Dense rotation invariant brain pyramids for automated human brain parcellation

The automatic parcellation of the human brain based on MR imaging isin several areas of high interest. In particular, identifying corresponding brain areasbetween different subjects is an indispensable prerequisite for any group analysis. Butalso, simple segmentations into different tissue types is an important preprocessingstep. We present a generic framework for describing and automatically parcellatinghigh angular resolution diffusion-weighted magnetic-resonance images (HARDI) ofthe human brain. Based on an initial training step our approach is capable to seg-ment the images into coarse parcellations or detailed fine grain regions of interest. Incontrast to existing model-free methods [SSK09] we are not only using the raw mea-surements at each position, but we are also including neighboring measurements in arotation invariant way. References[SSK09] S. Schnell, D. Saur, B.W. Kreher, J. Hennig, H. Burkhardt, and V.G. Kiselev. Fully auto-mated classification of HARDI in vivo data using a support vector machine. Neuroimage,46:642–651, 2009. ground truth regionsused in experiment 2prediction in data 1ground truth regionsused in experiment 1prediction in data 1 Figure 1: The ground truth regions that we used to train and evaluate our algorithm shown togetherwith our algorithm’s regions prediction. We can clearly see that our predictions are much moreconsistent with the data.