Computer-Aided Cervical Cancer Diagnosis Using Time-Lapsed Colposcopic Images.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical cancer is a leading cause of cancer deaths in women globally.
• Early detection of cervical intraepithelial neoplasia (CIN) is crucial for improving patient survival rates.

Purpose of the Study:

• To develop and validate a deep learning framework for accurate identification of LSIL+ (including CIN and cervical cancer) using time-lapsed colposcopic images.
• To compare the performance of different feature fusion approaches for improved diagnostic accuracy.

Main Methods:

• A deep learning framework with key-frame feature encoding and a feature fusion network was developed.
• Time-lapsed colposcopic images from 7,668 patients were used for training and validation.
• An Edge-feature Graph Convolutional Network (E-GCN) was employed as the optimal fusion method.

Main Results:

• The proposed framework achieved a classification accuracy of 78.33%, comparable to that of an experienced colposcopist.
• All tested fusion approaches outperformed existing single-time-slot automated cervical cancer diagnosis systems.
• The E-GCN approach demonstrated explainability consistent with clinical practice.

Conclusions:

• The developed deep learning framework shows significant potential for assisting in realistic clinical scenarios for cervical cancer diagnosis.
• Time-lapsed image analysis with advanced fusion techniques offers a promising avenue for improving early detection of cervical abnormalities.