Desktop version

Home arrow Computer Science arrow Computational Diffusion MRI: MICCAI Workshop, Athens, Greece, October 2016


Performance Evaluation

The proposed method was evaluated on MRI datasets of 15 healthy volunteers from the Human Connectome Project (HCP), Wu-Minn[1] database . The acquisition protocol included three shells (b-values of 1000, 2000, 3000), 96 unique directions for each, isotropic resolution of 1.25 mm and imaging matrix of 144 x 168 x 111 pixels [11]. The diffusion data were preprocessed using the HCP diffusion pipelines [12], which included susceptibility, eddy-current and motion distortions correction, and registration to a common space. WM fibers were obtained using Q-Space Diffeomorphic Reconstruction (QSDR) reconstruction and tractography by DSI- Studio.[2] Tractography was terminated upon reaching 1M fibers. In this section we evaluate the performance of the proposed scheme. There are three main aspects we would like to explore: how well does the sparse representation approximate the original fibers; how well is the original similarity measure approximated by CWDS; and is it possible to use a common dictionary for different fiber-sets. The experiments that are described next shed some light on these questions.

  • [1]
  • [2] Developed by Fang-Cheng Yeh from the Advanced Biomedical MRI Lab, National TaiwanUniversity Hospital, Taiwan, and made available at
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >

Related topics