Smaragd - NLP support for radiological reporting
The goal of the Smaragd project is to digitalise the screening and reporting process in radiology. This innovative approach improves the service for patients and referring physicians and opens up global market potential.
- Lead school School of Engineering and Computer Science
- Institute Institut für Medizininformatik I4MI
- BFH centre BFH Centre for Health technologies
- Funding organisation Innosuisse
- Duration (planned) 08.05.2022 - 07.05.2024
- Project management Prof. Dr. Kerstin Denecke
- Head of project Prof. Dr. Kerstin Denecke
Insel Gruppe AG
ID Suisse AG
- Keywords Radiology; Natural Language Processing; Conversational Agent
Radiology is a high-throughput medical discipline that serves different customer groups (referring physicians and patients). Due to the heavy workload and economic constraints, compromises have to be made in terms of the service: a) Referrers receive medical reports in prose, but would prefer a standardised, digitised form. b) For logistical reasons, personal contact between the doctors (radiologists) and their patients is possible only in exceptional cases.
Course of action
The SMARAGD consortium’s vision is to support the radiological reporting process with the help of artificial intelligence (AI): 1. Automatic structuring of radiological findings: Using a combination of a Natural Language Processing (NLP) pipeline and a new kind of template, reports written in German prose can be output in a structured, standardised form, ready checked for completeness. 2. User experience: The radiologists’ usual working methods are not impinged upon but supported. The strengths of free-text technical language are preserved. 3. Medical history chatbot: Before being examined, patients are given the opportunity to provide their medical history via chatbot. The chat, which is also processed with NLP, is provided to the radiologists for reference when making their diagnosis.
In a historical first, the chatbot now enables radiologists to include the patient-specific medical history in every report. The information is taken from the patient referral and supplemented by any information contained in the Clinic Information System (KIS). The chatbot will be rule-based and analyse free-text input using AI methods. There has never before been a chatbot for radiology that asks for the patient’s medical history. Unlike existing chatbots in the healthcare sector, the conversation with the SMARAGD chatbot is modelled in the ontology. This makes it possible to structure the patients’ free-text answers in the chat with the same methodology as the radiology findings.