Breast Cancer Risk Assessment using Gabor Filter Banks and Curvelet Transform

The goal of breast cancer risk assessment is to establish the risk that a healthy person has of developing breast cancer in the future. Recently, mammographic images have shown great potential for the extraction of imaging biomarkers for the construction of breast cancer risk models. This work aims to compare Gabor filters and the Curvelet transform for the extraction of texture features from mammographic images, evaluating their performance for the computerized assessment of breast cancer risk. Gabor filters and the Curvelet transform decompose texture information at different scales and orientations in the spatial frequency domain and they can be utilized for the analysis of mammographic texture features. We compared different configurations of both Gabor and Curvelet filter banks by analyzing the effect of the number of bands and architecture of the filters. Results in a pilot case-control study with digital mammography images from 54 women (27 cases and 27 controls) indicate that Gabor filter banks yield a superior performance than Curvelet analysis, with an AUC of 0.722 in the classification between high-risk and low-risk women.