Diagnostic Accuracy of AI-Assisted Colposcopy for Detection of Cervical Precancerous Lesions at Buddhachinaraj Hospital

Abstract

Objective: To evaluate the diagnostic accuracy of an artificial intelligence (AI)-assisted colposcopy imaging system for detecting cervical precancerous lesions (CIN2+ and CIN1+) compared with conventional colposcopy.

Methods: This prospective diagnostic accuracy study included women aged 25–60 years who had abnormal human papillomavirus (HPV) screening results requiring colposcopic evaluation at Buddhachinaraj Phitsanulok Hospital. The primary outcome was the diagnostic accuracy of AI-assisted versus conventional colposcopy for detecting CIN2+ lesions; the secondary outcome was diagnostic accuracy for detecting CIN1+ lesions. A total of 165 women underwent colposcopy. Histopathological findings from cervical biopsies served as the reference standard. Diagnostic performance was assessed using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and the area under the receiver operating characteristic curve (AUROC).

Results: The mean age of participants was 39.5 ± 8.3 years. Histopathological examination identified CIN1 in 50.3% and CIN2+ lesions in 41.8% of participants. For CIN2+, AI-assisted colposcopy had an accuracy of 69.1%, sensitivity of 26.1%, specificity of 100%, PPV of 100%, and NPV of 65.3%, whereas conventional colposcopy had an accuracy of 77.6%, sensitivity of 50.7%, specificity of 96.9%, PPV of 92.1%, and NPV of 73.2%. The AUROC was 0.630 for AI-assisted colposcopy and 0.738 for conventional colposcopy. For CIN1+, AI-assisted colposcopy demonstrated an accuracy of 84.8%, sensitivity of 84.9%, specificity of 84.6%, PPV of 98.5%, and NPV of 32.4%, compared with 88.5%, 88.8%, 84.6%, 98.5%, and 39.3%, respectively, for conventional colposcopy. The AUROC values were 0.847 and 0.867 for AI-assisted and conventional colposcopy, respectively.

Conclusion: AI-assisted colposcopy demonstrated less diagnostic performance for CIN2+ compared with conventional colposcopy but comparable performance for CIN1+. AI-assisted colposcopy should therefore be considered an adjunct to, rather than a replacement for, conventional colposcopic assessment in clinical practice.