Development of new schemes for detection and analysis of mammographic masses

We are developing automated-detection and analysis schemes of mammographic masses. The purpose of this study is to improve our previous schemes on the mass detection and analysis. In our detection scheme, pectoralis muscles area is firstly extracted. Digital mammograms are classified into four category and breast regions are segmented into dense and fatty parts. Low density areas as mass candidates are detected by several different thresholds. Feature analysis by size, circularity, standard deviation and contrast is finally employed for the detected areas to eliminate false positives. The residual candidates are detected as "true" masses and are classified into benign and malignant. In our analysis scheme, this classification is determined by the change of fractal dimension. The spicules on the detected masses are found by a proposed "pendulum filter". As results, the computerized method correctly localized 97% of the true masses with 3.5 false-positive detection per image. Performance of classification into benign and malignant masses by the fractal dimension was 100% in thirteen mammograms. The sensitivity and specificity of the pendulum filter were 93% and 73% in thirty mammograms, respectively. It was concluded that our methods have the potential to aid radiologist.