DoG-Based Detection of Architectural Distortion in Mammographic Images for Computer-Aided Detection

We propose a new method for accurate detection of architectural distortion that is a typical sign of breast cancer lesions in mammograms and necessary to be detected and diagnosed properly at an early stage for improvement of the survival rate of patients. An essential core of the proposed method is to efficiently extract a new general feature of the architectural distortions whose lesional intensities are not only higher than those of the surroundings as well known, but also often lower. While conventional features such as radial lines and higher intensities are difficult to be extracted and/or insufficient for accurate detection, the candidate area with such a new feature can be extracted accurately by using a difference of Gaussian (DoG)-based filter and after that a thresholding technique can reduce the number of false positives. The detection based on the new feature is expected to be more accurate than conventional ones because it reflects more general characteristics of the lesion. The experimental result using the database commonly tested worldwide shows that performance of the proposed method is superior to those of conventional ones.