Segmentation of MR Image Based on Support Vector Machine and Conditional Random Field

A novel segmentation algorithm which was called SVM-CRF for encephalic tissues in MR images was proposed. It is known that support vector machine (SVM) is a powerful machine learning algorithm for solving high dimension and non-linearity problems, and conditional random field (CRF) is effective in learning the dependency among the local data. The SVM and the CRF were combined to obtain the SVM-CRF segmentation algorithm, aiming to take their own advantages to improve the image segmentation accuracy. Experimental results show that the segmentation accuracy of the SVM-CRF is better than that of the SVM and the CRF, which is increased by 1.83% and 5.81% respectively for the cerebrospinal fluid, and increased by 1.84% and 7.60% respectively for the cancellous substance. Theoretical analysis and experimental results indicate that the SVM-CRF has better accuracy than the SVM and the CRF respectively for tissue image segmentation, especially for the tissue image which is difficult to be identified.