Towards Computer-Aided Diagnostics of Screening Mammography Using Content-Based Image Retrieval

Screening mammography has been established worldwide for early detection of breast cancer, one of the main causes of death among women in occidental countries. In this paper, we aim at moving towards computer-aided diagnostics of screening mammography. Tissue and lesion are classified using the methodology of content-based image retrieval. In addition, we aim at comprehensive evaluation and have established a large database of annotated reference images (ground truth), which has been merged and unified from different sources publicly available to research. In total, 10,509 mammographic images have been collected from the different sources. From this, 3,375 images are provided with one and 430 radiographs with more than one chain code annotations. This data supports experiments with up to 12 classes, and 233 images per class if a equal distribution is required. Using a two-dimensional principal component analysis with four eigenvalues and a support vector machine with Gaussian kernel for feature extraction and image retrieval, respectively, the precision of computer-aided diagnosis is above 80%. It therefore may be used as second opinion in screening mammography.