Deep convolutional neural network and emotional learning based breast cancer detection using digital mammography

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early breast cancer detection significantly improves patient survival rates.
• Mammography is a key tool for radiologists in identifying early-stage breast cancer.
• Automated systems can enhance diagnostic confidence and provide objective evidence.

Purpose of the Study:

• To propose an automated Diverse Features based Breast Cancer Detection (DFeBCD) system for classifying mammograms.
• To evaluate the effectiveness of static and dynamically extracted features for breast cancer detection.
• To compare the performance of Support Vector Machine (SVM) and an Emotional Learning inspired Ensemble Classifier (ELiEC).

Main Methods:

• Utilized four distinct feature sets: taxonomic indexes, statistical measures, local binary patterns, and dynamically extracted features via a highway-network based deep convolution neural network (CNN).
• Trained two classifiers, SVM and ELiEC, on these features using the IRMA mammogram dataset.
• Employed 5-fold cross-validation to ensure reliable performance evaluation.

Main Results:

• Dynamically generated features from the highway network-based CNN outperformed individual static feature sets.
• Hybridizing all four feature types improved system performance by 2-3%.
• The ELiEC classifier demonstrated superior performance compared to SVM when using both hybrid and dynamic features.

Conclusions:

• The proposed DFeBCD system effectively detects abnormalities in mammograms.
• Deep learning-based dynamic feature extraction significantly enhances breast cancer detection accuracy.
• The ELiEC classifier offers improved performance for automated breast cancer diagnosis, especially with complex feature sets.