Detection of abnormalities in digital mammograms based on Support Vector Machines / 医疗卫生装备

ABSTRACT

Objective To search an approach based on Support Vector Machine (SVM) for detection of different abnormalities including micro-calcifications and masses from digital mammograms. Methods Such detections were formulated as supervised-learning problems and SVM was applied to the detection algorithm. After the regions of interest were pre-processed by specific rectangular windows, three kinds of parameters were extracted, including the direct pixel value parameter, the parameters from Spatial Grey Level Dependency (SGLD) matrices and from Discrete Cosine Transform (DCT). At first, each kind of parameter was taken as the input of SVM to train and test the machine respectively. Then all the parameters were incorporated into the input of SVM. Results the classification accuracy is 92.28%, 90.35% and 91.12% respectively when only one parameter input. The classification accuracy reaches 99.23% when all the parameter incorporated. Conclusion The parameters extracted from the regions of interest in digital mammograms can reflect the characteristics of different regions and SVM is a powerful tool for the detection of abnormalities from digital mammograms.