Abstract
Introduction Colposcopy has limited sensitivity for identification of Cervical intraepithelial Neoplasia in need of preventive treatment (CIN2+) even in skilled hands and when digital technologies are used. Artificial Intelligence models offer the potential to enhance diagnostic accuracy and reduce the dependence on trained colposcopists.
Methods We recorded colposcopies by the Dysis Digital Colposcope from women with a Transformation Zone type 1 or 2. Images were mapped with location of 3-4 biopsies: - Biopsy at the colposcopist’s discretion (CD) - Biopsy marked by the Dysis Color Map (DCM) indicating low or high-grade lesions - Two random biopsies Biopsies were histologically analyzed separately and classified as Class 1 (normal, inflammation, CIN1) or Class 2 (CIN2+: CIN2/3,AIS or Carcinoma). A Cervix-AID-NET model was developed and tested on this material.
Results The median age of the 178 women was 30 years. Colposcopy indications were low grade cytology (n=80), high grade cytology (n=57), follow-up on CIN2 (n=31) and other reasons (n=10). The diagnostic accuracy of Cervix-AID-NET for CIN2+ was 99.8% compared to 58.4% for DCM and 55.1% for CD. The sensitivity, specificity and negative predictive value (NPV) of the AI model for CIN2+ were 99,9%, 99,8% and 99.9%. respectively.
Conclusion/Implications Conclusion and implication: Cervix-AID-NET demonstrated superior accuracy in detection of CIN2+ lesions compared to the performance of a digital colposcope and the subjective evaluation by the colposcopist. The high diagnostic accuracy support a ‘See and Treat’ strategy in patients with a Class 2 colposcopy. Women with a Class 1 colposcopy could be safely managed without biopsies.
Methods We recorded colposcopies by the Dysis Digital Colposcope from women with a Transformation Zone type 1 or 2. Images were mapped with location of 3-4 biopsies: - Biopsy at the colposcopist’s discretion (CD) - Biopsy marked by the Dysis Color Map (DCM) indicating low or high-grade lesions - Two random biopsies Biopsies were histologically analyzed separately and classified as Class 1 (normal, inflammation, CIN1) or Class 2 (CIN2+: CIN2/3,AIS or Carcinoma). A Cervix-AID-NET model was developed and tested on this material.
Results The median age of the 178 women was 30 years. Colposcopy indications were low grade cytology (n=80), high grade cytology (n=57), follow-up on CIN2 (n=31) and other reasons (n=10). The diagnostic accuracy of Cervix-AID-NET for CIN2+ was 99.8% compared to 58.4% for DCM and 55.1% for CD. The sensitivity, specificity and negative predictive value (NPV) of the AI model for CIN2+ were 99,9%, 99,8% and 99.9%. respectively.
Conclusion/Implications Conclusion and implication: Cervix-AID-NET demonstrated superior accuracy in detection of CIN2+ lesions compared to the performance of a digital colposcope and the subjective evaluation by the colposcopist. The high diagnostic accuracy support a ‘See and Treat’ strategy in patients with a Class 2 colposcopy. Women with a Class 1 colposcopy could be safely managed without biopsies.
Original language | English |
---|---|
Journal | International Journal of Gynecological Cancer |
Volume | 34 |
Issue number | Suppl. 3 |
Pages (from-to) | A51-A52 |
ISSN | 1048-891X |
DOIs | |
Publication status | Published - 2024 |
Event | IGCS 2024 Annual Meeting - Dublin, Ireland Duration: 16. Oct 2024 → 18. Oct 2024 |
Conference
Conference | IGCS 2024 Annual Meeting |
---|---|
Country/Territory | Ireland |
City | Dublin |
Period | 16/10/2024 → 18/10/2024 |