Significance of features in classification of breast abnormality patterns in digital mammography

The research work presented in this thesis is concerned with the computer-aided classification of breast abnormality patterns in digital mammograms to facilitate early diagnosis of breast cancer via digital mammography, a matter of great interest due to its benefits to society.

A novel neural pattern characterization and classification (NPCC) technique is proposed to determine the significance of different types of features for effective classification of malignant and benign class patterns of calcification and mass type of breast abnormalities patterns in digital mammography. Grey level features, BIRADS features, patient age features and the subtlety value features have been evaluated using the proposed NPCC technique to find their significance in the classification of breast abnormality patterns. The research also explores the auto-association and classification ability of the neural network approach to achieve the maximum classification on both types of breast abnormality patterns.

The proposed technique was implemented and tested on benchmark data. This provides a basis for performing comparison with previously reported results. The proposed NPCC technique attained high classification accuracy and identifies the significant types of features for consistent classification of breast abnormality patterns. The experimental results suggest that the proposed NPCC technique is promising and performs competently against the other existing techniques.