Artificial intelligence can predict a variety of diseases for years earlier

Scientists have developed a new productive artificial intelligence tool called “Delphi-2m” that can predict the risk of people with more than a thousand different diseases in the next twenty years.

According to the Guardian, the new Delphi-2M artificial intelligence processs a person’s medical history as a large language model that reads the text. Delphi-2m every medical diagnosis considers any important health event and even lifestyle factors such as cigarette or obesity as a “word” or “token” in a long sentence.

By learning the patterns in the medical records of millions, this artificial intelligence learns which “words” normally appear after other words and can even predict the approximate time of their occurrence. Built based on the modified GPT-2 architecture, the model has been able to prove its ability by training about 6,000 people in the UK and then testing on 1.2 million in Denmark.

Delphi-2m Artificial Intelligence to predict the disease

The results show that Delphi-2M in short-term health forecasts has achieved 2 % accuracy, and this accuracy will remain at 2 % for the next decade forecasts. This accuracy is recorded while simultaneously evaluating the entire spectrum of human diseases; The advantage that distinguishes it from current tools such as “QRisk” that predicts only one particular disease (such as myocardial infarction).

Despite the remarkable performance, this tool also has some limitations. For example, the accuracy of predictions for a 5 -year horizon decreases to about 2 to 5 percent, as many factors can change the situation over the past two decades. The data used to train the model have also been collected mainly from white, educated and relatively healthy people in the UK. This creates bias, and for use in more diverse populations, the model requires validation with wider data.

However, patients may enjoy the benefits of this tool over the next few years. Researchers imagine a future in which the doctor can tell the patient, “These are the four main risks of your health in the future, and you can do these two things to change the situation.”

Now, Delphi-2m is a research tool and is not publicly available. The findings of this study have been published in the journal Nature.