Stationary Wavelet Based Energy Feature Extraction for Detection and Classification of Mammographic Microcalcifications

Abstract

Since radiologists widely use mammographic microcalcifications as initial tool for breast cancer screening, accurate detection of microcalcifications is an inevitable stage to develop an effective diagnosis system. This paper proposes a Stationary Wavelet Transform (SWT) based novel technique for detection and classification of breast microcalcifications. To detect the suspected microcalcifications from mammograms, SWT was applied at multiple levels for decomposition purpose and Stationary Wavelet Energy (SWE) was then implemented for feature extraction from each obtained detailed SWT coefficient sub-bands. Four different Ensemble classifiers were employed for classification of microcalcifications as benign or malignant using these SWE features, conducting 10 fold cross validation. Mammographic Image Analysis Society (MIAS) mammographic database was used for experimental evaluation and at maximum a sensitivity of 94.12%, a specificity of 92.48%, a precision of 88.89% and an accuracy of 92.11% were obtained using Subspace Discriminant Ensemble classifier. Beside outcomes of comparative analysis prove the supremacy of our proposed approach over two state-of-the-art approaches.