AI trained on the treatment of blood poisoning and diagnosis of fractures

And although the AI ​​is not coping with every disease so far, the results of its work already look promising.

For effective treatment of patients requires a combination of training and experience. This is one of the reasons why people are looking forward to the prospects of using AI in medicine: algorithms can be trained to use the experience of thousands of doctors, giving them more information than any person could digest.

At the end of October, there was some evidence that the software might have come very close to meeting these expectations. Two papers were published describing the excellent preliminary results of using AI for diagnosis and treatment. The papers indicate completely different tasks and approaches, which suggests that the range of situations in which AI can be useful is very wide.

The choice of treatment methods

One study focused on sepsis (blood infection) that occurs when an overly strong immune system responds to an infection. Sepsis is the third leading cause of death worldwide, and remains a problem even after a patient is hospitalized. Methods of treating patients exist, but, judging by the statistics, there are significant opportunities to improve the situation. Therefore, a small team of scientists from Britain and the United States decided to check whether the software could provide this improvement.

They used reinforcement learning algorithmthat are considered effective in situations with “rare signals of reward”. In other words, such a large sample of the population will have many other things in the body, except for sepsis, which will affect the results of any treatment, and therefore the signals of effective treatment will be weak and difficult to distinguish. This approach was designed to increase the chances of their recognition.

A large base was used for training the software: more than 17,000 resuscitation patients and 79,000 hospitalized patients from more than 125 clinics. These patients contained 48 information parameters, from vital indicators and laboratory tests to demography. The algorithm used data to determine treatment that maximizes the patient’s chance of survival for 90 days. The researchers called the resulting software "AI-Clinician."

To assess the quality of the work of AI clinician used a separate set of medical histories of patients. The algorithm was used to select a treatment method, after which the actual treatment of patients was compared with the proposed algorithm. In general, the PO recommended lower doses of injections and higher doses of vasoconstrictor drugs. People whose treatment coincided with such recommendations survived more often than other groups of patients.

Diagnostics

The second paper evaluated the ability to detect problems requiring treatment, in particular, bone fractures. Often such problems are easy to see, but even a specialist is hard to notice a small chip or small crack. In most cases, the diagnosis falls on the shoulders not of a specialist, but of a doctor working in an ambulance. A new study does not seek to create an AI, replacing doctors, it only wants to help them.

The team asked 18 orthopedic surgeons to diagnose 135,000 images of potential wrist fractures, and then used this data to train the algorithm of a convolutional neural network with in- depth training.. The algorithm was used to mark areas that are worth paying attention to doctors who are not specialists in orthopedics. In fact, he helped them to concentrate on the areas that were most likely to have a fracture.

In the past, such tests gave too many diagnoses, and doctors recommended additional tests in harmless cases. But in this case, the accuracy of the diagnosis increased, and the false positives decreased. The sensitivity (or ability) to detect fractures has risen from 81% to 92%, and the accuracy (the ability to make a correct diagnosis) has risen from 88% to 94%. In sum, this means that among ambulance doctors, the number of incorrect diagnoses would be almost halved.

In both studies, the software was not used in a context that fully reflects the medical circumstances. Emergency doctors and doctors treating sepsis (and this may be the same people) will usually have many additional reasons for excitement and distractions, therefore integrating AI into their work will be difficult. But the success of these attempts suggests that clinical trials of AI can begin earlier than previously thought, and after that we will truly find out how much AI can help to make real diagnoses and prescribe treatment.