Estimation of the Underlying Fiber Orientation Using Spherical k-means Method from the Diffusion ODF in HARDI Data

This paper presents a new approach for detecting the underlying fiber directions in a voxel. The main idea is to use the principal direction (centroid of orientation class) of an orientation population instead of the classical maximal direction of diffusion orientation density function (ODF) for fiber orientation. Firstly, diffusion orientations from the ODF of raw data have been classified in accordance with the expected fiber populations. The centroids of diffusion orientations are then determined using the spherical k-means method so as to estimate fiber orientations. The proposed method is based on the reconstruction of diffusion ODF using spherical harmonic (SH) decomposition and the characterization of diffusion anisotropy in a voxel. It can approximate fiber orientations accurately and avoid the spurious detection of fiber orientation which is often observed with traditional methods. By using a variety of synthetic, phantom and real datasets, the experimental results demonstrate the effectiveness of the proposed method.