This article is more than 5 years old. Information may no longer be current.
Deep learning algorithm identifies cervical precancer, cancer
Click Here to Manage Email Alerts
A computer algorithm developed by Global Good and NCI has demonstrated accuracy in analyzing cervical images from modern digital cameras and detecting precancerous alterations that warrant further testing, according to study results published in Journal of National Cancer Institute.
“Our findings 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,” Mark Schiffman, MD, MPH, of the NCI’s division of cancer epidemiology and genetics, said in a press release. “In fact, the computer analysis of the images was better at identifying precancer than a human expert reviewer of Pap tests under the microscope.”
In the NCI-funded, population-based, longitudinal cohort study, Schiffman and colleagues sought to develop a deep learning-based visual evaluation algorithm that recognizes cervical precancer and cancer.
Researchers amassed images from roughly 30,000 screening visits of 9,406 women aged 18 to 94 years (median age, 35 years) in Guanacaste, Costa Rica. Women underwent screening at baseline and periodically from 1993 to 2000 using multiple tests — including cytology, HPV testing and cervicography — with histopathologic confirmation of precancer.
The researchers used tumor registry linkage to extend follow-up for cancers up to 18 years.
They trained the deep learning algorithm to distinguish conditions requiring treatment from those that did not using archived, digitized cervical images from screenings taken with a fixed-focus camera (cervicography). The algorithm produced an image prediction score (0-1) that could be categorized to balance sensitivity and specificity for identification of precancer and cancer.
Among the cohort, researchers observed 241 histologically verified cases of precancer (cervical intraepithelial neoplasia grade 2 [CIN2] or grade 3 [CIN3]) and 38 cases of cancer.
Results showed automated visual evaluation of the enrollment images had higher accuracy (area under the curve [AUC] = 0.91, 95% CI, 0.89-0.93) in detecting precancer/cancer than the original cervigram (AUC = 0.69; 95% CI, 0.63-0.74) among women of all ages (P < .001).
Moreover, the automated visual evaluation had better accuracy than conventional Pap smears (AUC = 0.71; 95% CI, 0.65-0.77; P < .001); liquid-based cytology (AUC = 0.79; 95% CI, 0.73-0.84; P < .001) first-generation neural network-based cytology (AUC = 0.7; 95% CI, 0.63-0.76; P < .001) and HPV testing (AUC = 0.82; 95% CI, 0.77-0.87; P < .001).
In a single visual screening sequence limited to women of prime screening ages (25-49 years), the automated visual evaluation detected 127 (55.7%) of 228 precancers diagnosed cumulatively in the whole adult population (CIN2, CIN3 and adenocarcinoma in situ) and referred 11% for management.
“When this algorithm is combined with advances in HPV 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,” Maurizio Vecchione, executive vice president of Global Good, said in the press release. – by Jennifer Byrne
Disclosures: The researchers report no relevant disclosures.
Click Here to Manage Email Alerts