AI beats doctors in spotting breast cancer
What takes a pathologist hours, machines do in a wink.
It’s a result that may further jangle the nerves of doctors already skittish in the face of machine medicos. Breast cancer is the latest disease that artificial intelligence (AI) can diagnose better than humans, according to a recent study in the Journal of the American Medical Association.
Led by Babak Ehteshami Bejnordi at Radboud University Medical Centre in the Netherlands, the study pitted different machine-learning algorithms against 11 pathologists in analysing 129 biopsies.
While the pathologists had years of experience, the algorithms were trained with just 270 digital scans of lymph node sections, 110 with malignant cells meticulously labelled by pathologists to show the cancers’ locations.
The human pathologists were given two hours to examine the slides, mimicking real-life workload in the Netherlands. On average they spotted just 31 of 49 cancers. One further pathologist, given no time limit and taking 30 hours, found 46. The top-performing algorithm, from the Harvard Medical School and Massachusetts Institute of Technology, significantly outperformed the timepoor doctors and performed on par with the pathologist given 30 hours – a time, the authors note, “infeasible in clinical practice”.