A new hybrid approach for medical image intelligent classifying using improved wavelet neural network

Abstract

Breast cancer is the second leading reason of fatality among all cancers for women. In this paper, we propose a novel method for early breast cancer intelligent classification. We combine wavelet theory with neural network theory to construct an improved wavelet neural network (IWNN) for digital mammography classification. Firstly, we combine redundant dyadic wavelet transform with ridgelet transform to enhance the image. Because most of the wavelet coefficients containing signals are retained, the image detail can be kept better. And then, the statistical coefficients of the source regions are extracted as features for classification. At last, the medical images are classified by using IWNN on real datasets MIAS(the Mammographic Image Analysis Society). The experimental results show that proposed IWNN classifier can achieve 86.71% accuracy, which outperform the traditional neural network method with 5.46% of increase of classification accuracy. The correct recognition rates are close to 100% averagely. © Springer-Verlag 2013.