Medizinische Universität Graz - Research portal
SHR
Neuro
Cancer
Cardio
Metab
Microb
Lipid
Simplification of medical reports using Artificial Intelligence.
- Abstract
- In recent years, there has been a trend in the medical landscape toward value-based healthcare. Value-based healthcare strives to achieve the best possible health outcomes or the best possible care at the lowest cost [Porter, 2010]. Studies proved that patients that better understand their health status have higher treatment compliance and a better outcome. Hence, well-informed and empowered patients play a key role in a modern patient-centered healthcare system.
The necessary information about a disease and the appropriate therapy is recorded in a medical patient report. This report is used to document the patient's state by treating physicians and to exchange information between physicians involved in the patient’s therapy. Originally, a patient report is not intended to provide information to the patient, so highly specialized medical language is used. To convey relevant information to the patient, a verbal conversation is the usual way. The quality of this information transfer is related to the doctor’s and patient’s communication skills, the time available for the conversation and the mental state of the patient, making this process highly user-dependent.
The availability of electronic health records (EHR) and the advent of Deep Learning (DL) technologies in natural language processing (NLP) now enable a semantic analysis to translate relevant content into language that medical laypeople can understand. Especially in the healthcare and clinical domain, DL techniques are rarely used as they lack available open data, which is mainly due to privacy concerns. This is even worse for German-language clinical data. In this project, we want to overcome these issues and apply text simplification for German-language clinical reports with a special focus on radiological reports to empower patients to take a more active role in managing their health.
To go beyond radiology reports, it is essential to extend the schema for medical entity and relation extraction to other medical fields. The generation of a large annotated medical corpus has to be supported by expert knowledge. To keep manual annotation costs as low as possible, the development of a novel easy to use, multiuser open-source annotation framework for human annotators is another goal of this project. It is important for this project to demonstrate the feasibility of report simplification in a real-world scenario. We are aiming to support existing healthcare providers with this technology, as well as enable patients to autonomously upload and analyze their medical reports independently.
- Project Leader:
- Reishofer Gernot
- Duration:
- 01.10.2022-07.02.2025
- Programme:
- IKT der Zukunft
- Type of Research
- basic research
- Staff
- Reishofer, Gernot, Project Leader
- Beger, Alexander Sebastian, Co-worker
- Renner, Roland, Co-worker
- Hassler, Eva Maria, Co-worker
- Gunzer, Felix, Co-worker
- MUG Research Units
- Department of Radiology
- Division of Neuroradiology, Vascular and Interventional Radiology
- Project results published
- > Information extraction from German radiological re... Sci Rep. 2023; 13(1): 2353
- > Reproducibility of artificial intelligence models ... Insights Imaging. 2022; 13(1):173
- > Active Learning Cycle for Information Extraction f... ECR Book of Abstracts, Insights into Imaging 13 (Suppl 4):205. 2022; 13: -ECR; 13-17 July, 2022; Vienna, AUSTRIA.