Automated Classification of Fatty and Dense Mammograms

Increased breast density was found to be associated with an increased risk of breast cancer growth, as high density makes it difficult for radiologists to see an abnormality which leads to decrease in sensitivity. Therefore, there is need for the development of highly efficient techniques for breast tissue classification based on density. This paper presents a scheme for the classification of fatty and dense mammograms based on texture analysis of Region of Interest (ROI) selected from the full mammogram. The work is based on 212 mammograms selected from the mini-MIAS database. Texture analysis is done by using Haralick's Spatial Gray Level Co-occurence Matrix (SGLCM), Statistical Feature Matrix (SFM), Gray Level Difference Statistics (GLDS), First-order Statistics (FoS), Law's Texture Energy Measures (TEM), Fractal and Fourier Power Spectrum (FPS) features. A total of forty five texture features are computed for fatty as well as dense ROIs. Since high dimensionality of feature vector may limit computational efficiency and accuracy of classification, therefore optimal feature selection is done to eliminate the less informative and redundant features. In this paper, feature selection is done by scalar feature ranking technique that employs Fisher's Discriminant Ratio (FDR) as well as cross correlation measure between pair of features in order to rank them. The highest ranked features are identified and exhaustive search is employed to select the best combination of features. It has been observed that feature selection results in three optimal features namely: Sum Average, Radial Sum and Angular Sum. The accuracy of these features in distinguishing fatty and dense mammograms has been evaluated by k-Nearest Neighbor (k-NN) classifier. From the analysis of results it was found that classifier gave an overall classification accuracy of 97.22% with 100% sensitivity. Hence these selected features can be used for building an efficient Computer Aided Diagnosis (CAD) system for breast density classification.