Breast Tumor Analysis in Dynamic Contrast Enhanced MRI Using Texture Features and Wavelet Transform

Abstract

Dynamic contrast enhanced MRI (DCE-MRI) is an emerging imaging protocol in locating, identifying and characterizing breast cancer. However, due to image artifacts in MR, pixel intensity alone cannot accurately characterize the tissue properties. We propose a robust method based on the temporal sequence of textural change and wavelet transform for pixel-by-pixel classification. We first segment the breast region using an active contour model. We then compute textural change on pixel blocks. We apply a three-scale discrete wavelet transform on the texture temporal sequence to further extract frequency features. We employ a progressive feature selection scheme and a committee of support vector machines for the classification. We trained the system on ten cases and tested it on eight independent test cases. Receiver-operating characteristics (ROC) analysis shows that the texture temporal sequence (Az: 0.966 and 0.949 in training and test) is much more effective than the intensity sequence (Az: 0.871 and 0.868 in training and test). The wavelet transform further improves the classification performance (Az: 0.989 and 0.984 in training and test).