Yue Li1, Hangyi Jiang1,2, Can Ceritoglu3, James Pekar2,4, Michael I Miller1,3, Susumu Mori1,2, and Andreia Vasconcellos Faria2
1AnatomyWorks, LLC, Baltimore, MD, United States, 2Department of Radiology and Radiological Science, Johns Hopkins University, Baltimore, MD, United States, 3Center for Imaging Science, Johns Hopkins University, Baltimore, MD, United States, 4F. M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States
Synopsis
Brain resting state fMRI
(rs-fmri) is a useful tool for research, although its clinical impact is still
limited, mainly because data from single subjects is low in power. We have
shown that this limitation can be addressed by “borrowing strength” from the
population, specifically by eschewing voxel-based analyses in favor of assessing
connectivity between atlas parcels [1]. We now present user-friendly software
for signal processing, integrated with structural analysis, under a web-based
platform "BrainGPS" (https://braingps.mricloud.org). Users can submit
data and download results (co-registered structural and functional images,
rs-fmri time-courses, and parcel-based correlation matrices) in a few minutes.Purpose
Brain resting state functional MRI (rs-fmri) is a highly useful resource for neuroscience, although its clinical impact is still limited. Reasons for such limitation include the technical details in the pre-processing, usually not trivial for non-experts, the no-intuitive biological translation of the signal analysis, and the large spatial dimension of the functional connectivity analysis in the widely used rs-fmri packages that work at the voxel level. The spatial dimension reduction by methods like filters or independent component analysis is a subject of active research. Alternatively, structure-based approaches have been proposed. These approaches, however, require incorporation of accurate structural identifications into the rs-fmri analysis pipeline. User-friendly software for signal processing and integration with structural analysis and other MRI modalities, would amilorate these issues. We created a web-based plataform for rs-fmri procesing and analysis, under BrainGPS (https://braingps.mricloud.org)[1]. By offering this tool in a Software as a service model (SaaS), we facilitate the rs-fmri analysis for non-experts and generate results in a more comprehensive format, which aids the dissemination of rs-fmri among the research and clinical communities
Methods
Users of MriCloud can either create their own accounts or simply login using their google accounts. After logging in, several available services are listed. For the rs-fmri processing, the users need to: 1) Submit T1-WI for a fully-automated, multi-atlas segmentation, region of interest (ROI) volume calculation and visualization pipeline based on LDDMM [2,3], and 2) Submit the rs-fmri images (Fig. 1). The rs-fmri dynamics will be coregistered, motion and slice-timing corrected based on SPM routines. The T1-WI and respective parcellation map obtained in the previous step are coregistered to the functional dynamics and time courses are obtained for more than 289 ROIs. The gray matter time courses from 78 ROIs are regressed for physiological nuisance by applying the CompCor [4] algorithm and the motion and intensity outliers are identified by ART, an SPM toolbox. For more details on the processing pipeline, please read [5]. The service accepts Analyze format data after eliminating patients' identity information, which can be converted from DICOM data using tools provided by the website. By now all the processing parameters, except by the repetition time, TR, are taken from the image header and default settings. In the future, we will allow the users to change settings as well as input parcellation maps obtained from other sources, so one can easily “customize” the analysis
Results
Results are available for download in about 2 minutes (for an rs-fmri protocol of about 200 dynamics) and consist co-registered images (rs-fmri / T1-WI / parcellation maps) for quality control and matrices of structural rs-fmri time-courses and correlations for statistical analysis (Fig. 2 and 3).
Discussion and Conclusion
By creating a Software as a Service Cloud-based Platform for rs-fMRI processing we make the rs-fMRI analysis much more accessible and comprehensive for non-expert users. Also, the structure-based format allows integrating multiple image features, such as volume, shape, Diffusion Tensor indices, perfusion values, etc. These pipelines for different image modalities are available – or will be shortly available – under our platform, providing a precise and global characterization of individual brains. This will be of great benefit of both research and clinical studies, enabling, for instance, the organization of big databases for archiving and retrieving image and non-image information, needed for high- throughput analysis and personalized medicine.
Acknowledgements
NIH/NIBIB
grant R03 EB014357 and AHA grant 12SDG12080169 (AVF), NIH P41 RR015241 (JJP, SM, MIM), NIH/SBIR grant 2R44NS078917-02A1 (YL,HJ)References
1. Faria A.V., et al. Atlas-based analysis of resting-state functional
connectivity: evaluation for reproducibility and multi-modal anatomy-function
correlation studies. Neuroimage 2012. 61(3): 613-21.
2. Li, Y., et al. BrainGPS: A Cloud-based
Platform for Neuroimage Analysis and neuroradiological Studies. Proc.
Annu. Mtg. Intl. Soc. Mag. Reson. Med. 23(2015), 3498.
3. Beg, M.F., et al., Computing large deformation
metric mappings via geodesic flows of diffeomorphisms, Int J Comp Vision 2005.
61(2), 139-157.
4. Tang, X., et al., Bayesian Parameter
Estimation and Segmentation in the Multi-Atlas Random Orbit Model. PLoS ONE
2013. 8(6): e65591.
5. Behzadi, Y. A component based
noise correction method (CompCor) for BOLD and perfusion based fMRI. NeuroImage
2007. 37 (1): 90–101