Segmentación adaptativa de lesiones isquémicas cerebrales a partir de imágenes de difusión de resonancia magnética

Abstract

The magnetic resonance imagenology (MRI) has become in one of the most important medical image modalities for diagnosis, prevention and monitoring several medical disorders. In particular, the diffusion weighted imaging (DWI) is extremely sensitive to achieve an early detection of ischemic changes in the acute phase of a brain infarct. In this study, it is presented the application of an adaptive segmentation method which has been validated and developed previously. The method uses a no parametric estimation based on the bandwidths or variable intensity radius. The main objective of the proposal method is to quantify the brain region which has been affected by an infarct but using the information contained in the DWI images. The segmentation algorithm with constant parameters was applied in the whole set of real images belonging to the previously acquired database. A comparison between the adaptive technique of DWI images segmentation and no parametric method with fixed radius was developed. This comparative study shows the benefits achieved by the adaptive method: the automatically processing and the robustness under different brain ischemical regions in acute phase. Even the sensitiveness is improved because the adaptive method was able to obtain the segmentation of images with small affected volumes (< 1 cm3). Comparing with the reference control segmentation method, the considered methods evaluated in this study improved the joint correlation: r=0.8863 for the fixed radious and r=0.9693 when the radious is variable. The adaptive method showed the best results among the other alternatives. Indeed, the averaged tanimoto index obtained in the adaptive version of the segmentation algorithm was superior to the one achieved when the radius was fixed (0.729 and 0.638 respectively)