Tumor Segmentation From a Multispectral Mri Images by Using Support Vector Machine Classification

The goal of this paper is to present a supervised system aimed at tracking the tumor volume during a therapeutic treatment from multispectral MRI volumes. Four types of MRI are used in our study: T1, T2, proton density (PD) and fluid attenuated inversion recovery (FLAIR). For decreasing the processing time, the proposed method employs a multi-scale scheme to identify firstly the abnormal field and extract then the tumor region. Both steps use support vector machines (SVMs). The training is carried out only on the first MRI examination (at the beginning of the treatment). The tracking process at the time point t takes the tumor region obtained in the examination at t-1 as its initialization. Only the second step is performed for others examinations to extract the tumor region. The results obtained show that the proposed system achieves promising results in terms of effectiveness and time consuming.