Iman Aganj1, Christophe Lenglet1,2, Renaud Keriven3, Guillermo Sapiro1, Noam Harel2, Paul Thompson4
1Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, USA; 2Center for Magnetic Resonance Research, University of Minnesota Medical School, Minneapolis, MN, USA; 3CERTIS, cole Nationale des Ponts et Chausses, Champs-sur-Marne, Marne-la-Valle, France; 4Laboratory of Neuro Imaging, University of California-Los Angeles, Los Angeles, CA, USA
Tractography algorithms based on local fiber orientation estimates are vulnerable to noise, partial volume effects, and above all the fiber crossing, since recovering connectivity in regions where fiber bundles mingle is particularly difficult. In this work, we present a global approach based on Hough transform. Our tractography algorithm essentially tests all possible 3D curves in the volume while giving a score to each of them, then chooses the curves with the highest scores and returns them as the potential connections. We present experimental results on both artificial and real DTI and HARDI data.