• Top page
  • Timetable
  • Per session
  • Per presentation
  • How to
  • Meeting Planner



Image/Sound Processing

開催日 2014/9/12
時間 11:00 - 12:00
会場 Poster / Exhibition(Event Hall B)

ラット脳のVoxel-based MorphometryにおけるNon-local Meansノイズ除去フィルタの効果
Effects of Non-local Means Denoising Filter on Voxel-based Morphometry in Rat Brain

  • P2-375
  • 大石 直也 / Naoya Oishi:1 吉井 崇喜 / Takanobu Yoshii:2 福山 秀直 / Hidenao Fukuyama:1 
  • 1:京都大院・医・脳機能統合研究センター / Human Brain Research Center, Kyoto University Graduate School of Medicine 2:京都府立医大・精神機能病態学 / Department of Psychiatry, Graduate School of Medical Science, Kyoto Prefectural University of Medicine 

Although recent advances in magnetic resonance imaging (MRI) can achieve high spatial resolution for voxel-based morphometry (VBM) in small animals, more spatial resolution causes lower signal to noise ratio (SNR). To increase SNR, not only a longer acquisition time but also a post-processing denoising method is important. Recently, a non-local means (NLM) filter has been proposed, which can effectively remove noise with preserving edge in return for computational burden. We have therefore developed an accelerating scheme of the 3D NLM filter by general-purpose graphics processing units (GPGPU), which enables massively parallel computing. The purpose of our study is to evaluate the effect of the filter on VBM in rat brains. Sixteen male Sprague-Dawley (SD) rats were divided into two groups: seven rats that received single-prolonged stress, which caused posttraumatic stress disorder, and nine control rats that only received ether anesthesia. Ex-vivo 3D isovoxel MRI was conducted on a 7 T scanner with a 3D fast spin echo sequence (voxel size=137μm). MR images were segmented into gray matter and spatially normalized by Unified Segmentation and diffeomorphic anatomical registration using exponentiated lie algebra (DARTEL) in SPM8 with an in-house SD rat brain template, probabilistic segmentation maps, and templates for DARTEL. The processed gray matter images were then modulated with Jabocian determinants from the DARTEL procedure. They were smoothed with an isotropic Gaussian kernel with 50, 100, 200, 400, and 800 μm full-width at half maximum and compared the two groups by VBM analyses. All the steps were also done with denoised images using an in-house GPGPU-based 3D NLM filter software. As a result, gray matter regions were better segmented in the denoised images compared with the original ones. Furthermore, VBM analyses revealed more consistent results in the denoised images regardless of degrees of smoothing. The results suggest that the NLM filter improves the quality of the segmentation and spatial resolution for VBM analyses in rat brains.

Copyright © Neuroscience2014. All Right Reserved.