Notes from Spinal Cord Hack – Milan 2014
The 44 researchers and clinicians attending “Hack” participated in a small series of oral presentations intended to provide background for newbies and identify the pressing issues and state of the art in spinal cord MRI. Following this, break-out groups met to discuss attendee-determined topics. At least 40 abstracts on advanced MR applications in spinal cord were presented at ISMRM’14, so we are pleased to have more than 40 attendees at this “Hack”.
Paul Summers (UniMoRE – Modena, IT) reviewed problems he’s encountered in over seven years of MRI of the spinal cord, and particularly functional MRI.
A copy of Paul’s slides is available here.
– anatomy – separation of the GM from the larger superficial draining veins may make the distinction of large and small veins – i.e. spin-echo vs gradient echo imaging a consideration – which has received little attention per se
– physiology – spinal cord and CSF motion are present – but how to best account for them, especially as Fourier transform based MRI leads to non-local manifestations of artifact.
– hardware – with both existing and emerging hardware, MR signal reception from the spinal cord is significantly penalized relative to brain studies.
– geometry – to accommodate the cord size, we tend to push for higher resolution, which together with above coil sensitivity and accentuated physiological noise contributions lower signal to noise ratios that need to be accounted for in study design
– homogeneity 1 – a large scale valley in the magnetic field is typically seen in the cervical spine; as noted by Anke Henning, even higher order shims may not be adequate (due to their limits) to correct for this globally – a slicewise correction may be a more robust approach. The use of susceptibility matching materials might help, but attendee experience with rice, antacid, and rubber pads have been disappointing. An e-pub paper on pyrolitic graphite foams looks promising.
– homogeneity 2 – superposed on the above inhomogeneity are variations at the scale of the vertebrae, which are not strictly dependent on vertebra disc susceptibility differences, but also on the inter-verterbal tissue posteriorly and laterally, as well as the shape and positions of the vertebral processes and laminae. A further reason to correct for magnetic field at slice level.
– homogeneity 3 – respiration creates position dependent variations in magnetic field which reaches tens of Hertz (>70 reported by Julien Cohen Adad). Notably this locks changes in signal intensity into the time-series by modulating slice selection and spatial encoding. This is well seen in the non-motional deformation of the spinal cord seen in this video. Dynamic shim correction may be advisable.
Julien Cohen-Adad (Ecole Polytechnique, Université de Montréal, Montreal, CA) presented the Spinal Cord Toolbox, which is a comprehensive and open-source library of analysis tools for multi-parametric MRI of the spinal cord.
A copy of Julien’s slides is available here.
The toolbox includes a template and several atlases (white and gray matter, spinal tracts, spinal and vertebral levels, etc.). Although other groups have proposed spinal cord templates, none of them included a method to estimate diffeomorphic transformations to register data between the native space and the template (forward and backward), as done in classical brain analyses. The toolbox fills this gap by providing methods for estimating such warping fields. It also includes useful scripts for data preprocessing: extraction of centerline, automatic segmentation, slice-wise motion correction, etc.
Spinal cord template constructed from state-of-the-art methods (using MNI tools)
Probabilistic atlas of gray and white matter
Atlas of spinal pathways (30 tracts)
Probabilistic location of spinal levels
Robust registration methods to the template (output warping fields compatible with ANTs, FSL, SPM)
Fast, robust and automatic segmentation of spinal cord
Slice-wise motion correction for dMRI and fMRI data
User interface: command line (like FSL or AFNI).
Environment: OSX, Linux
The software has been released on June 15th 2014 after being evaluated by 20 beta testers. It is in active development. Since its release it has been downloaded 159 times. Various links to the projects:
Patrick Stroman (Queens U, Kingston, CA) provided a recorded video presentation in which he provided an update on the techniques use in his lab for spinal cord fMRI, and the software they have developed for the analysis of said data.
A copy of Patrick’s presentation at SpinalCordMRHack14 can be seen here.
A more general introduction is available here.
– sagittal, single shot spin echo imaging allows TE optimized functional imaging with extended coverage and greatly reduced sensitivity to field homogeneity issues respect to gradient-echo BOLD methods
– a standardized cervical spine geometry has been developed to support group analysis of fMRI data.
– a software package tailored to the motion correction, co-registration and analysis of sagittal cervical spine fMRI data has been developed.
The software package will be available from mid June – please visit: http://post.queensu.ca/~stromanp/software.html
or email: patrick.stroman (at) queensu.ca.
Moreno Pasin (UCL, London, UK) performed a walk-through of the use of DRIFTER for the creation of physiological confound variables from cardio-respiratory log-file data.
– the program can be used to do data classing directly or to prepare a series of confounds.
– audience members noted that other software packages are available that can do similar, and it may be better to incorporate confound extraction into the data analysis rather than rely on a pre-processing data-cleansing. Moreno noted that the confound files should be usable in other fMRI analysis packages, and that DRIFTER could perform analyses with task regressors included.
– a concern was raised for how, the possibly large numbers of confounds available impact upon the degrees of freedom of the data.
