M2P2

Table of Contents

Medical, Multilingual and Privacy-Preserving Natural Language Processing in the clinical domain (M2P2)

abstract

M2P2-graphic-summary.png

Analysis of large quantities of clinical notes offers a great potential for medical knowledge discovery. Several factors limit their potential scientific usage, primarily the private and sensitive nature of the information they contain. Coupled with that, the actual amount of annotated text available at any hospital in a specific medical area might be insufficient to enable the identification of relevant clinical signals. Additionally, the different languages in Switzerland necessitate the development of multilingual analysis capabilities. In this project, we propose to develop de-identification techniques that will help to overcome these problems and leverage them in the development of natural language processing (NLP) systems such as Large Language Models for medical records, with two specific areas of applications: (1) extraction of clinically-relevant information from notes in oncology, and (2) summarization of discharge communication. The first case study focuses on the detection of adverse drug reactions in oncology. The second case study is about methods for generating and summarizing discharge letters, in order to simplify the administrative burden for doctors, and to adapt the language to the patient's needs. Thanks to the participation of three major hospitals (EOC, USZ, CHUV) we will implement solutions valid across Switzerland, using three national languages (Italian, French, and German).

core information

snf_logo_0.gif idsia-logo.png logoUZH.png

chuv.png eoc-logo.png usz-logo.jpg

  • Duration: Jan 2025 - Dec 2027
  • Funding: 1'070'390 CHF
  • Funding agency: Swiss National Science Foundation
  • PI: Dr. Fabio Rinaldi (IDSIA USI-SUPSI, Lugano)
  • Co-proposers: Prof. Michael Krauthammer (UZH/USZ, Zurich) and Prof. J.L. Raisaro (CHUV, Lausanne)
  • Participating hospitals: USZ (Zurich), CHUV (Lausanne), EOC (Lugano)
  • Partners: Ricardo Pereira Mestre (EOC), Andreas Wicki (USZ), Frederic Erard (UNIL), Lorenzo Ruinelli (EOC), Mary-Anne Hartley (EPFL), Alessandro Ceschi (EOC), Francois Bastardot (CHUV)
  • Main goal: de-identification of medical records to enable secondary usage for scientific studies
  • Secondary goals: detection of adverse drug reactions, generation of discharge summaries

Contact

E-Mail: fabio.rinaldi AT idsia.ch

NLP group page

Author: Fabio Rinaldi

Created: 2024-11-07 Thu 23:28

Validate