A hierarchical tissue segmentation approach in brain MRI images

Magnetic resonance imaging (MRI) is a widely used approach to obtaining high quality medical images of the brain. Post-processing MRI images with segmentation algorithms chances the visualization and measurement of soft tissues and lesions. Segmented brain images contain information amenable to quantitative analysis (e.g., tissue component percentage in a region of interest (RO])) and diagnostic interpretation (e.g., total lesion volume). A number of different segmentation algorithms have been developed for this purpose. In this paper, we propose a novel automated segmentation technique, hierarchical structure weighted probabilistic neural network (HSWPNN), based on multi-scale feature extraction, hierarchical labeling structure, and a modified weighted probabilistic neural network (PNN). Compared to other clustering algorithms, our method is relatively robust to noise and accurate. We compare our results to a model of ground truth.