Automatic Grading For B-Scan Placenta Images Based On Support Vector Machine

Grading of ultrasound placenta images is important reference onexpressing the maturity of placenta. It is one of important fields onstudying the placenta in medicine. This article firstly uses the SVM torealize automatical grading for the B ultra images of placenta This paperalso proposes a kind of automatical grading method: Firstly, featureselection of the B ultra images of placenta is implemented based onfeature extraction; Then the new feature parameters are used as inputvectors to grade the B ultra images of placenta in multi-class supportvector machines.This article firstly analyzes the characteristic of placenta B ultraimages. Then three kinds of feature extraction methods (the statisticfeatures based on gray information, feature extraction based on GLCMand feature extraction based on Gabor wavelet) are proposed to expressfeatures of the placenta B ultra images from spatial domain, frequencydomain and texture respectively. Then the method of principal componentanalysis (PCA) is used to analyze these feature parameters in order toobtain new feature parameters with lower dimension and better validity. The KPCA based on PCA is firstly applied in this paper. The non-linearityrelationship of original feature parameters is explained and the effectionof kernel parameters is analyzed here. Finally, the multi-class SVM basedon binary tree is used to identify and classify feature parameters. Thefunction of automatic grading for the B ultra images of placenta isrealized. In the meantime, the merit of binary tree is analyzed and a kindof algorithm of parameters (C,σ) which decides performance of SVMis proposed too.The research work in this paper shows that using multi-class SVMbased on binary tree is of very good effection in Automatic Grading forB-Scan Placenta Images.