Mammographic Classification Using Stacked Ensemble Learning with Bagging and Boosting Techniques

Abstract

The classification of mammographic images is an important process in Computer-aided diagnosis (CADx) system for an automatic detection of breast cancer. CADx helps radiologists in providing a second opinion for an accurate diagnosis. Therefore, an intelligent classifier is required in classifying suspicious areas in digital mammograms. A number of breast cancer classification techniques have been proposed over years, however, these techniques suffered from limitations that impact its performance. This paper aims to investigate an approach using stacked ensemble method that can reduce total classification error by reducing the variance and integrating the outputs of several classifiers to obtain a better classification performance. The proposed stacked ensemble classifier uses three algorithms Classical Soft Independent Modelling by Class Analogy (CSimca), Generalized Linear Model (GLM), Gaussian Process with Radial Basis Function Kernel (gaussprRadial) at the bottom layer and Stochastic Gradient Boosting (GBM) at the top layer. It is observed that the ensemble approach performs better than all the other single classifiers across the experiments that has achieved an overall classification accuracy of 85%, with sensitivity 86.67%, precision 83.87% and specificity 83.33% in the testing stage. The stacked ensemble learning classification techniques can be widely used in CADx that have proved their efficiency in mammogram classification.