Refining cell classification for cervical cancer detection using an improved high dimensional feature fusion approach

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Accurate cervical cell classification is crucial for early cervical cancer detection.
• Traditional methods like Pap smear analysis are time-consuming and prone to errors.
• Existing deep learning models often lack dataset generalization and struggle with multi-class classification.

Purpose of the Study:

• To develop a robust deep learning framework for improved cervical cell image classification.
• To overcome limitations of existing models in generalization and multi-class performance.
• To enhance the accuracy and reliability of automated cervical cancer screening.

Main Methods:

• A Modified High-Dimensional Feature Fusion (HDFF) framework was introduced.
• Feature vectors from seven diverse pre-trained Convolutional Neural Network (CNN) architectures were integrated.
• A fully connected classifier with dropout and batch normalization processed the fused features.

Main Results:

• The model achieved up to 99.85% accuracy in binary classification tasks.
• High precision, recall, F1-score, specificity, and AUC were maintained in multi-class settings.
• Demonstrated lower misclassification rates and stable performance across datasets and class imbalances.

Conclusions:

• The Modified HDFF framework is robust and adaptable for cervical cancer screening.
• The model's ability to generalize across datasets highlights its clinical relevance.
• This approach offers significant diagnostic value for early cervical cancer detection.