Multi-resolution wavelet-transformed image analysis of histological sections of breast carcinomas

Multi-resolution images of histological sections of breast cancer tissue were analyzed using texture features of Haar- and Daubechies transform wavelets. Tissue samples analyzed were from ductal regions of the breast and included benign ductal hyperplasia, ductal carcinoma in situ (DCIS), and invasive ductal carcinoma (CA). To assess the correlation between computerized image analysis and visual analysis by a pathologist, we created a two-step classification system based on feature extraction and classification. In the feature extraction step, we extracted texture features from wavelet-transformed images at 10x magnification. In the classification step, we applied two types of classifiers to the extracted features, namely a statistics-based multivariate (discriminant) analysis and a neural network. Using features from second-level Haar transform wavelet images in combination with discriminant analysis, we obtained classification accuracies of 96.67 and 87.78% for the training and testing set (90 images each), respectively. We conclude that the best classifier of carcinomas in histological sections of breast tissue are the texture features from the second-level Haar transform wavelet images used in a discriminant function.