This article is more than 1 year old

Breast cancer screening AI app OK'd by watchdog

When putting software on trial is a good thing

An AI application that aids early stage breast cancer screening has been approved by the US Food and Drug Administration for commercialization, software developer MedCognetics announced this week.

MedCognetics, based in Texas, has built a cloud-based software-as-a-service system for radiologists that analyzes mammograms. The systems runs QmTRIAGE, the company's breast cancer screening tool, which just received 510(k) clearance from the FDA.

QmTRIAGE is powered by computer vision algorithms trained to estimate the density of breast tissue to detect things like cysts or tumors associated with cancer in mammography scans. Radiologists have to scrutinize these scans carefully to determine whether a patient requires a follow-up biopsy to test for breast cancer. MedCognetics believes its tool will aid radiologists by highlighting breast tissue that needs further examination.

"It will help with cost and time savings, reducing the amount of time a radiologist needs to review a case by 30 to 50 percent," MedCognetics' CEO, Debasish 'Ron' Nag, claimed in a statement to The Register. This is an important factor, given that demand for radiologists vastly outpaces supply, and burnout in the profession is rampant.

"The vast majority of early stage cancers do not present with any clinical signs – they are detected on a screening mammogram before the woman has any symptoms. The only cure to breast cancer today is early detection. To enable that you need to see beyond the naked eye, and this is what MedCognetics does," he added.

The FDA 510(k) clearance allows the company to start marketing QmTRIAGE and working with healthcare providers to get the tool working in clinical settings. Nag told us trials of the software will be brought to market in the US in 2023, and trials in other countries are on the agenda.

MedCognetics teamed up with researchers from the University of Texas Southwestern Medical Center (which has licensed some of its IP to MedCognetics) and the University of Texas at Dallas to obtain clinical data to train its AI-powered algorithms. 

Nag said AI medical models are often trained on small datasets that aren't representative of patient populations, making them less effective for people from different ethnicities. He said the data used to train MedCognetics' systems is more diverse, and the company is keen to partner with more medical institutions to train and test its systems on more breast cancer scans. 

"Being able to expose the AI system to different sources and further variation of patient mammographic cases will help continue to solidify performance," MedCognetics states on its web page. This can be done through providing anonymized patient cases and accompanying results from radiology and pathology reports where applicable."

The company hopes to start deploying its systems in commercial settings, and publish peer-reviewed articles demonstrating QmTRIAGE's performance in 2023. ®

More about


Send us news

Other stories you might like