Cervix Type and Cervical Cancer Classification System Using Deep Learning Techniques

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Cervical cancer is a leading cause of cancer death globally, particularly in developing nations.
• Current screening methods like visual inspection, Pap tests, and HPV tests face challenges including observer variability and limited accessibility.
• Accurate and timely diagnosis is crucial for effective cervical cancer management.

Purpose of the Study:

• To develop an integrated deep learning system for automated cervix type and cervical cancer classification.
• To overcome limitations of manual diagnosis, such as inter- and intra-observer variability.
• To create a robust tool for supporting cervical cancer diagnosis, especially in resource-limited areas.

Main Methods:

• Utilized a dataset of 4005 colposcopy and 915 histopathology images.
• Employed MobileNetv2-YOLOv3 for region of interest (ROI) extraction in cervix images.
• Applied EfficientNetB0 for both cervix type classification and cervical cancer classification on histopathology images.

Main Results:

• Achieved 99.88% mean average precision (mAP) for ROI extraction.
• Attained 96.84% test accuracy for cervix type classification.
• Reached 94.5% test accuracy for cervical cancer classification.

Conclusions:

• The developed deep learning system demonstrates high accuracy in classifying cervix types and detecting cervical cancer.
• The system can serve as a valuable decision support tool, particularly in regions with limited healthcare resources and expertise.
• Automated classification holds promise for improving cervical cancer screening and diagnosis efficiency.