[Application of artificial intelligence-assisted diagnosis for cervical liquid-based thin-layer cytology]

Area of Science:

• Gynecologic Oncology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Cervical cancer screening relies on accurate cytological analysis of liquid-based thin-layer smears.
• Traditional methods can be time-consuming and subject to inter-observer variability.
• Advancements in AI offer potential to augment diagnostic capabilities.

Purpose of the Study:

• To evaluate the clinical utility of an AI-assisted diagnostic system for cervical cancer screening based on The Bethesda System (TBS) reports.
• To compare the diagnostic performance and efficiency of the AI system versus traditional cytological analysis.

Main Methods:

• A deep learning-based AI system for TBS reports was developed using convolution neural networks.
• 16,317 clinical samples of cervical liquid-based thin-layer cytology smears were analyzed.
• Sensitivity, specificity, accuracy, and time consumption were evaluated against the 2014 TBS standard.

Main Results:

• The AI system demonstrated high sensitivity (92.90%) and specificity (87.02%) for predicting cervical intraepithelial and other lesions.
• Cytologists using the AI system achieved superior sensitivity (99.34%) and specificity (97.79% to 99.10%).
• AI assistance reduced smear reading time by approximately six-fold for cytologists.

Conclusions:

• The AI-assisted diagnostic system for cervical cytology exhibits high sensitivity, specificity, and generalization capabilities.
• Integrating AI into the screening workflow significantly enhances cytologists' accuracy and efficiency.
• AI holds promise for improving the overall effectiveness of cervical cancer screening programs.