Combining HPV testing with artificial intelligence–based image analysis has shown efficacy in detecting precancerous lesions and guiding treatment decisions, according to a study led by the Barcelona Institute for Global Health (ISGlobal), a centre supported by “la Caixa” Foundation and the National Institutes of Health in the USA (NIH). The research, conducted within the HPV-Automated Visual Evaluation (PAVE) Consortium across nine countries, has been published in The Lancet Obstetrics, Gynaecology & Women's Health .
Despite global efforts such as the WHO cervical cancer elimination initiative, screening coverage remains below 15% in many low- and middle-income countries . Existing approaches, including visual inspection with acetic acid, often lack accuracy, while access to high-quality HPV testing and follow-up services remains uneven. To address these gaps, the PAVE study evaluated a strategy that integrates self-collected HPV testing with extended genotyping and automated visual evaluation (AVE) of cervical images . This approach is designed to be scalable and adaptable to settings with limited infrastructure, while maintaining high diagnostic performance.
The study enrolled more than 50,000 women aged 25 to 49 years between 2022 and 2025 in Brazil, Cambodia, Dominican Republic, El Salvador, Eswatini, Honduras, Malawi, Nigeria, and Tanzania. Participants provided self-collected samples for HPV testing capable of identifying thirteen carcinogenic virus types. Women who tested positive underwent cervical imaging using a portable colposcope, followed by biopsy to confirm diagnosis. The images were analysed by a deep-learning Automated Visual Evaluation (AVE) algorithm , which classified findings as severe, indeterminate, or normal.
“Among nearly 49,000 women with valid results, more than one in four tested positive for HPV , and most attended follow-up assessment,” explains Silvia de Sanjosé , associate researcher at ISGlobal and first author of the study. “We identified over 500 cases of severe precancerous lesions —known as CIN3+ , the most advanced stage before cervical cancer—with higher prevalence among women living with HIV.”
The risk of severe lesions increased according to both HPV type and the severity of abnormalities detected by the AI system. Infections with HPV16 (the most carcinogenic type) showed the highest risk, particularly when combined with severe findings from automated image analysis.
The findings have proved the efficacy of the strategy that combines self-sampling, HPV extended genotyping and AVE in detecting CIN3+ lesions . The model can improve the efficiency of cervical cancer screening by enabling risk-based management. This is particularly relevant in low-resource settings , where large numbers of HPV-positive women require triage but healthcare capacity is limited.
“Our results demonstrate that HPV testing and automated visual evaluation provide complementary information that can guide clinical decisions,” adds de Sanjosé . The approach also enables self-sampling , rapid testing, and real-time image analysis, facilitating a same-day screen–triage–treat model, an important advantage in settings where follow-up visits are difficult.
“Even as HPV vaccination expands globally, effective screening will remain essential for decades , particularly for populations that have not benefited from vaccination”, explains Mark Schiffman , epidemiologist at the National Health Institute (NIH) and co-leader of the study. “The PAVE approach offers a feasible solution to extend high-quality cervical cancer screening to underserved populations, contributing to global efforts to reduce cervical cancer incidence and mortality.”
Reference
de Sanjosé, S., Inturrisi, F., Perkins, R. B., Egemen, D., Rodriguez, A. C., Befano, B., Guillen, D., Jeronimo, J., Campos, N., Ribeiro, A., Ajenifuja, O., Yeates, K., Cremer, M., Goldstein, A., Chen, H., Figueroa, J., Madeleine, M., Norris, T., Raiol, T., . . . Schiffman, M. (2026). Extended genotyping of human papillomavirus and AI-assisted cervical visual evaluation for prediction of cervical precancer in nine resource-limited settings (PAVE): A cross-sectional study. The Lancet Obstetrics, Gynaecology, & Women’s Health, 2 , e985–e994. https://doi.org/10.1016/S3050-5038(26)00231-1
The Lancet
Observational study
People
MC reports being the principal investigator of the National Cancer Institute (NCI) and National Institutes of Health grants (grant numbers R01 Ca266059 and R44CA272177) paid to the Cleveland Clinic. That grant support includes a travel stipend paid to her institution. MC is also President and founder of the non-governmental organisation Basic Health International. IP received a contract between the NCI and the Enlace Hispanoamericano de Salud (EHAS) Foundation, where EHAS provides data collection tools for this work. IP declares funds received from the Spanish Agency for International Development Cooperation (AECID) for the development projects related to cervical cancer prevention in Guatemala but not related to this work; and an unpaid role as a director of the EHAS Foundation in Spain. KA is an investigator of the National Institute of Health, USA (grant numbers R44CA272177 and RO-CA266059). RB received from the NCI, USA, a contract for gynaecology expertise and travel as well as meeting expenses for the roundtable on cervical cancer prevention in March, 2026, from the American Cancer Society; and reports an unpaid role as leadership related to cervical cancer prevention guidelines at the American Cancer Society and Enduring Guidelines Committee. ZR is Vice-President of HPV Global Action and runs the Scientific Affairs section as an unpaid service. ZR also received travel meeting support from HPV Global Action. AL has been contracted by Liger Medical through Brooksee for engineering services. Salary was partially funded by Liger Medical from 2023 to 2024 and full-time funded since 2025, and Liger Medical paid for travel to Cambodia, Malawi, and Brazil and attendance to international papillomavirus conferences in 2024 and 2025. All other authors declare no competing interests.