A Textural Wavelet Quantization approach for an efficient breast microcalcifcation’s detection

Abstract

Microcalcifications are very small deposit of calcium. Their detection is a crucial task. Their presence affects the texture of a breast tissue. Texture information has the ability of mapping microcalcification’s characteristics. Thus, texture based features allow to carry in a more accurate analysis for microcalcification detection. In this paper, a texture based microcalcification detection method based on Textural Wavelet Quantization (TWQ) is proposed. It is based on a quantization of textural information on a wavelet transform domain. Firstly, to further highlight microcalcification details, we apply a nonlinear enhancement technique. The resulting enhanced image will be subsequently used to extract textural features in order to detect the presence of microcalcifications by means of local information. A product between wavelet coefficients of the enhanced image and those of the Gaussian Derivative filter is done highlight microcalcification’s frequency and remove the other frequencies. The resulting coefficients are subsequently quantized in a feature vector. This feature vector is considered subsequently as input vector for the classification step of the corresponding breast tissue. Indeed, our proposed texture descriptor allows to distinguish breast tissue with microcalcifications from safe one. A comparative study illustrates the efficiency of such approach, among existing ones, in classifying breast tissue. The proposed approach yields an area Under the Receiver Operating Characteristic (ROC) curve (AUC) of 99.93%.