Multi-stage feature selection (MSFS) algorithm for UWB-based early breast cancer size prediction.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Breast cancer is a leading cause of death in women globally, necessitating early detection for optimal treatment outcomes.
• Accurate breast cancer size detection is critical for effective medical intervention and patient management.
• Existing detection methods require enhancement for improved accuracy and efficiency.

Purpose of the Study:

• To propose a novel multi-stage feature selection method for enhanced breast cancer size detection.
• To investigate the efficacy of ultra-wideband (UWB) signals in breast cancer detection.
• To evaluate the performance of different machine learning classifiers using novel hybrid feature datasets.

Main Methods:

• Utilized ultra-wideband (UWB) signals transmitted through breast phantoms, captured in time and frequency domains.
• Developed a four-stage feature selection algorithm: data normalization, feature extraction, dimensionality reduction, and feature fusion.
• Created three hybrid feature datasets (8-HybridFeature, 9-HybridFeature, 10-HybridFeature) for classification.

Main Results:

• The 8-HybridFeature dataset demonstrated superior performance in breast cancer size classification.
• The Naïve Bayes classifier achieved the highest accuracy (91.98%) with the 8-HybridFeature dataset.
• Compared to Support Vector Machine (90.44%) and Probabilistic Neural Network (80.05%), Naïve Bayes showed better classification accuracy.

Conclusions:

• The proposed multi-stage feature selection method, utilizing UWB signals and hybrid features, is effective for early breast cancer size detection.
• The 8-HybridFeature dataset combined with the Naïve Bayes classifier offers a promising approach for accurate breast cancer diagnosis.
• The developed method, visualized in MATLAB, provides a robust tool for breast cancer research and clinical application.