Stanford researchers built an algorithm that recognizes pneumonia better than human radiologists

A group of Stanford researchers says it has built an algorithm that can diagnose pneumonia in chest x-rays better than the average human radiologists.

The group, which includes deep learning researcher Andrew Ng, developed the algorithm using more than 112,000 chest x-rays released by the National Institutes of Health in September. A little over a month later, the algorithm could identify 14 types of medical conditions and diagnose pneumonia better than a group of four radiologists.

The researchers published their results in a paper (PDF) posted the open access website arXiv.  

“Automated detection of diseases from chest X-rays at the level of expert radiologists would not only have tremendous benefit in clinical settings, it would be invaluable in [the] delivery of healthcare to populations with inadequate access to diagnostic imaging specialists,” the researchers wrote.

Pranav Rajpurkar, a graduate student in the Stanford Machine Learning Group who co-authored the paper, told Stanford News the algorithm can help radiologists detect an illness that is difficult to diagnose but hospitalizes more than 1 million people each year, according to the Centers for Disease Control and Prevention. The group has done similar work developing and testing algorithms that can identify heart arrhythmias.

RELATED: Artificial intelligence is transforming radiology, but it still can't make human judgment calls

The researchers used their work to build a tool that uses heat maps to visualize areas that are most likely to represent pneumonia.

On social media, Ng stoked the flames of a contentious debate about whether machine learning and AI will replace radiologists altogether. Eric Topol, M.D., director of the Scripps Translational Science Institute and a professor of genomics at The Scripps Research Institute, argued that radiologists aren't losing their jobs anytime soon and questioned whether four radiologists were representative of the entire profession.

Topol has previously acknowledged the impact AI could have on image-based specialties and suggested that radiology and pathology merge into one specialty to make room for high-tech advancements.

RELATED: Stanford researchers—Bring expectations for artificial intelligence back down to earth

That conversation is likely to intensify as researchers and AI companies test out new data-driven mechanisms for disease identification. Researchers have done similar work in image heavy specialties like dermatology, where Stanford researchers used machine learning to identify instances of skin cancer as accurately as 21 board-certified dermatologists.

“There is massive potential for machine learning to improve the current healthcare system, and we want to continue to be at the forefront of innovation in the field,” Jeremy Irvin, a graduate student in the Machine Learning Group and the other co-lead of the paper told Stanford News.