Artificial intelligence has already proven its ability to analyze images of medical devices, and pass medical student tests. Now, it's the turn of a new AI-based tool to demonstrate its ability to read doctors' reports and accurately predict the risk of death, rehospitalizations and other potential complications.

The program was developed by a team from New York University's Lagon School of Medicine and is currently being tested at a number of NYU partner hospitals with the aim of spreading the technology to the medical community in the future.

The journal Nature published a study on the potential benefits of using the program.

The projection model, called NYUTron, was created from millions of medical observations contained in the files of 387,2011 patients treated between January 2020 and May <> at hospitals associated with New York University.

These included written doctors' reports, notes on the progress of the patient's condition, X-ray images and medical devices, and recommendations given to patients when they leave the hospital, totaling 4.1 billion words.

The most prominent challenge of the program was the success in interpreting the language used by doctors, as each of them has its own terminology that differs significantly from the other, especially in terms of abbreviations.

The program was tested in real conditions, particularly through training to analyze reports from a hospital in Manhattan, and then compare the results to those of a hospital in Brooklyn for different patients.

By studying what actually happened to patients, the researchers were able to measure how often the program's predictions were correct.

Not a substitute

The result was surprising: Neutron predicted 95 percent of those who had already died later before patients were discharged from the study hospitals, and its predictions for 80 percent of those who were readmitted less than a month after their discharge were correct.

These findings were more accurate than most doctors expected and also exceed the expectations of currently used non-AI IT models.

To the surprise, however, a well-experienced physician widely respected in the medical community gave expectations "even better than those given by the program," explains Eric Orman, a neurosurgeon and computer engineer at New York Medical School and lead author of the study published in the journal Nature.

The program also succeeded by 79% in predicting the length of stay of patients in the hospital, 87% in predicting cases of failure of guarantors and insurance companies to cover medical care expenses paid by patients, and 89% in predicting cases in which patients suffered additional health problems.

Dr. Orman stressed that AI will never replace the patient-doctor relationship, but it may allow "more information to be provided. for doctors to enable them to make "informed decisions".