Brain tumor extraction using graph based classification of MRI time series for diagnostic assistance

Abstract

Automatic extraction of brain tumor regions from temporal sequence of MRI images has a great contribution for diagnostic assistance since it helps the expert to reduce the area of the region to analyze. However, it presents a challenging task for medical applications. Compared to single images, MRI images time series carry more rich information. Therefore, extracting the appropriate information among these data is more difficult. Moreover, spatio-temporal relationship modeling using graphs offers a simple way for powerful analysis. In this paper, a graph-based classification of MRI image time series is applied. It uses the advantages of graph representation of such data in order to extract the relevant information. This framework proved its efficiency in the context of Satellite Images Time Series (SITS) classification in order to monitor land cover evolution. Therefore, we intend to take advantage of this SVM graph based classification of image time series in medical imaging context. Using graph representation, an adapted labeling and temporal neighborhood definition are used for MRI time series' regions modeling. Then, SVM classification using graph kernel is applied to extract brain tumor regions. The experimental results have been conducted on real MRI data proving the accuracy of the proposed approach.