Abstract
Architectural distortion is one of the most important findings in screening mammograms. Since this finding is related to the risk of malignancy and a poor prognosis, improvement of the detection rate is strongly desired. In this paper, we propose a novel method for detecting architectural distortion by analyzing mammary gland structure using multiple Gabor filters. In our method, multiple Gabor filters with different characteristics are applied, and the filter that matches the mammary gland structure is selected. Then, the degree of concentration is calculated using a filter output image, and a candidate region of spicula and traction of the mammary gland structure caused by the architectural distortion is obtained. In the experiments, we compared the results of the proposed method against interpretation by a physician using 22 images of 6 cases in a digital database of screening mammograms. The results showed a true positive rate of 80% with 1.06 false positives per image. We conclude that this initial candidate detection method for architectural distortion provides an acceptable level of performance.
