Algoritmo supervisado de segmentación automática para imágenes de resonancia magnética

Abstract

An exponential improvement in computation capacity throughout the last decades has allowed the use of further computationally demanding algorithms to solve many kinds of problems in real time. In this project both computer vision and machine learning techniques are applied to support a neuroimage study. In particular, a fully automatic method to segment the caudate nucleus of the brain from a magnetic resonance image (MRI). Studies have shown that neuroanatomical abnormalities in the caudate nucleus are strongly related to pediatric attentiondeficit/hiperactivity disorders (ADHD). Therefore, providing an automatic subjectiveless tool to segment its volume not only improves the diagnose, but it also speeds up the process, freeing the experts from the arduous segmenting task. In order to achieve this purpose Graphcut unsupervised was developed. In this project a supervised term is added to the currently existing envoirment. Particularly a support vector machine classifier is trained with a set of MRI slices, which is used to refine the segmentation algorithm results.