Classification of breast tumors in ultrasound using biclustering mining and neural network

Ultrasound imaging is now becoming a frequently used tool in clinical diagnosis. In this article, a novel diagnosis scheme is proposed to aid the identification of breast lesions. In this approach, feature scoring scheme is first applied to product feature data. Biclustering mining is then helpful to discover the effective diagnostic patterns, and those found patterns are utilized to transform the original features into advanced hidden features. Finally those advanced features are treated as the input data for back-propagation (BP) neural network algorithm to train an efficient classifier for recognizing benign and malignant breast tumors. The proposed approach has been validated using a database of 238 breast tumor instances (including 115 benign cases and 123 malignant cases) with its performance compared with other conventional approaches. Experimental results indicate our proposed method yielded good performance in tumor classification, with the accuracy, sensitivity, specificity of 96.1%, 96.7%, 95.7%, respectively.