SVM Based Follow-up System for Brain Tumor Evolution from Magnetic Resonance Images

Abstract

Listen

Translate article

In this paper, a semi-automatic therapeutic evaluation system of brain tumor is proposed, which consists in following up the same patient during a therapeutic treatment to evaluate the state of the tumor and to quantify the effectiveness of the therapeutic treatment. Three types of MRI (Magnetic Resonance Imaging) sequences, T2-weighted, PD (Proton Density) and FLAIR (Fluid Attenuated Inversion Recovery), are used to obtain maximum information about the tumor. This system is based on SVM (Support Vector Machine) classification to segment the brain tumor with a novel kernel-based feature selection process which allows to efficiently fusing the three source images, to increase class separability and to reduce data dimensionality. The procedure consists of three steps for each MRI examination of the patient: the first step is to automatically carry out the SVM training except the first examination which is manual, the second one is to classify the tumor tissues, the last step is to improve the classification results by a region growing technique using a criterion combined with the probability belonging to the tumor tissue and the distance from the contour of the tumor. The procedure repeats for new examinations and gives the quantitative variation of tumor volumes between two examinations. Our system is tested on real patient images. The comparison with the manual results of experts demonstrates the good performance of the proposed method.