NEW YORK ─ Researchers at the National Institutes of Health's National Cancer Institute are leading the development of a machine-learning-based dual-staining technology with a goal of enabling more accurate cervical cancer screening.
Their method ─ which uses an algorithm to identify and analyze protein biomarkers associated with abnormal cells on slide images ─ takes aim at helping clinicians reduce unnecessary follow-up testing and invasive procedures, according to Nicolas Wentzensen, a researcher at NCI who is leading the development of the technology with colleagues.
The new approach combines the algorithm with cloud-based whole-slide imaging to analyze the simultaneous expression of p16 and Ki-67 ─ protein markers associated with cervical cancer, according to a study published recently in the Journal of the National Cancer Institute.
Wentzensen said that the machine-learning approach has the potential to reduce the number of false-positive results stemming from standard pap cytology testing and decrease the number of unnecessary follow-up procedures for HPV-positive women. In their study, the researchers cut clinician referrals to colposcopy by about one third when they compared their technology with Pap cytology in the context of triage testing for cervical cancer following a positive HPV result.
Early identification of women who are most at risk for cervical cancer is vital.
Persistent infection with HPV is the principal cause of cervical cancer, and the virus is implicated in more than 99 percent of those cancers worldwide. While most HPV infections resolve on their own, some women who test positive or whose co-testing results are inconclusive may develop pre-cancerous cervical lesions that, if left untreated, may progress to cervical cancer. In the US, about 13,800 new cases of invasive cervical cancer will be diagnosed and about 4,290 women will die from the disease this year, according to the American Cancer Society.
Given the limits of current testing approaches, researchers, in general, have "a strong focus on finding methods that can be applied after HPV testing to identify women who are at the highest risk for precancer," Wentzensen said.
In the new study, he and his colleagues evaluated the performance of their fully automated dual-stain test and compared it to Pap cytology as well as a manual dual-staining approach.
The NCI researchers collaborated with colleagues at the Steinbeis Transfer Center for Medical Systems Biology, which is associated with the University of Heidelberg, to train the machine-learning algorithm to identify the presence of p16 and Ki-67 on slides, and they tested the algorithm using slide images with cells from a separate set of participants. Overall, they compared the methods using samples from 4,253 people who had been participating in an epidemiological study of HPV-positive cervical and anal precancers at Kaiser Permanente Northern California and the University of Oklahoma.
"In addition to automating the dual-staining process and making it independent from subjective interpretation, we also improved the clinical performance," Wentzensen said. "We detected more disease than the Pap cytology test and sent fewer women for diagnostic evaluation."
Wentzensen added that that automated dual-staining approach "contributes greater specificity" to the identification of women at risk of progressing to cervical cancer after a positive HPV test.
Molecular tests are highly sensitive and specific in detecting whether women have HPV infections, and women who test negative for HPV are at low risk for cervical cancer for the following decade. Though most cervical HPV infections that cause positive test results will not lead to precancer, clinicians still need a way to identify women who are at risk of progressing to cervical cancer following a positive HPV test, Wentzensen noted.
A higher-performing test, such as the one under development, can reduce referrals for additional HPV or Pap cytology tests to assess the need for colposcopy, biopsy, or treatment by trained professionals for women with positive HPV test results.
The technology to identify women at risk for cancer following a positive HPV test is improving and manual dual-stain testing for the presence of p16 and Ki-67 in cervical samples is already showing better performance than Pap cytology, Wentzensen noted.
In March, the US Food and Drug Administration approved the Roche Diagnostics Cintec Plus Cytology test as the first biomarker-based triage test for women whose primary cervical cancer screening results are positive for HPV using the firm's Cobas 4800 HPV Test.
The technology, which detects the simultaneous presence within a single cell of p16 and Ki-67, simplifies clinical decision making by providing easy-to-understand results so that clinicians and women are clear on the next steps, according to Roche. The firm declined to comment for this article, saying it doesn't comment on ongoing research.
However, the FDA approved dual-stain test involves some subjectivity in that it requires that a cytotechnologist analyze slides through a microscope to determine the result, Wentzensen noted.
"Dual-staining done by manual evaluation of slides is more reproducible, less subjective, and more sensitive than Pap cytology," he said. "Now, we have automated that evaluation, removed the subjective component, and further reduced the number of false-positive results."
Through cloud-based implementation, the method could be made accessible to a broader set of clinicians, and the study findings may lead to the adoption of a digital pathology-based machine-learning approach in clinical practice, the researchers noted.
The automated approach can enable evaluations by clinicians lacking a comprehensive infrastructure to conduct cervical cancer screening, Wentzensen said. With the method, staining and scanning of slides can be completed in a central laboratory, and storage and evaluation can be done in the cloud.
The technology under development at NCI "offers promise to transform an industry segment affecting millions of women testing HPV-positive each year," Abhi Gholap, founder of OptraScan, a provider of digital pathology products and services, said in an interview.
Such technologies can help mitigate limitations associated with standard cervical cancer screening strategies, including cervical cytology, HPV testing, and colposcopy that "are often not possible in resource-limited settings due to economic and other infrastructure issues," said Gholap, who is familiar with the NCI project but not involved in it. "Along with dual staining, an ability to scan cytology slide samples at an affordable price with computer-aided analytics and data sharing is an important consideration," he added.
Still, a fully automated platform that is commercially available will require additional studies and regulatory approvals in different countries. Wentzensen didn't want to speculate on when such a system could be commercially available but added that the group's goal is to see the approach progress to clinical use. He added that while the NCI researchers are participating in additional studies to support the development of the technology, a manufacturer would need to commercialize the platform.
The group is currently working to better understand the criteria associated with how the machine-learning algorithm determines a positive or negative result. "What exactly drives the platform's score requires a lot of additional work and is part of what we aim to do going forward," Wentzensen said.