Tumor detection in digital mammograms

A novel approach for detecting microcalcifications in digital mammograms is proposed. The noisy and impulsive nature of normal breast tissue in mammograms leads to high false-positive rates, thereby impairing detection performance. We address this issue by first observing that the distribution of pixel values is heavy-tailed. A statistical-physical noise model, based on the generalized central limit theorem and simple models for X-ray attenuation, distribution of breast tissue, and the digitization process, is presented to explain the presence of outliers. We use this model to derive an optimal statistical test to detect breast abnormalities in symmetric alpha-stable (SαS) noise. The resulting algorithm yields a constant false-alarm rate (CFAR) SαS microcalcification detector that is robust in impulsive noise environments. Experimental results on mammogram images from the Digital Database for Screening Mammography (DDSM) are provided to demonstrate the usefulness of the proposed approach.