科技部
補助出席國際會議報告
(補助編號：101-2320-B-004-001 –MY2 )
Joint Annual Meeting of ISMRM-ESMRMB 國際醫用磁共振學會及歐洲生醫核磁共振醫學會
共同年會
2014/05/10~2014/05/16, Milan, Italy
會議心得報告
蔡尚岳 助理教授
政治大學應用物理研究所
2
一、參加會議經過
此會議乃磁振造影界一年一度的重要會議，今年為國際醫用磁共振學會第22屆 年會於義大利米蘭舉行，因舉辦地點位於歐洲，因此與歐洲生醫磁共振醫學會31 屆年會一起舉辦，上一次共同舉辦是在2010年於瑞典斯德哥爾摩，因此不僅發表 的論文水準相當高，此年度的論文數和參予人員通常會是往年的1.5 倍左右，此 次本人共有四篇文章發表。
會議於米蘭市附近的 Convention Center 舉行，議程共分五天進行，之前再外加 兩天的 educational courses。五天內的 Scientific Meetings 總共涵蓋不同主題的 oral presentations session。每天自早上七點開始一個小時的「Sunrise educational course」，針對 MRI 各領域邀請傑出研究專家學者演講與進行座談。每天有十個 不同主題同時進行，此部分主要請一些相當有經驗的學者，已上課的形式，介紹 各領域技術的基礎和發展，雖然涵蓋很基礎的部分但是也會有很深入的探討。本 人今年主要專注於高磁場應用部分。晨間的 session 結束後接下來會議當中安排 了幾場特別的演講 (Plenary Lectures)，請到領域中相當資深的研究人員主講，內 容涵蓋現今較先進之研究現況，如：Non-alcoholic fatty liver disease, Emerging biomarker of obesity 等等，特別是禮拜一和禮拜四的 Lauterbur 以及 Mansfield lecture, 請到 MRI 領域的資深的研究者談論目前的發展和未來，今年由 Prof.
Thomas M Grist 談到 MRI 的革新，禮拜四則請到 Prof. Denis Le Bihan 教授談到 關於利用擴散性影像了解生物體微結構的部分，受益頗多。
每天大會都會安排一家磁振造影設備大廠進行最新產品的介紹，對於目前 硬體上的發展，也提供相當多資訊。而且，由於世界各地優秀研究人員均會參 加，於休息時間，像 coffee break ，亦提供了一個跟其他國家研究人員交換意見 與討論的機會。會議中參與的人士涵蓋醫界、工程界與業界。不僅可以得到工 程學術上的研究經驗，對於臨床上一些應用也更深入的瞭解 這樣的交流與學 習，不僅可以對目前頂尖 MRI 技術有更深認識，對於未來研究方向也有所助 益。
3
二、與會心得
此次會議本人延續過去研究主題，以磁振頻譜以及相關量化分析等議題
(Quantitative Methods in Musculoskeletal Tissues、Normal Brain Physiology by MRS
& Other Modalities 、MRS of the CNS)；擴散影像應用及技術(Diffusion: Novel Acquisition、Diffusion Tractography)；fMRI 連結性相關議題(Functional
Connectivity、)；肝臟相關議題(Hepatobiliary )，幫助了解目前的發展趨勢。同時 關注的還有在目前 MR 硬體技術上的發展。在一般議程中可參與晚間的研究論壇 (study group)，由有興趣的研究人員共同參與討論，由資深研究人員發表實驗的一 些經驗，並由在場人員互相討論自身的使用情形，本人參與關於肝臟與肌肉脂肪 量測的議題。此外當然還有一些其他如硬體設備上的研究，以及廠商設備上的研 發現況。透過這樣的交流，使我對目前磁振造影界的發展與需求有更進一步的認 知。此行可為收穫頗豐。
三、攜回資料名稱及內容
會議論文隨身碟
四、發表論文附檔
共計發表論文四篇 僅附上論文摘要。
Figure 1. (Top) Averaged structural connectivity among 78 GM regions from all subjects. (Bottom) number of elements in groups with different connectivity probability
Figure 2. inter-subject CV, inter-scan CV and ICC for groups with different connectivity probability.
