Abstract
Background: Cytology-based cervical cancer screening is limited by its low sensitivity, especially in low-resource settings. Methods: We trained a deep learning system (DLS) using deep convolutional neural network models on a large dataset of 188,542 microscopic cytological images. We enrolled 2145 women aged 25-64 years to evaluate the algorithm by a clinical, multicenter, blinded case-control study. Cervical specimens were collected for liquid-based cytology by DLS analysis or by cytologists reading, and for Cobas HPV test, respectively. Colposcopy-directed biopsy were performed on all women. We compared the diagnostic performance of the DLS for detection of histopathologically confirmed CIN grade 2 or worse (CIN2+) with cytologists and with Cobas HPV test. Findings: The DLS identified 92·6% of CIN grade 2, 93·0% of CIN grade 3, and 94.6% of cervical cancer, with DLS scores significantly increasing with severity of cervical lesion (p 0·05 for both), respectively, whereas DLS specificity was higher than skilled cytologists (p<0·001) and similar to HPV test (p=0·30). In HPV positive women, the DLS showed similar sensitivity (p=0·79) but significantly higher specificity (p<0·001) compared to skilled cytologists in CIN2+ detection. Compared with cytologists at primary hospitals, the DLS increased sensitivity and specificity in CIN2+ detection by 49% and 84% (p<0·001), respectively. Interpretation: DLS could have similar diagnostic performance for CIN2+ compared with skilled cytologists and Cobas HPV test. Funding Statement: This study was supported by grants from the Association of Maternal and Child Health Studies (2017AMCHS006), National Natural Science Foundation of China (81903328). Declaration of Interests: The authors declared no conflicts of interest. Ethics Approval Statement: The study protocol was approved by the ethical review committee of the National Center for Chronic and Non-communicable Disease Control and Prevention, China CDC. Written informed consent was obtained from all study participants.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.