Can Bilateral Asymmetry Analysis of Breast MR Images Provide Additional Information for Detection of Breast Diseases?

Abstract

This paper presents a new method for bilateral asymmetry analysis of breast MR images that uses directional statistics of the breast parenchymal edges, obtained from a multiresolution local energy edge detector, and image texture information derived from local energy maps, obtained by using a bank of log-Gabor filters. Classification of MRI scans into cancer and non-cancer categories was performed by linear discriminant analysis and the leave-one-out methodology. A total of 40 cases, 20 normal/benign (BI-RADS 1 and 2) and 20 malignant, taken from a high risk screening population,were used in this pilot study. Average classification accuracy of 70%(k=0.45 +- 0.14) with sensitivity and specificity of 75%and 65%, respectively, was achieved. The results obtained support the idea that bilateral asymmetry analysis of breast MR images can provide additional information for detection of breast tissue changes arising from diseases.