3274
The reproducibility of diffusion tensor imaging on brain connectivity measures between cortical regions using probabilistic tractography
Chun-Hao Huang1, Woan-Chyi Wang2, Yi-Ru Lin1, and Shang-Yueh Tsai2
1Electronic and Computer Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan, 2Graduate Institute of Applied Physics, National Chengchi University, Taipei, Taiwan
Introduction
Diffusion Tensor Imaging (DTI) associated with tractographic method has been used to investigate the structural or anatomical connectivity in the brain, which can be linked to dynamic behavior (functional connectivity) of the brain1,2. Recently, small world model in network analysis have been widely used to study the structural networks accessed by connectivity between spatially isolated grey matter (GM) regions1,3. Although streamline tractography has been used to track the continuity of fiber orientation along the principle diffusion tensor, the ambiguity of principle diffusion direction in GM regions make tractography strongly affected by noise1. Probabilistic tractography is an extension of streamline method that calculates the probability of connectivity between regions with consideration of the uncertainty of major diffusion direction. Therefore, this method should be more robust and suitable for study the structural connectivity between GM regions. In this study, we investigate the test-retest reliability of structural connectivity among cortical regions parcellated in automatic anatomic labeling (AAL) template using probabilistic tractography.
Methods
Fourteen healthy subjects, 7 male and 7 female, with ages raging from 20 to 25 years old (21.8± 2.1), were recruited. Data were collected on a 3T MR system (Skyra, SIEMENS Medical Solutions, Erlangen, Germany) with a 32-channel head coil array. All subjects were scanned twice for the assessment of test-retest reproducibility. For each subject, a high-solution 3D T1 images were performed for the anatomical information. DTI protocols were performed using spin echo EPI sequence. We used 30 gradient directions with b-value 1000 s/mm2 and five additional images with minimum diffusion weighting. Experiment parameters were TR = 8800 ms, TE = 90 ms, FOV=256x256 mm2, MAT=128x128, slice thickness = 2 mm, slice = 61, NEX = 2, acceleration factor = 2. The total acquisition time were 15 minutes including T1 and DTI scans.
The image preprocessing, estimation of diffusion parameters and tractoraphy for DTI was carried out using FSL (www.fmrib.ox.ac.uk/fsl/).
Preprocessing includes motion correction, eddy current correction and brain extraction. Then we estimate the transformation matrix between standard MNI space and DTI space for each subject. A total of 78 cortical regions (39 in each hemisphere) in AAL template can be warped from MNI space into native DTI space. The local probability distribution of fiber orientation for each voxel in brain was estimated. Then probabilistic tractography was applied by sampling 5000 streamlines fibers per voxel within regions to estimate the connectivity probability between seed GM region to target GM region. Connectivity between two GM regions was displayed as number of streamlines passing target regions divided by total streamlines from seed region. Structural connectivity matrix in the order of 78 by 78 can be generated for each subject for each scan. Coefficient of variance (CV) and intra class coefficient (ICC) were calculated to characterize the inter-scan reproducibility.
Results and Discussion
Structural connectivity matrix of 78 GM regions from average of all subjects is shown in figure1. The connectivity is low between most of GM regions, which is in agreement with previous reports3. Connectivity probability among all regions is summarized by separating connectivity into groups according to the probability. Inter-subject CV range from 33% to 13% and Inter-scan CV range from 22% to 9% for groups with connectivity strength over 0.05. Better reproducibility and less variability between subjects is found in regions with higher connectivity. For connectivity over 0.2, inter-scan CVs are at level of 10% and inter-subject CV at 13%. This implies that the DTI with probabilistic tractography can provide stable estimation on connectivity between GM regions with higher connectivity. Further, structural connectivity network of brain can be similar across subjects because network analysis is constructed majorly based on regions with higher connectivity strength. ICCs for all groups range between 0.7 to 0.8. In conclusion, we have investigated the inter-scan reproducibility of DTI on the estimation of structural connectivity between AAL cortical regions. Probabilistic tractography can be successfully applied to calculate the connectivity between GM regions at the presence of the ambiguity of principle diffusion direction in GM regions. Thus network analysis based on DTI and probabilistic tractography can be feasible.
Reference
1. Vaessen M.J., Hofman P.A.M., Tijssen H.N., et. al. The effect and reproducibility of different clinical DTI gradient sets on small world brain connectivity measures.
NeuroImage 2010.
2. Gong G., Yong H., Concha L., et. al., Mapping Anatomical connectivity patterns of human cerebral cortex using in vivo diffusion tensor imaging tractography.
Cerebral cortex, 2009
3. Gong G., Rosa-Neto P., Carbonell F., Age- and gender related differences in the cortical anatomical network. Journal of neuroscience. 2009
Proc. Intl. Soc. Mag. Reson. Med. 22 (2014) 2662.
