Level set fiber bundle segmentation using spherical harmonic coefficients

Classifying brain white matter fibers into bundles is of growing interest in neuroscience. Quantification of diffusion characteristics inside a fiber bundle provides new insights for disease evolutions, therapy effects, and surgical interventions. In this paper, we present a novel method for segmenting fiber bundles using spherical harmonic coefficients (SHC) that describe diffusion signal obtained from High Angular Resolution Diffusion Imaging (HARDI) protocols. Based on SHC, we define a similarity measure and use it as a speed function term in level set framework. We show advantages of the proposed measure over similarity measures based on Diffusion Tensor Imaging (DTI) indices. Without any assumptions about diffusion model, we deal with diffusion signal instead of orientation distribution function (ODF) calculated using complicated mathematics, inaccurate simplifications, and time-consuming implementation. By applying the proposed algorithm on synthetic data, its superior accuracy and robustness in low SNR conditions are shown. Application of the proposed method on real HARDI MRI data also illustrates its superior performance, especially in heterogeneous diffusion areas with low traditional diffusion anisotropies.