ARE MRI TECHNOLOGIES CRUCIAL TO OUR UNDERSTANDING OF SPINAL CONDITIONS?

Low back and neck pain are disabling conditions,¹²⁴ projected to dramatically increase personal and public socioeconomic burden as the world ages.^{12; 26; 50} International experts have urgently called for renewed explanations and strategies to mitigate the persistence of spinal pain.^{12; 50} To that end, skeletal muscles, the spinal cord, and brain are receiving more attention as advancements in magnetic resonance imaging (MRI) technologies and analysis methods allow for improved visualization and quantification of their morphology.³⁷ Growing in parallel is radiomics, the field of study that aims to mine large amounts of quantitative features from medical images using data characterization algorithms.^{51; 135} As such, large datasets characterizing both healthy and diseased soft tissues with the patient’s pain experience may add to the biopsychosocial model in understanding spinal pain.

Changes in soft tissue morphology may represent 1) biological markers of pathology/trauma/pain,^{31; 32; 59} 2) ‘normal’ senescence,^{24; 122} or 3) an imaging variant with little, if any, prognostic or clinically relevant value.^{3; 16; 49; 87; 94; 120} Such conflicting evidence has resulted in suspicion of the value of imaging, particularly clinically where over-imaging has seeded doubt in the merit of image-identified (pathoanatomical) outcomes. However, readers will recognize that such findings may be influenced by inconsistent methods across research studies, limited image-resolution with conventional devices, use of historical imaging sequences, and the limitations of a purely biomedical approach for explaining persistent symptoms. As technologies advance, so too does our potential to understand the complexity of spinal disorders. MRI advancements^{27; 66} are allowing for prize-winning phenotypic discoveries regarding the vertebral column (intervertebral disc,^{81; 92; 108} vertebral end-plates,^{93; 139} and vertebral bone marrow.^{28; 102}) Studies examining soft tissues are similarly promising.^{1; 31; 55; 56; 67; 70}

The potential to bring together (and compare) distinct types of bio-psycho-social data, including the data generated by MRI, represents a substantial departure from the status quo of isolated pathoanatomical imaging. As clinical scientists employing MRI methodologies in our research, we recognize the exciting potential of exploring diverse data sources, bringing them into a common computational environment where big-data associations can be explored, further scrutinized, and settled.

Therefore, the purpose of this clinical commentary is two-fold; to summarize key and new literature and to highlight future directions in imaging spinal skeletal muscles, the spinal cord, and brain toward better understanding spinal conditions and the individual experience. We intend to make a case for the promise that advancing MRI technologies and radiomics have in bringing soft tissues forward in a collaborative clinical and research effort to limit the economic, societal, and personal burden of conditions affecting the spine.

SPINAL SKELETAL MUSCLE IMAGING

MRI is the gold standard for examining the integrity, size, and quality of spinal muscles due to its higher resolution, greater soft-tissue contrast, superior visualization of spinal landmarks, and potential for efficient semi- or automated segmentation. Water and fat images derived from multi-echo acquisitions (like Dixon) are superior in soft tissue analysis;^{40; 105; 136} yet, population-based ‘traditional’ T1-weighted and T2-weighted images remain widely used, and represent a valuable resource for investigators.^{48; 57; 91; 112; 113} Quantification techniques²⁴ achieved by segregating representative fat pixels within a selected muscle region of interest, are more accurate and have higher reliability than earlier qualitative¹ and semi-quantitative⁹⁰ methods. Further, quantification permits improved longitudinal assessment of temporal changes in participants with varying levels of pain, disability, and function (FIGURE 1). A downside to such an approach remains the time required to manually segment muscles of interest in post-hoc fashion. Depending on the muscle and region of the body, it is not unusual for an experienced user to spend approximately 45 minutes to well over an hour to manually segment each targeted muscle (based on the authors’ personal experiences). However, a new landscape of semi-automated and automated platforms is becoming available^{69; 70; 72; 90; 134} and we are only recently realizing the potential of machine learning algorithms towards streamlining the measurement of soft-tissue morphology. We, and others, expect such applications to dramatically and positively change our research capacity, output, and potential translation to clinical practice.

Figure 1a-b -.

Example of axial T2-weigthed lumbar paraspinal muscle morphology and composition measurements. a) illustrates the cross-sectional area measurements of the psoas (1,2), quadratus lumborum (3,4), multifidus (5,6) and erector spinae (7, 8), and also demonstrates current methodological differences in ROI definitions for the erector spinae muscle; the fat-filled “tent-region” under the lumbosacral fascia (posteriorly) was included in (7) and excluded in (8). b) illustrates the functional cross-sectional area (in green) representing the area of lean muscle mass (excluding fatty infiltration) of the multifidus muscle using a thresholding technique.

Change to paraspinal muscle composition (typically characterized by fatty infiltration) is widely examined and an accepted degenerative feature.^{38; 106; 112; 113} However, strong confounders like age,^{22–24; 122} and sex,^{24; 57; 109}, in addition to inconsistencies in methodologies, confuse the clinical relevance to pain.^{53; 103} Body composition,^{48; 122}, physical activity levels,^{49; 57} pain duration^{52; 127} and location (bilateral versus unilateral),^{14; 46} and co-existing pathology^{43; 100} may also influence spinal muscle composition. While limited evidence suggests that reduced multifidus cross-sectional area (CSA) predicts future (12 months) low back pain (LBP) in men,^{44; 103} their role and clinical relevance in the development of LBP and neck pain is far from explicit. Important limitations pertaining to understanding causation remain – does reduced muscle quality cause spinal pain, or vice versa? And further, can spinal muscle composition be modified with intervention(s) in health or disease, and if yes, what kind(s) of interventions, and at what age or injury-degenerative stage is intervention optimum?