Noncontact physiological measurements using video recording inside an MRI scanner Shang-Yi Yang1, Hsaio-Hui Huang1, Chi-Wei Liang1, Shang-Yueh Tsai2, and Teng-Yi Huang1
1National Taiwan University of Science and Technology, Taipei, Taiwan, Taiwan, 2The Graduate Institute of Applied Physics, National Chengchi University, Taipei, Taiwan, Taiwan
Target audience: MR physicists and researchers working on correcting physiological noise.
Purpose
Previous investigations showed that physiology parameters such as heart rate and respiratory rate can be measured using noncontact video recording1. This method is potentially useful in the MRI enviroment because it is an optical-based remote sensing which minmally interfers with the fast switching MRI gradient system. However, this method requires normal
ambient light as illumination source2. It is generally a low-light enviroment inside a conventional MRI scanner. This study attempts to evaluate the feasibility of the noncontact measurement method inside the MRI enviroment and optimize the computation algorithm for MRI applications.
Methods
We performed this study in a mock MRI scanner in Taiwan Mind and Brain Imaging Center, National Chengchi University. We used a conventional digital camera (16M pixels, 4/3-inch CMOS sensor, focus length: 20mm, aperture size: F1.7) for video recording. Figure 1(a) displays the position of the camera mounted at the top of a head coil. Three volunteers participated in this experiment. During experiments, we asked the subjects to keep their heads still and recorded one minute video for each session. The video file format was 640×480 MPEG-4. During the experiment, an operator recorded the radial pulse of the volunteer’s wrist using the pads of two fingers.
After experiments, we transferred the video files to a personal computer and performed data analysis using MATLAB® (Mathworks, Natick, MA, USA). Figure 1(c) shows the flow of image analysis. First, the three color channels, red (R), green (G) and blue (B) were separated. For each time frame, we averaged all the pixel intensities of the three images and produced three values. The procedure produced 3 signal-time curves corresponding to 3 color channels. We then used spline-based fitting to estimate the baseline drift of each curve and detrended the three curves by removing the baseline drifts from the raw signal curves. The three detrended curves underwent independent component analysis, component selection process and bass-pass filtering (0.1 - 5 Hz). Finally, we used peak detection to identify local maximum of the pulsatile curve to calculate heart rates (beats per minute) using 60 divided by time interval between two adjacent peaks.
Results
Figure 2 demonstrates the curves normalized to their initial values. Notice that the blue-channel curve fluctuates prominently. Figure 3 demonstrates the detrending procedure (top: estimating baseline drifts, bottom: the detrended curve). Figure 4 displays temporal heart-rate variations obtained from one of the volunteers.
Discussion
This study attempts to use video recording as a tool for physiology monitoring in an MRI scanner. During a cardiac cycle, facial skin blood perfusion changes alter optical path of ambient light emitted to the subject’s face. Using a conventional digital camera to capture the changes of the reflected light and using ICA analysis to remove other sources in light, we identify that this method is feasible in a low-light MRI bore.
This method is an optic-based technique which avoids the problem associated with switching gradient system. This method requires off-line computation and thus is not suitable for real-time applications such as triggering cardiac imaging or cine flow measurements. Nonetheless, it is potentially useful in physiology noise corrections such as the RETROICOR method3 to remove cardiac-related noise in resting state fMRI experiment. We performed the experiments in a mock MRI system, which is the limitation of this study. Utilizing a MRI-compatible digital camera for video recording during scans merits further investigations. In conclusion, physiology monitoring using video recording is a potential useful tool in MRI applications.
References
[1] Poh MZ, McDuff DJ, and Picard RW, “Advancements in Noncontact, Multiparameter Physiological Measurements Using a Webcam,” IEEE Trans Biomed Eng. 2011 Jan;58(1):7-11.
[2] Verkruysse W, Svaasand LO, and Nelson JS “Remote plethysmographic imaging using ambient light” Opt Express. 2008 December 22; 16(26): 21434–21445.
[3] Glover GH, Li TQ, Ress D. “Image-based method for retrospective correction of physiological motion effects in fMRI:
RETROICOR.” Magn Reson Med. 2000 Jul;44(1):162-7.
Figure 1 (a) Position of the camera (b) Separating RGB color channels (c) Data analysis flow diagram
Figure 2 Normalized intensity of 3 color channels
0 5 10 15 20 25 30 35 40 45 50
Figure 3 (top) Estimating baseline drift of the signal-time curve (bottom) the curve after baseline-drift removing
Figure 4 An example of the obtained heart rate changes during the experiment.
Proc. Intl. Soc. Mag. Reson. Med. 22 (2014) 1299.