Because Moreno provided a live demo, interested readers are referred to the Drifter website (http://becs.aalto.fi/en/research/bayes/drifter) for further information. A reference for Drifter is:
Simo Särkkä, Arno Solin, Aapo Nummenmaa, Aki Vehtari, Toni Auranen, Simo Vanni, and Fa-Hsuan Lin (2012). Dynamical retrospective filtering of physiological noise in BOLD fMRI: DRIFTER. NeuroImage, 60(2): 1517-1527.
Torben Schneider (UCL, London, UK) described their experience in establishing a spinal cord volumetry protocol, with considerations of sequence optimisation, motion control and reproducibility.
A copy of Torben’s slides is available here.
Tools and sequences:
– Whole cord volumetry is pretty established and works very well
– Many tools for (semi-) automatic segmentation are available (Lossef, cord finder in Jim6, Spinal Cord Toolbox); we use Jim6
– We advocate PSIR (standard on our Philips scanner) for great high resolution GM/WM contrast in cervical spine; immobilisation is key (i.e. collar); we get very reproducible GM/WM segmentation results
– Use same methods in lumbar spine: lumbar-sacral enlargement as reference point
GM/WM segmentation in DTI: Registration is very unstable; we use average DWI image for optimal contrast
– We are currently comparing spinal cord toolbox segmentation results with our established pipelines
– Julien Cohen-Adad raised the issue of inter-modality (PSIR vs T1) biases in whole cord segmentation;
– There is no real ground truth for in-vivo volumetry results – which method is “better”?
– Nico Papinutto reported that PSIR is available for Siemens systems (as used in his ISMRM’14 poster).
Alex Smith (Vanderbilt U, Nashville, USA) – B0, B1, M0 etc in Cervical Spine.
A copy of Alex’s slides is available here.
– applied the single-point qMT method to cervical spinal cord in vivo.
– evaluated assumption validity in single-point model in disease and healthy tissue.
– SC qMT can provide MPF maps in vivo with excellent grey/white matter contrast.
– observed MPF values were higher in healthy WM and GM than in MS lesions.
– this technique may provide a reliable outcome for evaluating spinal cord disease.
Robert Barry (Vanderbilt U, Nashville, USA) – 7T spinal cord fMRI
The material covered in this talk s largely covered by the recently published article: “Resting-state functional connectivity in the human spinal cord“.
– presentation of 7T functional images (voxels = 0.91 x 0.91 x 4 mm3) and movies, and relevant pulse sequence parameters
– discussion of approaches to mitigate image distortions and physiological noise, which are non-trivial issues at 7T
– resting state spinal cord fMRI were acquired in 22 healthy subjects
– seed-based functional connectivity analyses were performed, and both single-subject and group-level results were presented
– it was noted that between the sagittal spin echo fMRI, diffusion tensor analyses and spinal volumetry work to date, it appears that at least three reference geometries are emerging. A request was made and supported by the attendees that steps be taken to facilitate interchangeable between formats, and consistency of reporting results between systems; ideally by having a unified atlas across applications.
– An enquiry as to who would be interested in purchasing pyrolytic graphite pads for magnetic field inhomogeneity correction (5 groups expressed interest). Regarding this, enquiries are in course with the authors of the JMRI paper cited above, and with a foam padding producer as to possible sourcing of said materials.
– A show of hands indicated that people were interested to have a future Spinal Cord MR event (possibly not with “Hack” in the title due to limited software development foreseen). Suggestions included Hamburg at OHBM14, Boston – Resting State Meeting (Sept), ISMRM’15, and a separate dedicated event. This will be discussed via email.
– While support of a spinehack page on the brain-hack site has been organized, there remains to be established a site / medium through which the communication needs of the spinal cord MR community can be met. To be discussed via email.
Side Session 1: AFNI users
A mini break-out group briefed afni developers Robert Cox and Ziad Saad on issues encountered in the post-processing of spinal cord fMRI data (e.g. need for slice-wise corrections, large surround of non-interest, variable contrast between cord and CSF, limitation of motion-correction search-space, …). The afni developers indicated that some of these issues could likely be addressed with careful use of existing afni routines (see afni site: http://afni.nimh.nih.gov/afni), but a few would likely require more detailed consideration. They further encouraged us to communicate our questions with them via the afni-users discussion board.
Side Session 2: Q&A: how to and how not in Spinal Cord MR
An experience-sharing discussion was led by Julien Cohen-Adad covering topics emerging from the attendees, including:
motion correction strategies
registration to template
identification of nerve roots (spinal levels)
H Lundell (DRMCR, DK), who was not able to attend the Hack has provided some slides regarding the high resolution post-mortem monkey spinal cord DTI dataset his group has collected. The dataset is available for use by others via their website (http://dig.drcmr.dk).
List of Attendees:
Niels Peter Bergsland
Davide DuzziFalk Eippert
Maria A Rocca
Johanna Vannesjö Signe
Comments and Corrections, contact: paul.summers (at) unimore.it