Computer-based identification of breast cancer using digitized mammograms

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• High-quality mammography is crucial for breast cancer screening.
• Improving mammography technology, administration, and interpretation enhances diagnostic accuracy.
• Computer-based intelligent systems offer potential benefits for breast cancer diagnosis and management.

Purpose of the Study:

• To compare the performance of two distinct computer-based classifiers for mammogram analysis.
• To evaluate the effectiveness of a feedforward neural network and a Gaussian mixture model (GMM) in classifying mammographic images.
• To assess the diagnostic accuracy of these systems in differentiating between normal, benign, and cancerous breast conditions.

Main Methods:

• Feature extraction from raw mammogram images using image processing techniques.
• Classification of extracted features using a feedforward architecture neural network.
• Classification of extracted features using a Gaussian mixture model (GMM).
• Comparative analysis of the two classifiers' performance on a dataset of 360 subjects.

Main Results:

• Both the neural network classifier and the Gaussian mixture model (GMM) demonstrated high performance.
• Sensitivity and specificity exceeding 90% were achieved for both classification methods.
• The study provides a comparative evaluation of two advanced computational approaches for mammogram analysis.

Conclusions:

• Computer-based intelligent systems, specifically neural networks and GMMs, are highly effective for mammogram classification.
• These systems show significant potential in aiding the diagnosis and management of breast cancer.
• The developed methods achieve high sensitivity and specificity, supporting their clinical utility in breast cancer screening.