Two promising new ventures involving AI were announced in May, one that could improve the detection of breast cancer in women and another that could reduce the number of unnecessary tests that doctors run on patients complaining of chest pains.
The first study makes use of AI’s deep learning techniques – basically, an ability to absorb a whole wealth of past information, learn from it it and draw its own conclusions for future patients – to better read mammograms that check for breast density.
Researchers at the Massachusetts Institute of Technology (MIT) and Harvard Medical School found that using deep learning “substantially improved risk discrimination” compared to traditional models used by radiologists to evaluate mammograms.
"There's much more information in a mammogram than just the four categories of breast density," says Adam Yala, the study’s lead author and a PHD candidate at MIT. "By using the deep learning model, we learn subtle cues that are indicative of future cancer."
An added benefit – the AI model knows no colour. In other words, it works equally “across diverse races, ages and family histories,” says Regina Barzilay, an AI expert and professor at MIT who collaborated on the study. That could cut down on the number of misdiagnoses often suffered by minorities in the United States, particularly African American women.
"Until now, African-American women were at a distinct disadvantage in having accurate risk assessment of future breast cancer. Our AI model has changed that," says Regina Barzilay, an AI expert and professor at MIT who collaborated on the study.
A second study released this month uses AI in a different way: To avoid unnecessary testing for diseases and cut down on health costs.
Researchers in Europe found that artificial intelligence can be a far better predictor of whether patients who complain of chest pain really need to go through costly and time-consuming examinations of their heart and blood vessels.
"AI has the potential to save costs and staff time by identifying patients with chest pain who do not have significant coronary artery disease and therefore do not need expensive cardiac imaging," says the study’s author, Marco Mazzanti of the Royal Brompton Hospital in London.
One glaring example cited by researchers: whether to use a CTA scan to check for blocked blood vessels. Cardiologists recommended the scan for 83% of patients with chest pains while the AI model, dubbed ARTICA, suggested it for just 10%. Despite recommending fewer tests overall, researchers found that ARTICA made the right call 97% of the time.
Still, Mazzanti seemed to acknowledge the world might not be ready for a machine to take full control of the decisions-making process just yet, calling it a “second set of eyes” rather than a replacement for doctors.
"As doctors we order a lot of tests which cost a lot of money and waste time. ARTICA is like a second set of eyes to make sure we follow recommendations," says Mazzanti. – dpa
Did you find this article insightful?