Modified Normal Vector Voting Estimation in neuroimage

The normal vector is one of the metrics that are sensitive to the property of the curved surface. In neuroimage studies, normal vectors are used to strip skulls, to analyze cortical thickness, to assess the principle direction of diffusion tenor and to estimate the location of neural activities. However, conventional methods to estimate the vertex normal on the tessellated mesh are sensitive to irregular triangulation and noise signals occurred from the previous processing to construct a cortical surface model form MRI volume data. In the present project, an algorithm is proposed to estimate a normal vector on a vertex in a more precise fashion. Following the idea Normal Vector Voting algorithm [1], Modified Normal Vector Voting algorithm is implemented. Simulations on a sphere and a complex surface were performed with various amount of noise to assess the proposed method comparing to conventional ones. The errors of estimation from the proposed method were the least in all performed cases demonstrating the robust estimation to noise of the proposed algorithm. The large amount of computational load of the proposed algorithm remains as a further problem for practical application.