CURRENT ROLE OF AI IN THE MEDICAL FIELD
The idea of Artificial Intelligence replacing human doctors has become more and more acceptable over the years. We are getting to a point where AI can outperform humans in a large number of domains. Healthcare diagnostics is no exception. AI has demonstrated itself to be a very important innovation in this area, because it can quickly learn to recognize abnormalities that a doctor would also label as a disease with more accuracy. However, at the moment these systems tend to work in a very opaque manner and because it can be difficult to “look under the hood” to see what these AI systems see human doctors tend to have a better overall idea of the big picture when making a diagnosis. This is similar to the “black box problem” that has been discussed on this blog before.
The Radbound University Medical Center has described in a new publication ways that AI can become more relevant to clinical practice.
DOCTORS vs AI
Artificial Intelligence is being utilized for diagnostic work. Human doctors can examine an X-ray or an MRI to identify abnormalities but these tasks can also be accomplished by an AI with arguably more accuracy and in less time. AI also tends to outperform human doctors when attempting to diagnose complex chronic illnesses, whose causes are more cryptic.
Despite the diagnostic performance gap between human and AI doctors, the human doctor is better able to look at the big picture when diagnosing patients. They may consider more thoroughly a treatment plan weighed against the patient’s quality of life or whether an abnormality should be removed through surgery or treated some other way. The AI tends to be “lazy” in the sense that it makes an accurate diagnosis but does not offer anything else in the way of help.
MAKING AI MORE RELEVANT TO CLINICAL PRACTICES
To make AI systems more useful for clinical practices researchers at the Radboud University Medical Center have developed a two-sided innovation for diagnostic AI. The idea is deep learning algorithms that analyze medical images are taught to parse over the image again and again ignoring areas that it has already passed despite the fact that an accurate diagnosis can be made by parsing the image only once. This makes the diagnosis more transparent as the AI will provide a more detailed summary of the problem, which will help human doctors enormously.
The method described above is equivalent to a human doctor being able to interpret the same medical image over and over with a fresh pair of eyes, something we know humans can’t actually accomplish. This not only allows the AI to make a more detailed diagnosis similar to how a human doctor would, but it also allows for even more accuracy.
For further reading the full report from the Radboud University Medical Center is available here.