Breast tumor classification in ultrasound images using texture analysis and super-resolution methods

Ultrasound images can be used to detect tumors that do not appear in the mammograms of dense breasts. Several computer-aided diagnosis (CAD) systems based on this type of images have been proposed to detect tumors and discriminate between benign and malignant ones. To characterize those lesions, many of the aforementioned systems rely on texture analysis methods. However, speckle noise and artifacts that appear in ultrasound images may degrade their performance. To tackle this problem, and contrary to the state-of-the-art methods that utilize a single image of the breast, this paper proposes the use of a super-resolution approach that exploits the complementary information provided by multiple images of the same target. The proposed CAD system consists of four stages: super-resolution computation, extraction of the region of interest, feature extraction and classification. We have evaluated the performance of five texture methods with the proposed CAD system: gray level co-occurrence matrix features, local binary patterns, phase congruency-based local binary pattern, histogram of oriented gradients and pattern lacunarity spectrum. We show that our super-resolution based approach improves the performance of the evaluated texture methods and thus outperforms the state of the art in benign/malignant tumor classification.