Quantitative Comparison between Hybrid Diffusion Imaging and Diffusion Spectrum Imaging

Y-C. Wu, A. L. Alexander Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, Psychiatry, University of Wisconsin-Madison, Madison, WI, United States Background At higher levels of diffusion-weighting, the diffusion tensor model is inadequate for describing non-Gaussian diffusion that arises from complex tissue architecture (e.g., crossing white matter (WM) fiber groups). Complex diffusion behavior in brain tissue may be studied with different approaches including diffusion spectrum imaging (DSI) [1], q-ball imaging (QBI) [2], and diffusion model fitting (CHARMED) [3]. Hybrid diffusion imaging (HYDI) combines the approaches of diffusion tensor imaging, high angular resolution, multiple diffusion-weighting levels [4]. The HYDI q-space method consists concentric shells of icosahedral encoding directions at different diffusion-weightings (DW) (b values). While the whole dataset may be used for DSI, the inner shell is suitable for diffusion tensor imaging (DTI), and the outer shell with higher angular sampling resolution may be used for QBI. In this study, several diffusion distribution measurements derived from HYDI, DSI and QBI were compared quantitatively. Monte Carlo noise simulations with different complex diffusion patterns (single and crossing WM with “fast” and “slow” diffusing signal components) were used to study the compare of image noise on DSI, HYDI, QBI for different SNR and q-space ranges. Methods Reported diffusion measurements in corpus callosum were used to simulate fast and slow Gaussion diffusion functions [5]. Simulations included single-fiber groups with four spatial orientations (azimuthal angle: 0°, 30°, 60° and 90°) and two-crossing WM fiber groups at four intersection angles (45°, 60°, 75° and 90°). For each geometry, six levels of SNR were evaluated using either 90 or 150 Monte Carlo trials. The q-space noised diffusion signal was sampled and processed using both DSI and HYDI encoding schemes. The DSI q-space sampling was a 7x7x7 Cartesian lattice with a spherical aperture giving 100 total diffusion-encoding directions. The HYDI sampling scheme (Table 1) consisted of four icosahedron shells also with 100 encoding directions. The outer shell was used to evaluate a QBI experiment with 50 encoding directions. Four maximum b values (4000, 6000, 8000 and 10000 sec/mm) were studied. The b value of each HYDI shell was decreased propotionally with the maximum b value. For both DSI and HYDI, the diffusion distribution probability density function (PDF) was estimated by Fourier Transform of the q-space samples [1]. Three displacment PDF measures were calculated including the zero