'Deep Learning'–Based Visual Evaluation for Cervical Cancer Screening
In an observational study reported in the Journal of the National Cancer Institute, Hu et al found that an automated 'deep learning'–based visual evaluation algorithm permitted identification of cervical precancer/cancer cases with greater accuracy than other screening methods.
Study Details
The study involved a population-based longitudinal cohort of 9,406 women aged 18 to 94 years in Guanacaste, Costa Rica, who were followed for 7 years (1993 to 2000) with multiple cervical screening methods and histopathologic confirmation of precancers. Cancers identified through 18 years were identified from tumor registry linkage. Archived digitized cervical images from screening taken with a fixed-focus camera (cervicography) were used for training/validation of the deep learning–based algorithm.
Accuracy of Methods
There were 241 histologically confirmed cases of precancer (CIN2/CIN3) and 38 cases of cancer observed among the 9,406 women during follow-up. The automated visual evaluation of cervigrams using the deep learning algorithm identified precancer/cancer cases with greater accuracy (receiver operating characteristic area under the curve [AUC] = 0.91), compared with original cervigram interpretation (AUC = 0.69, P < .001), conventional cytology (AUC = 0.71, P < .001), liquid-based cytology (AUC = 0.79, P < .001), first-generation neural network-based cytology (AUC = 0.70, P < .001), and human papillomavirus testing (AUC= 0.82, P < .001).
It was estimated that a single visual screening round using the automated method among women at the prime screening ages of 25 to 49 years would identify 55.7% of all precancers found in the entire population.
The investigators concluded, “The results support consideration of automated visual evaluation of cervical images from contemporary digital cameras. If achieved, this might permit dissemination of effective point-of-care cervical screening.”
Mark Schiffman, MD, MPH, of the National Cancer Institute, is the corresponding author for the Journal of the National Cancer Institute article.
Disclosure: The study authors' full disclosures can be found at academic.oup.com.
The content in this post has not been reviewed by the American Society of Clinical Oncology, Inc. (ASCO®) and does not necessarily reflect the ideas and opinions of ASCO®.
