A recent study performed by researchers has revealed that a computer algorithm enhanced the efficiency and precision of cervical cancer screening when compared to the Pap test (cytology).
The latter is the present standard used to monitor women who tested positive with primary human papillomavirus (HPV) screening.
The latest method employs artificial intelligence (AI) to automate dual-stain assessment and holds distinct implications for clinical care.
The study findings were published in the Journal of the National Cancer Institute on June 25th, 2020. The computer algorithm was developed and the research was performed by scientists at the National Cancer Institute—part of the National Institutes of Health—in association with investigators from a number of other institutions.
We’re excited to show we have a fully automated approach to cervical cancer screening as a follow-up to a positive HPV test that outperformed the standard method in our study. Based on our results, it could increase the efficiency of cervical cancer screening by finding more precancers and reducing false positives, which has the potential to eliminate a substantial number of unnecessary procedures among HPV-positive women.
Nicolas Wentzensen, MD, PhD, Study Lead, Division of Cancer Epidemiology and Genetics, National Cancer Institute
In the recent past, clinicians have been hoping to leverage the developments in machine learning and digital imaging to enhance cervical cancer screening. For women who tested negative for HPV, they face a low risk of cervical cancer for the next 10 years, and even a majority of the cervical HPV infections—which result in positive HPV tests—will not lead to precancer.
But determining which women who tested positive for HPV are most likely to experience precancerous cervical changes is rather difficult; hence, colposcopy should be performed on these patients to assess their cervix and collect biopsy samples, or on those who require instant treatment.
Women who tested positive for HPV may have further Pap cytology tests or HPV tests to evaluate the requirement for treatment, biopsy, or colposcopy.
In Pap cytology, exclusively trained laboratory professionals, or cytotechnologists, examine the stained slides to search for atypical cells. This test is employed to detect precancers before they develop into cancer. However, such methods are not suitable. For instance, Pap cytology tests are highly sensitive, time-intensive, and may lead to false-positive findings.
Dual-stain testing is another method that has evolved over the years. It provides a more accurate way to predict whether a woman who tested positive for HPV has precancerous changes in her cervical cells. This test quantifies the presence of a pair of proteins—Ki-67 and p16—in cervical samples.
In a couple of earlier studies, Dr Wentzensen along with his collaborators observed that women who tested negative on the dual-stain approach had a decreased risk of developing cervical precancer in the next five years and that fewer women tested positive for dual-stain when compared to the Pap cytology test.
Back in March 2020, the U.S. Food and Drug Administration approved the manual dual-stain cytology test for women who have obtained a positive result on a primary HPV screening.
There is a subjective component in the manual dual-stain test, that is, a cytotechnologist should peer at the slide to establish the outcomes. In the latest work, the scientists wanted to find out whether a completely automated dual-stain test could surpass or match the performance of the manual method.
In association with Niels Grabe, PhD, and Bernd Lahrmann, PhD, from the Steinbeis Transfer Center for Medical Systems Biology—associated with the University of Heidelberg—the scientists have now created a whole-slide imaging platform. After being trained with deep learning, this platform could establish if any cervical cells were stained for both proteins—that is, Ki-67 and p16.
The researchers compared this technique with both manual dual-stain testing and traditional Pap cytology in samples taken from as many as 4,253 people who took part in one of three epidemiological analysis of HPV-positive anal and cervical precancers conducted at the University of Oklahoma and Kaiser Permanente Northern California.
The scientists eventually observed that the AI-based dual-stain test exhibited a lower rate of positive tests when compared to the manual dual-stain and Pap cytology tests, with more improved sensitivity (the potential to correctly identify precancers) as well as considerably higher specificity (the potential to accurately detect those without precancers) when compared to the Pap cytology test.
AI-based dual-stain decreased colposcopy referral by almost one-third when compared to the Pap cytology test (about 42% versus 60%). The testing approach was also powerful, demonstrating similar performance in anal cytology.
In brief, the automated test exceeded the performance of the present standard, Pap cytology, decreasing the proportion of false-positive outcomes and considerably decreasing referral to unwanted colposcopy processes.
The outcomes also support further assessment of the test as an alternative for screening of anal cancer. The scientists observed that their method has evident clinical application, and via cloud-based implementation, it would be accessible worldwide. The platform is also suitable for other applications, like quality control, assisted evaluation, and second opinion.
The FDA has only recently approved the manual dual-stain test for screening women who have tested positive for HPV, and hence, its use is only being started. Extra regulatory approval will be required to enable the screening of HPV-positive women with a completely automated dual-stain test.
According to the scientists, their findings act as a crucial example for introducing deep learning and digital pathology into the clinical setting, and their method has the ability to considerably enhance cervical cancer screening, impacting scores of women testing HPV-positive every year.
Wentzensen, N., et al. (2020) Accuracy and Efficiency of Deep-Learning–Based Automation of Dual Stain Cytology in Cervical Cancer Screening. Journal of the National Cancer Institute. doi.org/10.1093/jnci/djaa066.