Loading...
 
[Show/Hide Left Column]

COT 

Coréférence événementielle cross-document dans les dossiers électroniques patient 


Axis : DataSense, tâche 2
Subject : Coréférence événementielle cross-document dans les dossiers électroniques patient
Directors : Aurélie NÉVÉOL, LIMSI, Xavier TANNIER, LIMSI, Olivier FERRET, IHRIM
Institution :LIMSI & CEA list
Administrator laboratory : LIMSI
PhD Student : Julien TOURILLE
Begin :10/01/2015
Defense date :12/18/2018

Scientifics production :
  • LIMSI-COT at SemEval-2017 Task 12: Neural Architecture for Temporal Information Extraction from Clinical Narratives - Julien Tourille, Olivier Ferret, Xavier Tannier, Aurélie Névéol - 2017 - Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)
  • Tourille, Julien, Ferret, Olivier, Tannier, Xavier, Névéol, Aurélie. LIMSI at SemEval-2017 Task 12: Neural Architecture forTemporal Information Extraction from Clinical Narratives in Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval 2017).
  • Tourille, Julien, Ferret, Olivier, Tannier, Xavier, Névéol, Aurélie. Temporal information extraction from clinical text in Proceedings of the European Chapter of the ACL (EACL 2017, short paper). Valencia, Spain, April 2017.
  • Tourille, Julien, Ferret, Olivier, Névéol, Aurélie, Tannier, Xavier. LIMSI-COT at SemEval-2016 Task 12: Temporal relation identification using a pipeline of classifiers in Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval 2016). San Diego, USA, June 2016. Selected for the "Best of SemEval 2016"
  • Tourille, Julien, Ferret, Olivier, Névéol, Aurélie, Tannier, Xavier. Extraction de relations temporelles dans des dossiers électroniques patient. in Actes de la Conférence Traitement Automatique des Langues Naturelles (TALN 2016, article court). Paris, France, July 2016.

Others résults :
  • Campagne d'évaluation ClinicalTempEval 2016 & 2017.
    • 2016, résultats très encourageants,
    • 2017, meilleurs résultats de l'ensemble des participants, dans la plupart des tâches.

Ressources :
Image


Context :
The retrospective analysis of health pathways gives a synthetic vision of the management of a pathology. It makes it possible to compare actual management with recommendations for good practice in order to evaluate the quality of care pathways, identify elements of the pathway requiring specific management and improve the diagnostic or therapeutic approach.

The TOC project proposes to use automatic language processing methods to facilitate this analysis. Julien Tourille will work on the textual content of patient files to develop tools adapted to the biomedical language that will make it possible to identify the salient events in the patient's medical history and the associated time markers in order to aggregate them into a synthetic chronology.

Scientific objective:
The first part of the work will consist in correctly detecting the temporal expressions used in the corpus. These temporal expressions can be absolute dates (e.g. January 14, 2008) but also relative dates that will require a resolution or normalization phase (e.g. January 14, 2008). This study will then make it possible to adapt existing time analysis tools, built mainly for the journalistic field. The team can build on similar studies conducted for the English language (Leaman et al. 2013, Fan & Friedman 2011, Zhu et al. 2013, Mork et al. 2013).

The next step will be cross-document co-reference. All the information contained in the documents will be filtered in order to establish a summary chronology of the most important events.

The chronologies obtained will be compared with the reference courses (developed with the help of doctors) and with standard management protocols.

Perspectives :
The project presents many interests from the point of view of e-Health issues. Three main objectives have been established:
obtaining a synthetic and visual document of the patient's evolution;
the comparison of monitoring protocols in force and actual management
Analysis of the differences in treatment between similar cases and the discovery of new links between phenotypes and treatments.