Machine learning-enabled smartphone CRISPR-Cas12a lateral flow platform for sensitive detection of circulating HPV DNA.

Persistent infection with high-risk human papillomavirus (HPV) is a major cause of cervical cancer, and improved point-of-care (POC) detection is critical for early intervention. Although PCR-based assays are highly sensitive, their reliance on centralized laboratory infrastructure limits accessibility in decentralized settings. CRISPR-Cas diagnostics combined with lateral flow assays (LFA) offer a rapid alternative; however, visual interpretation of faint test bands remains subjective and variable. Here, we developed a smartphone-based CRISPR-Cas12a LFA platform integrated with an interpretable machine learning (ML) framework for quantitative detection of circulating HPV DNA in plasma. Standardized image acquisition was implemented using a light-controlled enclosure, and radiomics-inspired features were analyzed using a multivariable logistic regression model. The system was trained on 150 plasma samples and validated in an independent cohort of 60 samples. The optimized model achieved 96.7% sensitivity and 100% specificity, outperforming visual interpretation, particularly for low-signal samples. Performance remained stable across different smartphone models, lighting conditions, and operators, with rapid on-device inference enabling consistent and reliable operation. This integrated CRISPR-LFA platform demonstrates accurate and reproducible detection of circulating HPV DNA and supports feasibility for POC applications, pending further validation in broader clinical settings.