Scientists announced Wednesday that they had created a model of artificial intelligence capable of predicting medical diagnoses years in advance, based on the same technology as Chatgpt of OpenAI.
Starting from the patient’s clinical history, The delphi-2m “predicts the prevalence of more than 1,000 diseases” in advanceaccording to a study published by the Nature Magazine.
The scientists, from institutions from the United Kingdom, Denmark, Germany and Switzerland, resorted to the medical history of about half a million patients from the UK Biobank biomedical database. To treat all this data, they used a technology similar to that of Chatgpt, a language model initially based on textual content.
Understanding a sequence of medical diagnoses is equivalent to “learning the grammar of a text,” Moritz Gerstung, a specialist at the German Cancer Research Center, told the press.
Thanks to your training, Delphi-2M “Learn to detect schemes in health data, before diagnoses, and in what combinations and successions occur”Gerstung added, allowing “very significant and relevant predictions for health.”
The specialist presented diagrams that suggest that AI could identify people at a risk of suffering a more or less high infarction than factors such as age or other data could predict.
Delphi-2M’s capabilities were tested from the health data of almost two million people in Denmark.
In the future, models like this could help “guide monitoring and, possibly, lead to earlier clinical interventions” in preventive medicine, Gerstung said.
The research teams indicated, however, that this model of AI had to undergo more evidence and that it was not yet ready for use.
On a larger scale, these tools could help “optimize resources in a health system under pressure,” said Tom Fitzgerald, of the European Molecular Biology and study co -author.
In many countries, computer devices are already used to predict the risk of diseases, such as the QRISK3 program, which are used by British doctors to evaluate the risk of suffering a heart attack or stroke.
But Delphi-2M “can manage all diseases at the same time and for a long period,” said another author of the text, Ewan Birney.
For Gustavo Sudre, professor at the King’s College in London specialized in medical, this advance “seems to be a significant step towards a predictive modeling in medicine that is scalable, interpretable and, what is more important, ethically responsible.”