MRI and machine learning to improve early prognosis and clinical management after spinal cord injury - PROJECT SUMMARY/ABSTRACT
Purpose/Hypothesis: Spinal cord injury (SCI) causes substantial social, economic, and health burden.1 For
individuals with motor incomplete SCI, some basic ability to stand or walk is expected during the recovery
process,2 and this is a top priority in rehabilitative programs.3 However, establishing a prognosis for recovering
community walking ability is extremely difficult.4 Within 72 hours after SCI,5 edema develops within the damaged
spinal cord. This edema is a hallmark of spinal cord injury, expressed as signal hyperintensity using T2 magnetic
resonance imaging (MRI).6 Correlations between the sagittal length of this spinal cord edema and walking ability
are generally poor.7,8 However, advanced but available high resolution axial T2-weighted MRI to quantify spinal
cord edema in people in the acute stage of SCI may improve prediction of walking ability.9,10 The early clinical
management and targeted rehabilitation of these individuals could be drastically enhanced, optimizing recovery
and rehabilitation outcomes. The main objective of this research project is to use early axial spinal cord
MRI sequences as neuroprognostic biomarkers to improve the prediction of residual motor function.
This objective will be realized by the implementing the following specific aims: Aim 1: To establish to what extent
the axial damage ratio biomarker, measured by high-resolution axial T2-weighted structural imaging, can predict
residual function in persons with SCI. Previously-collected axial T2-weighted spinal cord structural MRI data of
200 people with SCI from the US Model SCI System at Craig Hospital will be used to quantify cord damage. This
metric will be related to the primary 1-year status-post injury outcome measures, which are clinical records of
walking ability and function. Multivariate statistical analyses will be applied to create exploratory models to
determine the prognostic value of the MRI measures. We hypothesize that the axial damage ratio can be used
in the acute stage as an accurate and objective neuroprognostic biomarker of residual motor function. Aim 2: To
identify the relationship between damage to specific spinal cord regions and specific motor and sensory deficits.
MRI data from Aim 1 will be used. Spinal cord regional damage analysis will be related to right and left upper
and lower extremity motor and sensory scores. Correlational statistical analyses will be applied to analyze
relationships between specific tract damage and motor/sensory deficits. We hypothesize that damage to
descending lateral corticospinal motor regions is related to ipsilesional motor deficits, and that similar findings
exist for ascending sensory regions and sensory deficits. For both Aims, we will compare our manual damage
quantification to our machine learning approach to automatically detect spinal cord damage. Aim 3: Develop,
test, and distribute a machine-learning based analysis pipeline for spinal cord damage measures. We will use
functions included in the Spinal Cord Toolbox and the open-source NiftyNet deep-learning platform to develop
the machine-learning based analysis pipeline. The processing steps will include spinal cord detection, spinal
cord damage segmentation, registration to the spinal cord template, and the calculation and output of the axial
damage ratio and regional damage biomarkers.
Significance: Successful completion of these Aims will advance the NIH/NICHD NCMRR aim: “to enhance the
health, productivity, independence, and quality of life of people with physical disabilities.” The significance of this
outcome relates directly to improving the clinical management of SCI. This research may inform clinicians,
patients, and families, regarding the percentage chance of regaining walking ability. The healthcare team will be
able to determine, early-on, which people will optimally respond to locomotor training. This work will significantly
improve the prognosis for recovery of walking and specific motor/sensory function based on early imaging of the
damaged spinal cord.
