Automated visual evaluation (AVE) has long been seen as a potential game changer for cervical cancer screening in low resource settings, but progress has stalled due to a lack of robust clinical tools. A new study in *Nature Medicine* argues that frugal AI, efficient, multimodal models designed for real world workflows, could be the missing piece. The approach prioritizes explainability, local calibration, and practical deployment, offering a path to scalable and affordable screening solutions where they are needed most.
Cervical cancer remains one of the most preventable yet deadly cancers globally, with nearly 90% of deaths occurring in low and middle income countries. Traditional screening methods, such as Pap smears and HPV testing, require infrastructure, trained personnel, and laboratory resources that are often unavailable in these settings. Automated visual evaluation (AVE), which uses AI to analyze cervical images for precancerous lesions, has been proposed as a solution. However, despite years of research, AVE has not yet delivered clinically reliable tools for widespread use.
The latest analysis in *Nature Medicine* highlights why previous efforts have fallen short. Many AVE models were developed using high quality images from well resourced settings, making them ill suited for real world conditions where lighting, equipment, and patient populations vary. Additionally, most models operate as black boxes, providing little transparency in their decision making, a critical flaw for clinical adoption.
DISCUSSION (0)