In recent years a growing interest has grown in Magnetic Resonance images segmentation techniques, due to their usefulness in many applications. Within this manuscript, a novel segmentation approach is presented, based on two main innovations. First, it exploits the estimated proton density and relaxation times for each pixel, instead of its gray-level intensity. This feature makes the algorithm particularly robust and allows the classification of identified segments. Secondly, it implements a specifically evolved version of the DBSCAN approach, gaining advantages in the effectiveness of region estimation. The technique, compared to an euclidean distance based one, is able to improve the correct classification rate. The effectiveness of the approach is evaluated on a simulated case study, and will be extended to real data within next weeks.