In the current study, we propose to develop further and evaluate the accuracy of a recently developed, promising, low-cost visual screening approach—automated visual evaluation (AVE)based on computer-based deep learning of cervical images collected using a mobile device. Specifically, we will train and assess an algorithm for the diagnosis of cervical intraepithelial neoplasia 3 or greater (CIN3+) for: 1) primary screening and 2) triage of HPV-positive women for colposcopy. We propose to conduct a community-based cross-sectional screening study of 20,000 women aged 25 to 49 years in Varanasi, Uttar Pradesh and Sangrur, Punjab. Screening procedures will include VIA and HPV testing using BD Onclarity, an approved clinical assay that yields HPV type as well as positivity/negativity.  At the time of VIA, a digital camera image will be taken to train the AVE algorithm (which will not be used clinically).  Women positive for VIA or HPV will be referred to colposcopy, biopsy and treatment.  The scientific standard of disease will be based on histopathology of the colposcopically directed biopsies.

This study will result in the development and internal validation of an AVE score (range=0 to 1) for accurate and efficient identification of women with CIN3+. We hypothesize that an AVE algorithm using images collected through a mobile phone in India, Brazil, Nigeria, Zambia and few other small network sites will provide good discrimination and calibration.  Additionally, based on the data published to date, we hypothesize that AVE will outperform (increase sensitivity/specificity) VIA as both a primary screening method and triage method of HPV-positive women in the split-sample validation set, though we are not powering this study to specifically compare the performance of AVE in a blinded fashion to VIA and the primary aim of the study is to train the algorithm.

Upon successful completion of this study, we plan to independently validate the AVE in a separate study. This latter study will be conducted as a formal efficacy trial, with comparison of AVE to VIA. Importantly, if successfully shown to be more accurate as a screening method than VIA, AVE will be scaled up under the auspices of WHO as a mobile phone application for broad public health use, independent of patent/commercial interests.