Optimizing cervical cancer diagnosis with accurate cell classification using modified HDFF.

Area of Science:

• Oncology
• Medical Imaging
• Computer Science

Background:

• Cervical cancer (CC) remains a major global health concern, necessitating advanced diagnostic tools.
• Traditional cervical cell classification methods are labor-intensive and prone to errors, driving the need for automation.

Purpose of the Study:

• To introduce and evaluate a novel modified hierarchical deep feature fusion (HDFF) method for automated cervical cell classification.
• To assess the performance of the HDFF method on the SIPaKMeD and Herlev datasets for various classification tasks.

Main Methods:

• The study employed a modified hierarchical deep feature fusion (HDFF) approach.
• This method integrates hierarchical deep learning features to enhance classification accuracy and robustness.
• The HDFF method combines features from multiple deep learning model layers for improved performance.

Main Results:

• The Modified HDFF method achieved 98.88% accuracy in 2-class cervical cell classification, outperforming existing models.
• High precision, recall, and F1-scores were maintained in multi-class tasks, with 98.8% accuracy for 3-class and 98.5% for 7-class problems.
• The method demonstrated superior performance compared to RF-based hierarchical classification (98.43% in 2-class).

Conclusions:

• The Modified HDFF method presents a highly accurate and efficient tool for cervical cancer screening.
• Its strong performance across diverse classification tasks indicates significant potential for improving early detection rates.
• Further development with larger datasets could enhance its utility in automated cervical cancer detection systems.