Diffusion Signal Decomposition Using Periodical Sampling in Gradient Direction Domain and Fourier Approximation

Farshid Sepehrband¹, Jeiran Choupan¹, Nyoman Dana Kurniawan¹, David C. Reutens¹, Zhengyi Yang²

In this work we proposed a novel data driven approach to decompose restricted diffusion from diffusion signal, and validated it using ex-vivo mouse brain tractography. To achieve this goal, first, we impose a periodic spiral sampling in gradient direction domain. Then, we applied low pass filter based on finding the optimal cut-off frequency to decompose the diffusion signal. We assume that restricted diffusion is sensitive to sampling orientation, while free diffusion and noise are orientationally independent. Therefore, restricted diffusion contributes in the low frequency parts of diffusion signal, while free diffusion and noise contribute in high frequencies.