Cervical Net: A Novel Cervical Cancer Classification Using Feature Fusion

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Cervical cancer is a prevalent and curable disease affecting women globally.
• Pap smear imaging is a standard screening technique for early detection.
• Accurate and efficient diagnostic tools are crucial for improving patient outcomes.

Purpose of the Study:

• To develop a computer-aided diagnosis system for cervical cancer detection.
• To utilize novel deep learning structures (Cervical Net) and feature fusion with Shuffle Net.
• To enhance the accuracy and efficiency of cervical cancer screening from Pap smear images.

Main Methods:

• Image acquisition, enhancement, feature extraction, and classification were performed.
• Automated features were extracted using pre-trained Convolutional Neural Networks (CNNs) fused with Cervical Net.
• Principal Component Analysis (PCA) and Canonical Correlation Analysis (CCA) were employed for dimensionality reduction and feature selection.

Main Results:

• A total of 544 features were extracted and refined using PCA and CCA.
• Five distinct machine learning algorithms were evaluated on the selected features.
• The proposed system, utilizing a Support Vector Machine (SVM), achieved a record accuracy of 99.1% for classifying five Pap smear image classes.

Conclusions:

• The novel Cervical Net deep learning structure combined with Shuffle Net features significantly improves cervical cancer screening accuracy.
• The computer-aided diagnosis system demonstrates high potential for reliable and automated detection of cervical cancer.
• This approach offers a promising advancement in early cervical cancer detection, potentially leading to better patient management.