"After 30 years of reading cardiograms (心电图), I can never tell whether it's from a man or woman, or the age of the person," said Eric Topol, a cardiologist from Scripps Research in La Jolia, California. "A machine can detect if a person has anaemia (贫血)or other difficult diagnoses (诊断)."
Topol is excited not only about how machines are already better than experts at spotting problems, but how they can discover patterns that experts wouldn't even notice. "In Japan, doctors are using machine vision to pick up polyps (息肉) in real time, and detecting whether or not they could be cancerous and whether they should have a biopsy," he said. "Machines will not replace physicians—but physicians making use of AI will soon replace those not using it."
Still, these are early days for the application of AI in healthcare. Pearse Keane, a consultant doctor at Moorrelds Eye Hospital, has been leading a collaboration between Moor fields and Google's Deep Mind Health. In 2018, he famously published a proof-of-concept paper in Nature showing the erst successful AI diagnosis for eye disease. "The algorithm that we've developed isn't in clinical use at the minute, so we're trying to implement this now," Keane said.
Keane mentioned the INSIGHT study, which is looking into eye disease and its link to other conditions such as diabetes (糖尿病). "We are using the eye as a window to the rest of the body. With deep learning, we can now look at a retinal (视网膜）photograph and say: 'This is a woman, she's 58 years old, she's not a smoker or a diabetic, her BMI is around 25, and her blood pressure is around 150 over 85'. Now, to me, that's staggering." Keane said.
The INSIGHT study is analysing more than three million OCT scans from around 300,000 patients. "We now know, for every person having had a retinal scan here, who's gone on to develop a heart attack or diabetes," he said. "The reason why we're excited is that we think that if we can get the appropriate data sets and learn them deeply, we can find much more in the back of the eye about the health of the rest of the body."
"The application of AI for healthcare and medicine is about precision and accuracy, but that's not all," said Topol. The most important aspect is how AI can promote a stronger human connection between doctor and patient. "We see patients in single digit numbers of minutes. And that's not enough, you need the gift of time, which AI can give back," he said. "Next year this will be the standard," he believes. "Rather than doctors being data clerks, they will be making eye contact with patients. There's no algorithm for empathy. That's a human characteristic that we have to develop and get back in the way it used to be."
