Brain Tumor Image Segmentation Using Neural Networks
Villalta R, Martn-Landrove M
Centro de Resonancia Magntica, Facultad de Ciencias, Universidad Central de Venezuela
A new method that combines MRS with multiecho T2-weighted images is proposed to obtain nosologic maps with appropriate spatial resolution for treatment considerations. Neural networks are used to classify relaxation data according to spectroscopic information and combine it in the generation of nosologic maps. Results are compared with other segmentation methods based in the analysis of relaxation rate distributions by inverse Laplace transform algorithms showing a high degree of correspondence and an improvement in processing time, what makes this methodology suitable for the analysis of a high number of images, typically acquired fro 3D treatment planning in radiotherapy or radiosurgery.