Artificial intelligence in cervical cancer screening and triage: a role-stratified systematic review and bivariate meta-analysis.
Murat Cengiz, Onur Can Zaim, Bilal Esat Temiz et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and bivariate meta-analysis evaluated the accuracy of artificial intelligence (AI) tools in cervical cancer screening across different clinical roles. Among 97 studies, AI demonstrated the highest diagnostic accuracy during colposcopy (sensitivity 0.908, specificity 0.844). However, the triage of hrHPV-positive women showed the weakest performance, with a sensitivity of 0.805, indicating significant gaps in evidence for this application.
Key findings
- AI diagnostic assistance during colposcopy showed sensitivity of 0.908 and specificity of 0.844.
- Primary AI-cytology screening had high sensitivity of 0.934 but unstable specificity.
- Triage of hrHPV-positive women had a sensitivity of 0.805, below the commonly cited safety threshold.
- Only 10.9% of studies reported external validation.
Why it matters
Understanding the varying effectiveness of AI in cervical cancer screening can guide clinicians in selecting appropriate diagnostic tools and identifying areas needing further research, particularly in triage applications.