Novel network architecture and learning algorithm for the classification of mass abnormalities in digitized...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Digitized mammograms are crucial for breast cancer screening.
• Accurate classification of mass abnormalities is essential for early diagnosis.
• Existing methods face challenges in balancing memorization and generalization.

Purpose of the Study:

• To introduce a novel learning algorithm for classifying mass abnormalities in digitized mammograms.
• To enhance the diagnostic accuracy of mammogram analysis using artificial intelligence.

Main Methods:

• A new neural network architecture with additional neurons in the hidden layer was developed.
• A hybrid training strategy combining minimal distance-based similarity and direct output weight calculation was employed.
• Grey-level and Breast Imaging Reporting and Data System (BI-RADS) features were extracted from mammograms.

Main Results:

• The algorithm achieved 100% retrieval accuracy on training patterns.
• High generalization accuracy was obtained for unseen patterns, with a 94% accuracy on the test set.
• The approach demonstrated superior performance compared to existing classification methods.

Conclusions:

• The novel learning algorithm shows significant promise for mass abnormality classification in mammograms.
• The proposed method outperforms existing approaches in accuracy, generalization, and memorization.
• This technique offers a potential advancement in automated breast cancer detection systems.