One of the most common and cost-effective ways to detect cervical cancer is the pap smear, in which cells are scraped from a woman's cervix and sent to a laboratory for analysis. But this method requires equipment and medical expertise that are not always available in some low-income countries. Now scientists are making an app they hope could use artificial intelligence to identify precancerous or cancerous cells with just a photograph.

The app is being developed by researchers at the National Institutes of Health and Global Good; the latter is a joint effort by Bill Gates and invention firm Intellectual Ventures. Their preliminary results, published online in January in the Journal of the National Cancer Institute, suggest that such an approach could significantly improve cervical cancer diagnosis in low-resource settings.

Cervical cancer rates are higher in countries or regions that lack the resources to conduct pap smears. Health care providers in these areas often use a less accurate diagnostic technique, in which they swab the cervix with dilute acetic acid and visually inspect it for any white spots that might signal abnormal cells.

Over a seven-year period NIH researchers routinely photographed the cervixes of more than 9,400 women in Costa Rica. They used these images to train an AI algorithm to recognize characteristics of abnormal tissue—and to predict later cancer development. When the algorithm analyzed new images, it performed better than a clinical expert did by visual inspection.

“We were surprised to see that computers could see much more sensitively and clearly which cervixes are or are not precancerous,” says Mark Schiffman, a molecular epidemiologist at the National Cancer Institute and senior author of the paper. “I really thought [the AI was] cheating.” The scientists ultimately plan to implement their algorithm on mobile phones and aim to train future iterations of the program with digital camera photos.