Zemike moment feature extraction for classifying lesion's shape of breast ultrasound images

One of the methods that is often used to screening breast cancer is ultrasound examination but the result is very subjective. It depends on the ability the radiologist. Therefore, a tool that can help us to make it more objective needs to build. System on the device should be able to diagnose breast cancer malignancies of various parameters including the breast lesion shape parameter. Image classification of breast lesion shape begins with input image processing by filtering it with a median filter then performing segmentation with Chan-Vese active contour, conducting extraction of feature with Zernike moments and Invariant moment. Finally, undertaking classification by support vector machine (SVM) and Multilevel Perceptron (MLP). At the Zernike moment, the highest accuracy obtained by using SVM classifier is 84.80%, and the MLP classifier is 87.90% At invariant moment, accuracy obtained by using SVM classifier is 69.69 %, and the MLP classifier is 78.70%. On the other side, when Zernike moment and invarian moment are combined, the classification result achieves 93,90% for accuracy, 91.70% for specification, and 100% for sensitivity.