Breast cancer discriminant feature analysis for diagnosis via jointly sparse learning

Breast cancer is one of the most common malignant tumors among female. How to effectively discriminate the category of the cancers using the key features/factors is very important in the diagnosis and prediction. In this paper, Jointly Sparse Discriminant Analysis (JSDA) is proposed to explore the key factors in breast cancer and extract the key features for improving the accuracy in diagnosis and prediction. JSDA introduces the jointly sparse regular term (i.e. L2,1L2,1-norms term) to the criterion. A convergent iterative algorithm is designed to solve the optimization problem. It is shown that the proposed JSDA algorithm not only can learn the jointly sparse discriminant vectors to explore the key factors of the breast cancer in cancer pathologic diagnosis, but also can improve the diagnosis accuracy compared with the classical feature extraction and discriminant analysis algorithm. Experimental results on breast cancer datasets indicate that JSDA outperforms some well-known subspace learning algorithms in prediction accuracy, not matter they are non-sparse or sparse, particularly in the cases of small sample sizes. Data analysis shows that the key factors of the breast cancer explored by the JSDA are consistent with the practical experience.