Identification of Breast Tumours by Wavelet Based Adaptive Windowing Technique with Statistical Features Using Back Propagation Neural Network

Abstract

Mammography is the most efficient procedure for the early detection of breast diseases. Mammogram analysis refers the processing of an image with the goal of finding abnormalities. This paper proposes a new technique to identify whether the tumour is present in mammogram or not. This technique includes image segmentation, feature extraction and classification. A wavelet based adaptive windowing technique is used for segmentation. Coarse segmentation is done by wavelet based histogram thresholding and fine segmentation result obtained by window based adaptive thresholding. A Multi layer Feed Forward neural network is used for identification. The purpose of this paper is to classify the breast tissues into normal and abnormal classes automatically. This method saves the radiologist time, increases accuracy and yield of diagnosis. The algorithm is validated with mammograms in Mammographic Image Analysis Society Mini Mammographic database which shows that the proposed technique is capable of detecting tumours of very different sizes.