Inès Baleydier

Study protocol for a two-site clinical trial to validate a smartphone-based artificial intelligence classifier identifying cervical precancer and cancer in HPV-positive women in Cameroon

Introduction Cervical cancer remains a major public health challenge in low- and middle-income countries (LMICs) due to financial and logistical issues. WHO recommendation for cervical cancer screening in LMICs includes HPV testing as primary screening followed by visual inspection with acetic acid (VIA) and treatment. However, VIA is a subjective procedure dependent on the healthcare provider’s experience. Its accuracy can be improved by computer-aided detection techniques. Our aim is to assess the performance of a smartphone-based Automated VIA Classifier (AVC) relying on Artificial Intelligence to discriminate precancerous and cancerous lesions from normal cervical tissue. Methods The AVC study will be nested in an ongoing cervical cancer screening program called “3T-study” (for Test, Triage and Treat), including HPV self-sampling followed by VIA triage and treatment if needed. After application of acetic acid on the cervix, precancerous and cancerous cells whiten more rapidly than non-cancerous ones and their whiteness persists stronger overtime. The AVC relies on this key feature to determine whether the cervix is suspect for precancer or cancer. In order to train and validate the AVC, 6000 women aged 30 to 49 years meeting the inclusion criteria will be recruited on a voluntary basis, with an estimated 100 CIN2+, calculated using a confidence level of 95% and an estimated sensitivity of 90% +/-7% precision on either side. Diagnostic test performance of AVC test and two current standard tests (VIA and cytology) used routinely for triage will be evaluated and compared. Histopathological examination will serve as reference standard. Participants’ and providers’ acceptability of the technology will also be assessed. The study protocol was registered under ClinicalTrials.gov (number NCT04859530). Expected results The study will determine whether AVC test can be an effective method for cervical cancer screening in LMICs.

Automated Cervical Cancer Screening Using a Smartphone-based Artificial Intelligence Classifier

Cervical cancer remains a major public health challenge in low- and middle-income countries (LMICs) due to financial and logistical issues. The World Health Organization (WHO) recommendation for cervical cancer screening in LMICs includes Human Papillomavirus (HPV) testing as primary screening followed by visual inspection with acetic acid (VIA) and treatment. However, VIA is a subjective procedure dependent on the healthcare provider's experience. Therefore, an objective approach based on quantitative diagnostic algorithms is desirable to improve performance of VIA. With this objective and in a collaboration between the Gynecology and Obstetrics Department of the Geneva University Hospital (HUG) and the Swiss Institute of Technology (EPFL), our group started the development of an automated smartphone-based image classification device called AVC (for Automatic VIA Classifier). Two-minute videos of the cervix are recorded during VIA and classified using an artificial neural network (ANN) and image processing techniques to differentiate precancer and cancer from non-neoplastic cervical tissue. The result is displayed on the smartphone screen with a delimitation map of the lesions when appropriate. The key feature used for classification is the dynamic of cervical acetowhitening during the 120 second following the application of acetic acid. Precancerous and cancerous cells whiten more rapidly than non-cancerous ones and their whiteness persists stronger overtime. Our aim is to assess the diagnostic performance of the AVC and to compare it with the performance of current triage tests (VIA and cytology). Histopathological examination will serve as reference standard. Participants' and providers' acceptability will also be considered as part of the study. The study will be nested in an ongoing cervical cancer screening program called "3T-approach" (for Test, Triage and Treat) which includes HPV self-sampling for women aged 30 to 49 years, followed by VIA triage and treatment if needed. The AVC will be evaluated in this context. The study's risk category is A according to swiss ethical guidelines. This decision is based on the fact that the planned measures for sampling biological material or collecting personal data entail only minimal risks and burdens.