CerviNet: A Novel Approach for Cervical Cancer Classification Using Pap-Smear Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Cervical cancer is a significant global health concern, ranking as the fourth most common disease in women.
• Pap smear imaging is crucial for detecting cervical, vaginal, and vulvar abnormalities.
• Manual assessment of pap smears presents challenges, driving the need for automated diagnostic systems.

Purpose of the Study:

• To introduce an innovative hybrid deep learning model for accurate cervical cell categorization.
• To enhance the model's adaptability and reliability through advanced data enhancement techniques.
• To improve the precision of cervical cancer classification using deep learning.

Main Methods:

• Employed data enhancement: resampling for class imbalance and augmentation (flips, rotations) for diversity.
• Utilized Vision Transformer (ViT) with linear projection and position embedding for image patching and encoding.
• Integrated convolutional layers and a fully connected layer into a fusion architecture for feature enhancement.
• Applied median smoothing and Gaussian filtering for image preprocessing.

Main Results:

• The hybrid model demonstrated high accuracy in cervical cancer classification across multiple datasets.
• Achieved 98.07% accuracy for 2-state classification on the Herlev dataset.
• Achieved 98.08% accuracy for 3-state classification on the SIPaKMeD dataset.
• The model's performance indicates robustness and potential for clinical application.

Conclusions:

• The proposed hybrid deep learning model effectively categorizes cervical cells with high precision.
• Advanced data enhancement and a fusion architecture contribute to the model's accuracy and generalization.
• The model shows significant promise for practical clinical use in diagnosing cervical cancer.