The challenge dedicated to simplifying medical records.
Third place for the University of Udine in the first international scientific medical IT challenge dedicated to the simplification of medical records thanks to the use of artificial intelligence systems. They were registered for the competition, the “Snomed Ct Entity Linking Challenge”. over 500 researchers, universities and research centers from all over the world.
The aim of the competition is to find solutions that can help facilitate and speed up the work of healthcare personnel in compiling and understanding clinical documents. For the Friulian University one participated team of the Medical IT, Telemedicine and E-Health Laboratory (Mitel) of the Department of Mathematical, Computer and Physical Sciences. The group was made up of the laboratory coordinator, Vincenzo Della Mea, professor of medical informatics, the researcher Kevin Roitero and the doctoral student Mihai Horia Popescu. The first place in the challenge was an Israeli team from the “KI research institute”, according to a team from the Bauman Moscow State Technical University. The winners’ solutions are made available to everyone and can be freely used.
The competition is organized by a consortium, Snomed international, which includes 48 countries. The aim of the consortium is to manage a clinical terminology, the Snomed-Ct, made up of over 300 thousand concepts accompanied by codes intended to represent everything that can be described in a medical record, giving it a standardized form suitable for data analysis and support to the decision. Italy, for the moment, is not part of the consortium.
“At present – explains Professor Della Mea –, where Snomed-Ct is actually used to annotate medical records or parts thereof, such as medical reports pathological anatomy, the choice of terms and consequently of codes is carried out mainly by hand by healthcare personnel that enters the data. This makes its use complicated and time-consuming. The organizers wanted to challenge the scientific community regarding theidentification of machine learning techniques to simplify the work of healthcare personnel“.
The Medical IT, Telemedicine and E-Health Laboratory has, among its lines of research, a line relating to biomedical terminologies and classifications. In particular, both from the point of view of the systems for their management and the automatic coding of clinical documents with artificial intelligence methods.
“In the challenge our team – explains Professor Della Mea – has adopted a two-stage approach. In the first we have attempted to identify text segments of clinical interest, training a special artificial intelligence system, the Large language model Mistral. In the second, through a further model, we have associated the Snomed-Ct codes with the previously identified segments. As for our system, it appears to have identified more segments than human experts did.”
Do you want to remove national ads?
You can subscribe to only €1.10 per month by clicking here
Already a subscriber?
You can log in by going to the section Login from the site menu or by clicking here
