An Adaptive Behavioral Learning Technique based Bilateral Asymmetry Detection in Mammogram Images

Objectives: This work is used to efficiently analyze and identify the sign of the Bilateral Asymmetry presence in the mammogram images. Methods/Statistical Analysis: The proposed work performs the feature selection process prior to the classification of mammograms as Bilateral Asymmetry and architectural distortion. For the Bilateral Asymmetry the features used are based on the directional, morphological and density. For selecting potential features Artificial Bee Colony Optimization (ABCO) technique is used. The performance of each feature set obtained is measured using Artificial Neural Network classifier. The dataset for analysis is collected from MIAS database. Findings: The results of the ANN classification obtained by using the feature selection Particle Swarm Optimization and Ant Colony Optimization with the three features M1, M2 and H aligned are selected but its performance is relatively low while comparing with the selected features of the ABCO. The ABCO significantly reduces the false positive rates in the detection of the ROI in the Bilateral Asymmetry detection. In recognizing the sign of bilateral asymmetry, the results retrieved shows best performance which indicates features of directional features and ROI alignment. The experimental result shows that in the Bilateral Asymmetry the sensitivity and specificity using PCO is 0.79% and 0.83% and for ABCO it is 0.89% and 0.91% respectively. So the rate of true positive value increases in a substantial manner thus reducing the false positive rate. Application/Improvements: In this study, prelude results are exposed related to the segmentation of fibro-glandular discs, the two angular distributions categorizations and the features extraction. It reports the detection of Bilateral Asymmetry in Mammogram images that may be used by radiologists for earlier prediction of breast cancer. Future work will address the problem of the asymmetry assessment.