Integration of Global and Local Descriptors for Mass Characterization in Mammograms
Devi Vijayan, R. Vidhya Lavanya · Procedia Computer Science · 2023
Breast cancer is the most common cancer among women and is associated with high morbidity and mortality. Early detection has proven successful in effective treatment and increased survival rate. Mammography is the most common modality for screening breast cancer; nevertheless, it is associated with low sensitivity and specificity due to subtlety of abnormalities in early stages. A mass which is the most common and important indicator of breast cancer, is especially challenging to analyse and interpret. Computer aided diagnosis (CAD) systems can improve the diagnostic performance by aiding radiologists in providing an objective assessment. The proposed work aims to develop a CAD system to evaluate mammographic masses and characterize them into benign and malignant categories. The system employs an integrated approach based on fusion of global descriptors extracted from the segmented mass regions and local descriptors from a ribbon of pixels surrounding the mass boundary, to effectively characterize masses. The proposed approach is validated using the Digital Database for Screening Mammograms (DDSM) database, yielding a classification accuracy as high as 94.6% for mass characterization.