A team at Lehigh University in Bethlehem, Pennsylvania, has developed an artificial-intelligence screening tool for cervical cancer that appears to be better than traditional methods, and far less expensive.
The researchers at Lehigh’s Image Data Emulation & Analysis Laboratory spent 10 years developing the system, which is based on computer analysis of digital photographs of the cervix.
Their study, ”Multi-feature base benchmark for cervical dysplasia classification,” appeared in the journal Pattern Recognition.
AI refers to scientists programming computers to think like humans. Those who champion AI say it can make better decisions than humans in carrying out specific tasks.
In the Lehigh case, researchers programmed a computer to recognize abnormalities in cervix cells and tissues that are collectively known as cervical dysplasia.
The system already appears to interpret results better than experts, and the research team is refining it so it does even better. Because of the system’s low cost, it could be used in developing countries, where more than 80 percent of the world’s cervical cancer deaths occur.
Pap smears and HPV tests are commonly used to screen for cervical cancer, but they require specialized equipment that makes them too expensive in many places. Lehigh developed AI for an alternate screening method called cervicography that is so cheap that it could replace standard tests in the developing world.
Before AI, experts had to interpret the photos — called cervigrams — that were obtained through digital cervicography. Experts can disagree about results, however, raising concerns about the method’s effectiveness.
The Lehigh team developed algorithms to train the computers to recognize and classify the dysplasia seen in cervigrams. An algorithm is a mathematical formula that programmers develop to control artificial intelligence applications. The team developed seven series of algorithms, or classifiers, to grade dysplasia.
To come up with the algorithms and program them, the researchers used cervigrams from 1,112 patients. They included 345 patients with moderate or severe dysplasia, which is likely to develop into cancer, and 767 with low-grade, mild dysplasia, which the body can normally eliminate.
Researchers also built a database with patients’ diagnoses and outcomes as a reference for health professionals who want to use the technology.
The AI system outperformed Pap tests and HPV tests in diagnosing cervical cancer, while achieving an acceptably low rate of false positives. These occur when a test diagnoses a woman without cancer as having the disease.
AI was particularly adept at detecting moderate and severe cervical cancer. “Our lower-cost image-based classifiers perform comparably or better than human interpretation on traditional Pap and HPV tests, on our test dataset,” the researchers wrote.
“Our method would be an effective low-cost addition to a battery of tests helping to lower the false positive rate, since it provides 10% better sensitivity and specificity than any other screening method, including Pap and HPV tests,” Sharon Xiaolei Huang, an associate professor of computer science and engineering at Lehigh, said in a press release.
“Humans and computers are very complementary,” Huang said. “That’s what AI is all about.”
While refining its cervical cancer screening system, the team is also developing an AI system to detect breast cancer.