A CAD system to analyse mammogram images using fully complex-valued relaxation neural network ensembled classifier

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate breast cancer classification from mammograms is crucial for early detection and treatment.
• Existing methods may face challenges in achieving high classification accuracy and efficiency.
• The MIAS database provides a standard benchmark for evaluating mammogram analysis techniques.

Purpose of the Study:

• To develop and evaluate an improved classification technique for mammogram images.
• To enhance the accuracy and efficiency of breast cancer detection using neural networks.
• To classify mammograms into Normal, Benign, and Malignant categories.

Main Methods:

• Feature extraction from the MIAS database, including Binary object, RST Invariant, Histogram, Texture, and Spectral Features.
• Development of an ensemble classifier based on Fully Complex-Valued Relaxation Neural Networks (FCRN).
• Evaluation of the system's performance using Receiver Operating Characteristic (ROC) analysis.

Main Results:

• The proposed FCRN-based ensemble technique demonstrated superior classification performance.
• Ensembled FCRN networks improved the overall classification rate compared to individual networks.
• The system achieved accurate output approximation with lower computational effort.

Conclusions:

• The ensembled FCRN classifier offers a highly accurate and computationally efficient solution for mammogram classification.
• This technique shows significant potential for improving automated breast cancer detection systems.
• Further performance comparisons across different training and testing datasets are provided.