EMD Based Binary Classification of Mammograms with Novel Leader Selection Technique

Mammography is one of the primary radiography techniques that is used for detection of breast lesions which may range from benign to malignant pathologies. However manual analysis of a mammogram can be both time intensive and prone to unwanted error. Engineers off late has been actively contributing to the domain of medical image classification to ease the categorization process and make it efficient. This paper introduces a technique to identify the graveness of breast carcinoma from mammograms using three different sized training sets. In the current study we present an Earth Mover’s Distance (EMD) based binary classification of mammograms containing benign and malignant tumors. To facilitate the classification, a novel Leader Selection (LS) technique is used to identify the leader of each cohort in the training sets. The proposed model achieves a maximum sensitivity of 93.75% while producing a maximum F1 score of 83.33%. It is also observed that increasing the size of the training set improves the relevant performance metric.