AVE — AI-Assisted Cervical Precancer Screening in Zambia's Public Health Facilities

Automated Visual Evaluation (AVE) is an AI image-classification tool, originally developed with the U.S. National Cancer Institute, that analyses smartphone or digital-camera photographs of the cervix to flag possible precancerous lesions, intended to assist rather than replace the health worker performing the exam. Between October 2019 and December 2021, Zambia's Ministry of Health, working with NCI, Global Health Labs and Unitaid, ran an internal validation study across eight public health facilities (seven in Lusaka, one in Kitwe), enrolling 8,204 women aged 25-55 through the country's public-sector cervical cancer prevention programme.

Assess how accurately AVE identifies histology-confirmed precancerous and cancerous cervical lesions (CIN2+) in a real public-health-facility setting, and compare its diagnostic performance against standard visual inspection with acetic acid (VIA).

The Zambia study confirmed AVE results against histopathology for the 8,204 enrolled women. A separate prospective, observational diagnostic-accuracy study ran across government health facilities in Malawi, Rwanda, Senegal, Zambia and Zimbabwe between March 2022 and January 2023, comparing AVE's detection of CIN2+ lesions against naked-eye VIA.

In Zambia, AVE identified CIN2+ lesions with an AUC of 0.91, translating to 85% sensitivity (95% CI 81-90%) and 86% specificity (95% CI 84-88%). In the five-country study, AVE lifted sensitivity for detecting CIN2+ lesions to 60.1% (95% CI 55.5-64.5), compared with 36.6% (95% CI 32.2-41.1) for VIA alone, with an accepted trade-off in specificity.

Both studies were explicitly framed as diagnostic-accuracy research rather than a finished national screening rollout; the study authors say the specificity trade-off makes clinician oversight of every AI-flagged case essential, and no source found describes AVE as deployed beyond the study facilities.