False positive and false negative reduction in digital mammograms using binary rotation invariant and noise tolerant texture descriptor

Digital mammography is one of the most effective approaches to diagnose breast cancer at early stage. Several strategies have been proposed to identify breast tumor, yet these techniques endure with high rate of false positive and false negative. It is challenging task to reduce false positive and false negative. In this research work, a new approach is proposed that uses textural properties of the masses to reduce false positive and negative in mammograms. The thought of utilizing textural data for tackling this issue is not new and has been earlier presented in several works. However, the proposed method utilizes a recently developed texture descriptor called Binary Rotation Invariant and Noise Tolerant (BRINT) to characterize local-patterns (i.e. lines, edges, flat areas, spots). In the meantime, spatial structure of the masses is preserved. The proposed system is tested on 512 Regions of Interest (ROIs) extracted from the Digital Database for Screening Mammography (DDSM) database. Support Vector Machines (SVM) is used with radial basis kernel for classification. The comparison is performed with Local Binary Pattern (LBP) based feature extraction strategy. The results indicate that the proposed method has improved the classification accuracy.