External validation of AI assisted colposcopy using WHO dataset for cervical precancer and cancer detection

Artificial intelligence (AI)-guided colposcopy has previously been developed to assist colposcopists in detecting cervical intraepithelial neoplasia grade 2 or worse (CIN2+) and guiding biopsies, but internal testing of the AI model could overestimate diagnostic performance. We validated the performance of the AI-guided colposcopy on an independent WHO open-source set of 187 patients with 855 colposcopic images. Forty-five colposcopists with different levels of experience from 12 regions in China participated in the validation set. The sensitivity of the AI system on CIN2+ standalone diagnosis were 84.2% (95%CI 74.4–90.7%). With AI assistance, colposcopists sensitivity increased from 84.8% (95% CI, 82.7–86.9%) to 90.6% (95% CI, 88.7–92.1%). The improvement was most evident among low-experienced colposcopists, with sensitivity increasing by 6.5%, and AUC from 0.72 (95% CI, 0.68–0.77) to 0.76 (95% CI, 0.72–0.80; p = 0.043). AI-guided biopsy support was also associated with a reduction in the mean number of biopsies per case, from 2.48 to 2.02. These findings highlight the AI system’s robust performance on an independent external dataset and its capacity to bridge the diagnostic experience gap among readers, supporting its clinical implementation as a valuable adjunct for colposcopic assessment.