Can Computers Diagnose Cancer?

A Stanford study found that computers have some demonstrably serious advantages over human pathologists in diagnosing breast cancer.

Last year, researchers in Stanford University’s Computational Pathologist project trained computers to analyze tissue samples for breast cancer. When they compared the computers’ analysis against the readings of human doctors, they discovered something surprising. The computers provided significant additional information beyond what physicians could with a visual analysis and grading.

Daphne Koller, the Stanford computer science professor who wrote the paper published in November, says C-Path’s results strongly suggest that computers might be able to help improve doctors’ diagnostic decisions – for instance, by suggesting tailored treatment regimens.

“I personally foresee data-driven medicine, which is a step beyond evidence-based medicine”, Koller says. “Conclusions arise from the data directly, as opposed to having a human necessarily come in with a preformulated hypothesis”.

That may sound like a job killer, but doctors need not worry. “Computers are smart and very narrow in their thinking”, Koller says. “I don’t think this is going to end up replacing doctors in the foreseeable future” – K.F.