Investigating the reproducibility on the quantification of £^-aminobutyric acid (GABA) in visual cortex Tzai-You Wu1, Chun-Hao Fang1, Yi-Ru Lin1, and Shang-Yueh Tsai2
1Electronic and Computer Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan, 2Graduate Institute of Applied Physics, National Chengchi University, Taipei, Taiwan
Introduction
γ-Aminobutyric acid (GABA) is major inhibitory neurotransmitter in human brain. Recently, GABA has been found to have high relationship with various neuron degenerative disorders. To measure GABA in brain, a spectral editing MRS technique called MEGA-PRESS sequence has been widely used2. However, performance on the quantification of GABA, which may be related to spectral editing efficiency and can be sensitive to noise due to its low concentration, needs to be carefully investigated. In this study, the inter-subject and intra-subject reproducibility of GABA quantification were evaluated using three quantification methods including integration, fitting using two Gaussian shape and LCModel.
Methods
In this study, sixteen healthy subjects (10 male, 6 female, age = 21.9±1.4) were included and scanned on 3T MR system (Skyra, Siemens Medical Solutions, Erlangen, Germany). MRS voxels were positioned to cover both left and right visual cortex. Water suppressed spectra were acquired using MEGA-PRESS sequence with editing pulse (20 ms Gaussian) alternated between 1.9 ppm and 7.5 ppm. The scanning parameters are: voxel size = 30x25x25mm³, TR/TE=2000/70ms and measurements = 260. Interleaved editing-on and editing-off spectra were acquired in the scan. A non-water suppressed MRS was acquired using PRESS sequence with the same experiment parameters. For each subject, GABA scans were repeated twice after adjusting shimming and frequency to access reproducibility of this method.
The post-processing of MRS data was done by a self-developed program in MATLAB. Spectra with resonance frequency shifted more than 13 Hz were excluded to ensure the acceptable spectral editing efficiency. Frequency shift information was acquired base on the NAA location in editing-off spectra. Frequency shift correction was carried out by aligning the NAA peak on editing-off spectra and same alignment was applied on editing-on spectra. Editing-on and editing-off spectra were averaged respectively. Spectral-editing spectrum was obtained by the subtraction of editing-on and editing-off spectra. Quantification of GABA signal was done by LCModel, direct integration and fitting using two Gaussian functions after linear baseline correction. Quantified GABA signals were normalized to water signal acquired from NWS scan and creatine (Cr) signal quantified on the editing-off spectra. Inter-subject and intra-subject coefficient of variance (CV) were calculated for these three quantification methods.
Results
Inter-subject and intra-subject CVs of quantified GABA were summarized in Table1. In general, integration and fitting shows similar GABA+/H2O and GABA+/Cr level. Three quantification methods exhibit similar intra-subject CV ranging from 11.1% to 15.3%. Inter-subject CVs ranges from 19.7% to 26.2%. Among all quantification methods, intra-subject CV of GABA+ quantified by fitting were around 3% higher than those quantified by LCModel and integration. Lowest inter-subject CV were found in the of GABA+ quantified by integration. GABA+ normalized to water gives slightly lower intra-subject CV than those to creatine.
Discussion and Conclusions
In this study, reproducibility accessed by inter-subject CV on the quantification of GABA signal was investigated using three quantification strategies and two normalized referenced. The higher intra-subject CV found in GABA+ quantified by fitting can be attributed to unstable fitting of GABA from baseline and line shape distortion in some subjects. The CVs can be improved when a more complex fitting algorithms is applied as results in LCModel. However, according to the consistency of averaged GABA+ level quantified in fitting and integration, we think the quantification of GABA+ on GABA spectra using simple fitting or integration algorithm can be sufficient compared to LCModel. Improved intra-subject CV in GABA+/H2O than GABA+/Cr may be attributed to higher signal to noise ratio of water than creatine.
In this study, reproducibility accessed by inter-subject CV on the quantification of GABA signal was investigated using three quantification strategies and two normalized referenced. The higher intra-subject CV found in GABA+ quantified by fitting can be attributed to unstable fitting of GABA from baseline and line shape distortion in some subjects. The CVs can be improved when a more complex fitting algorithms is applied as results in LCModel. However, according to the consistency of averaged GABA+ level quantified in fitting and integration, we think the quantification of GABA+ on GABA spectra using simple fitting or integration algorithm can be sufficient compared to LCModel. Improved intra-subject CV in GABA+/H2O than GABA+/Cr may be attributed to higher signal to noise ratio of water than creatine.