The value of artificial intelligence in the diagnosis of precancerous cervical lesions

BACKGROUND: Cervical cancer is one of the leading causes of cancer-related mortality among women aged 30–34. Late diagnosis remains a critical issue, with >40% of cervical cancer cases being first detected at stage III–IV. Although extended colposcopy is an important step in the diagnosis of precancerous conditions and cervical cancer, obtaining a colposcopic report may be challenging due to a lack of equipment or qualified specialists in some settings. Therefore, AI-based analysis of colposcopic images is a promising component of comprehensive cervical disease diagnosis.

AIM: To analyze the efficacy of comprehensive diagnosis and detection of precancerous cervical lesions and cervical cancer.

METHODS: In 2024–2025, at the Department of Obstetrics and Gynecology No. 1 of the N.V. Sklifosovsky Institute of Clinical Medicine, Sechenov University, we analyzed data from 42 patients aged 18–45 years who had grade 2 abnormal colposcopic findings on routine extended colposcopy. All patients underwent a comprehensive workup, including history taking, gynecological examination, clinical description, HPV testing, Papanicolaou (Pap) cytology of ecto- and endocervical scrapings, and extended colposcopy. Anonymized colposcopic images were uploaded to AI software, which processed the images without a predefined diagnosis and generated an automated conclusion for each case without human intervention. The system’s conclusion was then compared with the consensus conclusion of a medical board comprising three obstetrician-gynecologists.

RESULTS: Agreement between the AI system and the medical board was 92.8% (39 cases). Discrepancies were observed in only 3 cases (7.2%), in which the medical board classified the colposcopic images as grade 2 abnormal, whereas the AI-based software identified grade 1 abnormal findings.

CONCLUSION: AI-based interpretation of extended colposcopy data can identify patients at risk of cervical disease for further examination and early detection of precancerous changes and cervical cancer. In our study, the sensitivity of the diagnostic method was 92.9%; however, it should be emphasized that for clinical diagnosis, AI findings in at-risk patients must be confirmed by histological examination.