The application of AI in healthcare is something I’ve covered numerous times in this blog over the years. Most of these projects have fed on huge quantities of data to make inferences faster and often more effectively than would otherwise be the case.
The latest comes from new research from the USC Viterbi School of Engineering. The study describes an AI-based system that mines electronic medical records to suggest diagnoses and recommend diagnostic tests, which the authors believe could lead to better and faster treatment for patients.
“The algorithm works just like a doctor – thinking about what to do next at each stage of the medical work-up,” the researchers say. “The difference is that it has the benefit of all the experiences in the collective healthcare records. This is a new form of artificial intelligence.”
A new approach
Most AI-based systems in healthcare use Bayesian Inference to suggest diagnoses based on the information available. It’s a probability thing. This research describes a slightly different approach and instead reverses the process and strives to find tests that would be most likely to correctly identify an illness. It’s an approach referred to as Bayesian Exploration.
It’s an approach that has a number of benefits. For instance, the authors believe that the algorithm could improve diagnostic and testing decisions by providing a number of good options. The system is also designed to continue learning as more data is added to the medical records.
It could even speed up the completion of the medical records as the system would be able to correctly input the relevant information into the system for the doctor. Of course, the system would also take account of any instances whereby the doctor rejects the AI recommendations.
“The algorithm isn’t meant to make decisions for doctors or replace them,” the researchers explain. “It’s meant to complement and support them.”
Changing behaviors
As with many of the new generation of technologies, the system could introduce new ways of working for physicians. The researchers use the example of a patient reporting a headache. In a traditional scenario the doctor might assess the patient for fever or other neurological symptoms and in their absence suggest that it’s a stress headache. The patient then returns and complains that the headache has worsened, which makes stress a less likely cause, so the doctor moves on to diagnoses such as brain tumors or viral encephalitis. These tests are not without risks of their own and are costly, so other options tend to be looked at first.
If the doctor was using the algorithm, however, the records of millions of patients could be mined for those with symptoms similar to those presented by the current patient. The system would then do a cost-benefit calculation in providing the most recommended screenings to the doctor.
Obviously, there’s a lot of barriers in the way of such a decision support system becoming commonplace in healthcare, but the researchers are optimistic that eventually, such support will be the norm.
“If the promise of success is great enough, then people are going to be motivated to do it,” they conclude. “And that’s what we think this algorithm provides: the possibility, the promise of offering a solution to a huge problem that wastes a lot of resources, trillions of dollars’ worth.”
