Doctors can't diagnose every single patient at risk of heart disease and even end up treating some who don't need treatment, but a team of researchers who decided to put machine learning on that case found that the new process was improved and results were notably more accurate with an AI helping doctors screen patients. To put it as shortly as possible, a machine learning AI that's fed routine clinical data about cardiac disease, in testing, was able to not only help doctors predict patients at extreme cardiovascular risk, but also screen patients that didn't actually need treatment, thus facilitating the process of diagnosing patients.
Funding from the National Institute for Health Research School for Primary Care Research allowed Stephen F. Weng and Jenna Reps, among others, to create and train an AI based on four different algorithm models. The models used conventional machine learning and were mostly trained by being fed data from real medical cases, primarily simple routine treatments. The models were based on a random forest method, logistic regression, gradient boosting machines, and neural networks. The four models all had an identical goal – predicting a patient's first cardiovascular event, which normally manifests as a heart attack. The freshly trained AI managed to beat an established AI based on guidelines from the American College of Cardiology and the new AI correctly predicted 355 patients out of a pool of 24,970 cardiovascular incidents.
The conclusions bode well for modern medicine, but the AI is only a bit ahead of the longstanding algorithm for now. Further advancement will come, of course, and the model will grow when fed more data over time, should the researchers choose to continue cultivating it. For the time being, no cardiologist will have to worry about having an AI shorten their day or even replace them despite the fact that even this experimental AI ended up diagnosing heart diseases somewhat more accurately than human doctors, but technology is certainly heading in that direction. Given the flexibility of machine learning and neural networking, this is essentially the natural progression of things, at least according to some Google executives. Medicine in particular is a segment that's not only ripe for an AI revolution but is already benefitting from the advancement of automated help in some areas.