DeepCervix

DeepCervix performs automated classification of cervical cell images using deep learning to improve cervical cancer screening and early detection of precancerous lesions.

Key Features:

• Hybrid Deep Feature Fusion (HDFF): Integrates deep features from multiple deep learning (DL) models using a hybrid deep feature fusion technique to improve classification accuracy and reliability.
• Multi-model feature integration: Combines features extracted by various DL models to capture comprehensive information from cervical cell images, addressing cell clustering and uneven data distribution.
• No requirement for pre-segmented images: Operates on raw cervical cell images without requiring pre-segmentation.
• Robustness to data imbalance: Is specifically designed to maintain high performance when class distributions in cervical cell datasets are uneven.
• Validated multiclass performance: Demonstrates high reported accuracies on benchmark datasets, indicating strong multiclass classification capability.

Scientific Applications:

• Computer-aided diagnosis (CAD) for cervical cancer screening: Provides automated cervical cell classification to support Pap smear analysis and reduce manual interpretation errors.
• Early detection and stratification of precancerous lesions: Enables multiclass cytological categorization relevant to early detection and clinical stratification of lesions.

Methodology:

DeepCervix trains multiple deep learning models to extract deep features and fuses them using a hybrid deep feature fusion (HDFF) technique; performance was validated on the SIPaKMeD dataset (99.85% 2-class, 99.38% 3-class, 99.14% 5-class) and the Herlev dataset (98.32% 2-class, 90.32% 7-class).