Currently, doctors can choose from a variety of MR imaging modalities to contrast different tissue structures (for example gray or white matter), or to highlight different neoplastic deformities. There are several techniques [1,6,7,8] for consolidating the intensity information from several modalities into one single image with improved highlighting of these features. However, these techniques are generally sensitive to input noise and initialization data [6,7], or require prior knowledge which is not practical in the medical environment[1]. Our research shows that with minor modifications, spatially coherent clustering [2,3] is an effective tool for producing segmentations out of multi modal data. The resulting images are smoother and show greater tissue detail. Our research also overcomes the problem of initialization data by periodically adjusting the number of clusters.

A copy of this paper can be found here