Utilising deep learning networks to classify ZEB2 expression images in cervical cancer

Area of Science:

• Oncology
• Biotechnology
• Artificial Intelligence

Background:

• Cervical cancer remains a leading cause of cancer deaths in women, particularly in underserved regions.
• Lack of accessible screening and follow-up care exacerbates the problem.
• Zinc finger E-box-binding homeobox 2 (ZEB2) is a potential biomarker for cervical tumor aggressiveness and progression.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning system for classifying cervical cancer images based on ZEB2 expression.
• To enhance the interpretability of deep learning models using visualization techniques.

Main Methods:

• A hybrid deep learning system was created by integrating EfficientNet, DenseNet, and InceptionNet models using ensemble voting.
• Gradient-weighted class activation mapping (Grad-CAM) was employed for model interpretability.
• A dataset of 649 annotated cervical cancer images was utilized for training, validation, and testing.

Main Results:

• The hybrid deep learning model achieved a high classification accuracy of 94.4% on the test dataset.
• Grad-CAM visualizations highlighted critical image regions influencing ZEB2 expression classification.
• The model demonstrated an effective and interpretable approach to ZEB2 expression analysis.

Conclusions:

• The proposed hybrid deep learning model offers a promising tool for the accurate and interpretable classification of cervical cancer based on ZEB2 expression.
• This approach has the potential to significantly improve early diagnosis rates, leading to better patient outcomes and reduced healthcare costs.
• Future research will focus on refining model accuracy and exploring its application to other cancer types.