The Performance of Artificial Intelligence in Cervical Colposcopy: A Retrospective Data Analysis
Yuqian Zhao1, Yucong Li2,3, Lu Xing2
1Center for Cancer Prevention Research, Sichuan Cancer Hospital and Institute, Sichuan Cancer Center, School of Medicine, University of Electronic Science and Technology of China, Chengdu 610041, China.
Journal of Oncology
|January 17, 2022
Summary
An artificial intelligence (AI) system demonstrated comparable sensitivity to human colposcopists for detecting high-grade precancerous lesions (CIN2+ and CIN3+). AI-assisted colposcopy significantly enhanced sensitivity for CIN2+ detection.
Area of Science:
- Gynecology
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Cervical cancer screening relies on detecting high-grade precancerous lesions, such as cervical intraepithelial neoplasia grade 2 (CIN2+) and grade 3 (CIN3+).
- Accurate detection of these lesions is crucial for effective treatment and prevention of invasive cervical cancer.
- Colposcopy, guided by human expertise, is a key diagnostic tool, but its performance can vary.
Purpose of the Study:
- To evaluate the diagnostic performance of an artificial intelligence (AI) system in identifying high-grade precancerous cervical lesions.
- To compare the AI system's performance against human colposcopists and AI-assisted colposcopy in detecting CIN2+ and CIN3+.
Main Methods:
- A retrospective diagnostic study analyzed anonymized medical records from 346 women at Chongqing Cancer Hospital.
- Data included cytology, HPV testing, colposcopy findings (images), and histopathology results.
- Sensitivity, specificity, and AUC were calculated for the AI system, AI-assisted colposcopy, and human colposcopists in detecting CIN2+ and CIN3+.
Main Results:
- In detecting CIN2+, AI-assisted colposcopy significantly increased sensitivity (96.6%) compared to human colposcopists (88.8%), but with lower specificity (38.1% vs. 59.5%).
- The AI system showed comparable sensitivity to human colposcopists for both CIN2+ and CIN3+ detection.
- For CIN3+ detection, AI-assisted colposcopy also showed higher sensitivity (97.5%) than human colposcopists (92.6%), though specificity remained lower.
Conclusions:
- The AI system offers sensitivity for detecting CIN2+ and CIN3+ that is on par with human colposcopists.
- AI-assisted colposcopy demonstrates a significant improvement in sensitivity for detecting CIN2+ lesions.
- Further research may explore optimizing AI algorithms to balance sensitivity and specificity in cervical precancerous lesion detection.
