Abstract
We have been developing a computer-aided diagnosis (CAD) system for mammograms. The system detects clustered area of mammographic microcalcifications based on density gradient analysis, and classifies the candidates into benign or malignant according to the feature analysis. Although the image digitized with 100-μm sampling distance (“100-μm image”) was used in our previous study, the image digitized with 50-μm sampling distance(“50-μm image”) would be used to improve the classification performance because the precise shape of microcalcification may be extracted from these images. In this study, we propose a technique of classifying clustered microcalcifications using 50-μm image. The re-detection processing by 50-μm image including noise reduction processing based on smoothing technique is newly added to our conventional method. As a result of applying this technique to 85 cases (137 ROIs),10% of improvement was achieved in terms of a correct answer rate, and the validity of this technique was investigated by ROC analysis. It was concluded that the classification of clustered microcalcifications with higher accuracy was obtained from these procedures of using a highresolution image(50-μm image)effetively.
