Study: AI approach to detect cervical precancer outperforms human experts

Research findings published Thursday in the Journal of the National Cancer Institute suggest that an artificial intelligence (AI) approach known as "automated visual evaluation" was able to analyse digital images of the cervix and accurately identify both cervical precancer and cancer. Senior author Mark Schiffman said the results "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." He added "in fact, the computer analysis of the images was better at identifying precancer than a human expert reviewer of Pap tests under the microscope."

To create the algorithm, investigators used more than 60 000 cervical images from an archive of photos collected during a cervical cancer screening study, which included a cohort of 9406 adult women ages 18 to 94 who were followed for as long as 18 years. Photos were digitised and used to train the deep learning algorithm so that it could differentiate cervical conditions requiring treatment from those not requiring treatment.

The authors reported that "automated visual evaluation of enrollment cervigrams identified cumulative precancer/cancer cases with greater accuracy than original cervigram interpretation or conventional cytology." Further, a single visual screening round restricted to women at the prime screening ages of 25 years to 49 years identified 55.7 percent of 228 precancers diagnosed cumulatively in the entire cohort, with 11 percent of the cases referred for management, the study found.

"When this algorithm is combined with advances in HPV [human papillomavirus] vaccination, emerging HPV detection technologies, and improvements in treatment, it is conceivable that cervical cancer could be brought under control, even in low-resource settings," said Maurizio Vecchione, executive vice president of Global Good, a fund at Intellectual Ventures that participated in the study.

The researchers plan to train the algorithm further on images of cervical precancers and normal cervical tissue from women around the world, using a variety of cameras and other imaging options, with the aim being to create the best possible algorithm for common, open use. Schiffman indicated that the method likely would not replace methods currently used in the US, saying he envisions it being applied in places where women need to be screened and treated on the same day if tests uncover areas of concern.

Did you like this article?