The Stanford Machine Learning group, based at Stanford University, recently launched a competition designed to compare AI’s capability in interpreting chest x-rays to the capabilities of human experts from Stanford.  And the humans lost.

The winner was an AI team from JF Healthcare, a medical diagnostic start-up based in Nanchang, China, which outperformed all three Stanford radiologists involved with an average AUC score (a measure of diagnostic accuracy) of 0.926.

According to a JF statement, the test demonstrates “the role that AI can play in providing precise medical diagnostics, especially in underserved areas of the world, with profound implications for the treatment of lung cancer, tuberculosis and other diseases of the thorax.”

Founded in 2015 by Frank Wu, longtime leader of the healthcare division of Siemens in Northeast Asia, JF offers remote diagnostic services, focusing on chest x-rays, for rural township hospitals in China where certified radiologists are often not available.  For 2019, the company plans to have its mobile screening truck screen more than ten provinces in China for tuberculosis, offering a half-million people the possibility of early detection while also, ideally, preventing the spread of disease to broader populations.

"The goal is to be better than the best human in making diagnoses and deliver low cost care with a level of quality that is equivalent to the best hospitals in the world. To do that, the secret is not math; it's a huge volume of well-curated data," said Wu, JF Healthcare's CEO. "There are 3 billion people in the world without access to quality healthcare. The first step in remedying this problem is providing extremely accurate diagnostics, which is at the core of our strategy and mission.”

According to the company, JF plans to develop a chest x-ray screening product to simultaneously screen for tuberculosis, lung cancer and pneumonia, covering the major thorax diseases based on more than 300,000 chest x-ray images with curated radiology reports collected from approximately 1,000 township hospitals. 

The team is also hoping to fully leverage the clinical value of chest x-rays through AI and deliver it to the massive township hospital patient population in China where radiologists are most needed.