Introduction Colposcopic examination requires sufficient training to detect cervical intraepithelial neoplasia (CIN) with (1) high diagnostic accuracy and (2) minimizing time and reducing tissue biopsies. This study aimed to develop an artificial intelligence (AI) based system which replicates expert colposcopic examination techniques, independent of examiner skill.

Methods A retrospective analysis was performed using 8341 colposcopic videos from 2013 to 2019, consisting of seven cases of early-stage cervical cancer, 203 cases of CIN3, and 456 cases of CIN1. An AI-based lesion detection model was developed to identify major abnormal colposcopic findings. The model was first trained using annotated colposcopic findings with the highest acetic acid intensity in cervical cancer and CIN3 cases whose histological diagnoses were confirmed by biopsies. The developed AI model was then applied to CIN1 cases and the diagnostic accuracy of the lesions was evaluated.

Results The AI-based model identified major abnormal colposcopic findings in cervical cancer and CIN3 cases with an area under the curve (AUC) of 0.89 for lesion area and 95% accuracy for number of lesions identified. The model also predicted minor abnormal colposcopic findings in CIN1 cases, with an AUC of 0.81 for detection of lesion area and 93% for identification of number of lesions. In addition, a heat map display based on the prediction results allowed visualization of the area of highest acetic acid intensity corresponding to the actual biopsy locations.

Conclusion/Implications Newly developed AI-based diagnostic system for colposcopy could identify CIN lesions with high accuracy and suggest appropriate biopsy sites.