Development of an automated detection system for microcalcifications on mammograms by using the higher-order autocorrelation features

The purpose of this work is to develop a new pattern recognition method using the higher-order autocorrelation features (HOAFs), and to apply this to our microcalcification detection system on mammographic images. Microcalcification is a typical sign of breast cancer and tends to show up as very subtle shadows. We developed a triplering filter for detecting microcalcifications, and the prototype detection system is nearly complete. However, our prototype system does not allow for the detection of three types of microcalcifications, two of which are amorphous and linear microcalcifications and the third is obscured microcalcifications which is often confused with the background or circumference that have almost the same density. We targeted the amorphous type of microcalcification, which has a low contrast and easily goes undetected. The various features of microcalcifications and false-positive (FP) shadows were extracted and trained using the multi-regression analysis, and unknown images were recognized as a result of this training. As a result, amorphous microcalcifications were successfully detected with no increase in the number of FPs compared with our existing detection method.