Multi-kernel SVM based classification for tumor segmentation by fusion of MRI images

Tumor segmentation, a significant application in the field of medical imaging and pattern recognition, is still a very difficult and unsolved problem up to now. In this paper, an improved SVM algorithm-multi-kernel SVM, integrated with data fusion process, is proposed to segment the tumors from the MRI image sequence. Three kinds of MRI image sequence-T2, PD, FLAIR are used as input sources in learning and classifying process. Then a region growing step is exploited for a refinement of the tumor contour. At last, according to the follow-up result of the same patient at five different periods, it is obvious that the tumor´s volume becomes smaller, and an evaluation percentage is given to prove the effectiveness of the therapy. The quantification of result demonstrates the effectiveness of the proposed method.