DeepCervix: A deep learning-based framework for the classification of cervical cells using hybrid deep feature fusion...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Cervical cancer is a leading cause of death among women, preventable through early detection via screening.
• Traditional Pap smear screening is prone to human error, leading to high false-positive rates.
• Existing machine learning and deep learning models for cervical cell classification face challenges with cell clustering, segmentation, and imbalanced datasets.

Purpose of the Study:

• To develop an accurate deep learning-based computer-aided diagnostic system for cervical cell classification.
• To address limitations of existing methods, including the need for pre-segmented images and poor performance on imbalanced data.
• To propose a hybrid deep feature fusion (HDFF) technique for enhanced cervical cell classification.

Main Methods:

• Developed DeepCervix, a hybrid deep feature fusion (HDFF) technique utilizing multiple deep learning models.
• Employed various deep learning architectures to capture diverse cellular features and improve classification performance.
• Tested the HDFF method on the SIPaKMeD and Herlev public datasets.

Main Results:

• Achieved state-of-the-art classification accuracy on the SIPaKMeD dataset: 99.85% (2-class), 99.38% (3-class), and 99.14% (5-class).
• Demonstrated high accuracy on the Herlev dataset: 98.32% (2-class) and 90.32% (7-class).
• Outperformed base deep learning models and late fusion methods in classification accuracy.

Conclusions:

• The proposed DeepCervix HDFF technique significantly enhances the accuracy of cervical cell classification.
• This AI-driven approach offers a promising solution to improve the reliability of cervical cancer screening.
• The open-source availability of the DeepCervix model facilitates further research and clinical application.