A Comparative Study of Breast Mass Classification based on Spherical Wavelet Transform using ANN and KNN Classifiers
Pelin Görgel, Ahmet Sertbaş, Osman Nuri UCAN · DergiPark (Istanbul University) · 2012
Breast cancer may be missed by the radiologists at the early ages because of the mammography artifacts. Computer aided diagnosis can decrease the mortality rate by providing a second eye. The artifacts exist due to the noise and the inappropriate contrast in mammograms. Therefore a study that classifies the cropped region of interests (ROI’s) as benign or malign and provides a second eye to the radiologists is proposed. The study consists of two steps: First step is the application of Spherical Wavelet Transform (SWT) to the original ROI matrix prior to feature extraction. Second step is to extract some predetermined pixel and shape features both from wavelet and scaling coefficients. Finally, for classification the prepared feature matrix is given to Artificial Neural Networks (ANN) and K-Nearest Neighbour (KNN) systems which are widely used in image processing. The algorithms are tested on 60 abnormal digitized mammogram ROIs acquised from The Mammographic Image Analysis Society (MIAS) which is a free mammogram database