Publication

Abstract : Breast cancer is the most fatal among the cancers detected in women. Mammography is the most common and efficient tool for early detection of breast cancer. In mammography images of breast are captured in two standard views namely Mediolateral oblique (MLO) view and Craniocaudal (CC) view. Radiologists generally analyze both the views during diagnosis. In recent years as the number of cases to be diagnosed is increasing significantly, computer aided diagnosis (CAD) systems were developed. In the datasets used by these systems, often there is a chance for the presence of missing values (MVs) in any one of mammographic views due to various reasons. Some of these include obscuration of mass by a dense breast tissue, region of interest being out of frame during the image acquisition etc. This results in the use of data from single view alone for diagnosis. But the diagnostic performance of CAD systems is better when multi-view data is used. In this paper the use of Iterative Singular Value Decomposition (ISVD) imputation is proposed to handle missing values in order to preserve the advantage of using multi-view data during the diagnosis. Classification accuracy and Kappa statistics are the metrics used to assess the performance of ISVD scheme for different percentages of MVs ranging from 1-15%. Experimental results demonstrated that diagnosis using multi-view following ISVD performed at least as well as and most of the time better than the systems using single view.