Can AI Outperform Doctors at Detecting Cervical Cancer?

A computer algorithm was able to accurately analyze digital images of cervical screenings and identify precancerous changes that required further medical follow-up, a new study showed. The new approach, called automated visual evaluation, has the potential to change point-of-care cervical screening.

“Our findings show that a deep learning algorithm can use images collected during routine cervical cancer screening to identify precancerous changes that, if left untreated, may develop into cancer,” Mark Schiffman, MD, MPH, of the National Cancer Institute's Division of Cancer Epidemiology and Genetics, and senior author of the study said in a press release. “In fact, the computer analysis of the images was better at identifying precancer than a human expert reviewer of Pap tests under the microscope (cytology).”

According to the release, the artificial intelligence (AI) method is easy to perform. Health workers can use a cell phone or similar camera device for cervical screening and treatment during a single visit. In addition, this approach can be performed with minimal training, making it ideal for countries with limited health care resources, where cervical cancer is a leading cause of illness and death among women.

To create the algorithm, Schiffman and colleagues used information from a population-based longitudinal cohort study of 9,406 women aged 18–94 years in Guanacaste, Costa Rica. During a 7-year period, women underwent multiple cervical screening methods and histopathologic confirmation of precancers.

A tumor registry was used to link identified cervical cancers up to 18 years out from the study. More than 600,000 archived, digital cervical images taken at screening were used for training and validation of the deep learning-based algorithm.

The automated visual evaluation method performed better than all standard screening methods at identifying cases of cervical cancer diagnosed during the Costa Rica study. The AI method had greater accuracy (area under the curve [AUC] =0.91) than human expert review (AUC = 0.69) or conventional cytology (AUC = 0.71).

Looking at a single screening round of women aged 25–49, the automated visual evaluation identified 55.7% of 228 precancers and referred 11% for management.

Maurie Markman, MD, president of Medicine & Science at Cancer Treatment Centers of America, who was not involved in the research, told Cancer Network that this was a provocative report that utilizes modern technology to help deal with a critical cancer screening issue in low-resource settings.

“While additional work is required to optimize this deep learning algorithm employing an automated evaluation of images of the cervix the potential exists that this strategy could substantially favorably impact cervix cancer-associated morbidity and mortality in these clinical settings,” he said.