

# Download the dataset homogenized
xls = pipe.to_excel()  
print(f"Excel file saved to : {xls}")

Literature

Back to top

  • Identification de profils patients à partir de notes cliniques non structurées.
    Corentin Laval, Catherine Combes, Rémi Eyraud, Virginie Fresse.
    PFIA 2025.

Français

Haut de page

CleanNote analyse des notes médicales brutes et les transforme en documents concis et structurés, centrés sur les symptômes, les conclusions médicales et les traitements, afin de faciliter leur analyse et la recherche clinique.

Cette solution a été développée dans le cadre d’un stage de recherche au laboratoire **** , et est actuellement démontrée à l’aide du jeu de données AGBonnet/Augmented-clinical-notes ainsi que du modèle mistralai/Mistral-7B-Instruct-v0.3.

Ce travail a été supervisé par trois personnes :

  • Mme ****, tutrice entreprise et maîtresse de conférences, qui a proposé le sujet du pipeline de prétraitement,

  • Mme ****, tutrice école et maîtresse de conférences, qui m’a encadré sur les aspects de frugalité et de mise en forme,

  • M. ****, tuteur entreprise et doctorant, qui m’a accompagné sur l’identification des solutions adaptées et l’apport de connaissances du domaine.


Installation

Haut de page

Tout d’abord, assurez-vous d’être dans votre environnement virtuel Python (par ex. venv).
Pour installer la dernière version disponible (voir le badge PyPI ci-dessus) :

 pip install -U cleanote

Si vous souhaitez installer une version spécifique (par exemple 0.2.1):

 pip install -U cleanote==0.2.1

La dernière version publiée est toujours affichée dans le badge PyPI en haut de ce README.


Utilisation

Haut de page

Après installation, vous pouvez utiliser CleanNote en seulement quelques lignes de code :

from cleanote.dataset import Dataset
from cleanote.model import Model
from cleanote.pipeline import Pipeline

# Charger un jeu de données
data = Dataset(name="AGBonnet/Augmented-clinical-notes", split="train", field="full_note", limit=1)

# Charger un modèle
model = Model(name="mistralai/Mistral-7B-Instruct-v0.3", max_new_tokens=512)

# Créer le pipeline
pipe = Pipeline(dataset=data, model_h=model)

# Lancer le pipeline
out = pipe.apply()

# Afficher le résultat
print(out.data.head())

# Exporter le jeu de données homogénéisé
xls = pipe.to_excel()  
print(f"Fichier Excel sauvegardé : {xls}")

Références

Haut de page

  • Identification de profils patients à partir de notes cliniques non structurées.
    Corentin Laval, Catherine Combes, Rémi Eyraud, Virginie Fresse.
    PFIA 2025.

License

This project is licensed under the MIT License.

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

cleanote-0.2.3.tar.gz (12.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

cleanote-0.2.3-py3-none-any.whl (8.2 kB view details)

Uploaded Python 3

File details

Details for the file cleanote-0.2.3.tar.gz.

File metadata

  • Download URL: cleanote-0.2.3.tar.gz
  • Upload date:
  • Size: 12.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.13.7

File hashes

Hashes for cleanote-0.2.3.tar.gz
Algorithm Hash digest
SHA256 698287b3c0d765dab8ab3d17ff6f6dc631489f2f08197d224ef7780911dd15d4
MD5 64d30cf75eceb5710dfd8ed1ac0066a3
BLAKE2b-256 33f04db8b12e4537287f7781c458444071e078174a905a8ae41c0fa48728d5d2

See more details on using hashes here.

File details

Details for the file cleanote-0.2.3-py3-none-any.whl.

File metadata

  • Download URL: cleanote-0.2.3-py3-none-any.whl
  • Upload date:
  • Size: 8.2 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.13.7

File hashes

Hashes for cleanote-0.2.3-py3-none-any.whl
Algorithm Hash digest
SHA256 f887638c1f377526464e4185a198d0cce2fea2ebab3715396714895bb3ed2210
MD5 0207a99193b0bafde247dbc7dcaa173e
BLAKE2b-256 36148d98fbcc09bb5f4c078de4aac3020d6a92ac8ba5cb456a9c42dcb01b1381

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page