Interestingly, non-uniform proportions and change to fatty infiltration between paraspinal muscles has been shown,^{22; 23; 48} alongside higher proportions of fatty infiltration in axial than peripheral skeletal muscles.²⁵ These findings may have important implications for muscle function and targeted therapeutic interventions. While intuitive that greater paraspinal muscle fatty infiltration negatively impacts muscle function, previous studies showed no clear association with percent thickness change during submaximal voluntary contraction,⁷⁶ muscle endurance²⁵ or muscle structure.⁵² However, greater paraspinal fatty infiltration has been associated with lower physical function,^{45; 47; 58; 114} and lean muscle mass (excluding fatty infiltration) to muscle strength.⁴² Furthermore, better surgical outcome^{73; 117} and effective therapeutic exercise⁵⁸ are achieved in those with superior muscle composition, which may reflect the superiority of muscle quality/composition rather than size in characterizing muscle tissues.³³ Clarification is necessary.

Future directions for muscle imaging: measurement and analyses

Although highly reliable,^{21; 35} current MRI techniques and measurement methods of paraspinal muscle composition are time-consuming and not always feasible for clinical settings. Moreover, the use of proprietary image analysis software and insufficient description of measurement protocols hinders replication efforts. Currently, most manual segmentation techniques are tedious and rater-dependent, motivating the development of automated or semi-automated segmentation methods. Automated segmentation of spinal muscles lags behind that of other soft tissues like the brain^{71; 110} and larger (thigh) muscles.⁹⁷ Automated thresholding algorithms and correction tools for lumbar paraspinal muscle analysis have been developed;^{47; 123} however, full automation requires advancements in computer-assisted imaging analysis involving atlas-based algorithms referencing a standardized coordinate system¹³⁴ as used for the heart¹³⁷ and brain.^{41; 133}

The next generation of developments in spinal muscle phenotyping and analyses are atlases that preserve typical spine anatomy while largely reducing the individual differences in muscle shapes and composition.¹³⁴ By way of example, lumbar multifidus is a complex transversospinales muscle comprising ‘typical’ orientation and length of fibers located superficial (approximately 4) and deep (approximately 2) to the intervertebral segments they span.²⁰ While individual differences may exist in overall multifidus shape and composition, the impact on function is largely distilled to their anatomical fundament and therefore can be employed in prediction methods to understand healthy ageing and spine conditions.^{64; 65; 112; 113} FIGURE 2 provides examples of current use of MRI in examining and detailing the composition of muscles traversing the cervical spine.

Figure 2 -.

Example of axial fat-only images examining composition of the multifidus and semispinalis cervicis traversing the cervical spine. a) illustrates the cross-sectional area measure of the multifidus and semispinalis muscles in the mid-cervical region (C3) and b) illustrates the cross-sectional area measure of the multifidus and semispinalis muscles in the lower-cervical region (C7). c) illustrates the propagated volume of the multifidus and semispinalis cervicis spanning those vertebral levels.

SPINAL CORD IMAGING

Conditions typically resulting from head/neck trauma (whiplash associated disorders (WAD)) from a motor vehicle crash (MVC), or spinal cord injury (SCI)), and degenerative cervical myelopathy (DCM) lead innovation for using MRI to examine spinal cord pathways. For example, the potential value of advanced MRI towards quantifying altered spinal cord anatomy has been found, albeit in a small number of patients, with chronic WAD^{36; 115} and incomplete SCI.¹¹⁵ Specifically, larger magnitudes of lower extremity muscle fat that seemed to correspond to altered spinal cord anatomy and reductions in the ability to maximally activate plantar flexor torques were identified in 3 patients with chronic WAD. In contrast, this was not found in a participant reporting full recovery.³⁶ Alarmingly, the lower extremity structural changes and volitional weakness in a discrete number of individuals with chronic WAD were comparable to small number of participants with incomplete SCI.¹¹⁵ Larger scaled prospective studies probing the integrity of white matter pathways with spinal cord imaging applications and quantitative measurements are warranted before stronger conclusions can be drawn. Certainly, tract-specific volume losses identified with high-resolution MRI in patients with DCM have been demonstrated⁶¹ and specific regional changes of the anterior spinal cord are associated with clinical outcomes.¹⁷

Spinal cord imaging applications for WAD, DCM, and SCI

Seminal work by Wolff and Balaban furthered MRI’s sensitivity to tissue composition.¹³¹ This work, based on cross-relaxation and saturation transfer methods, originated in single voxel spectroscopy (SVS). Protons associated with free water (FIGURE 3; blue line) are distinguished from another broad-spectrum proton pool of other hydrophilic macromolecular surfaces (eg, proteins, lipids: FIGURE 3; red line). These two proton pools mix constantly, resulting in cross-relaxation^{29; 132} and probing of the bound and free water pools. By saturating the broad resonance with a powerful off-resonance (1–2 kHz) pulse (green arrow in FIGURE 3), the bound protons lose their magnetization and exchange with the free water pool resulting in macroscopic saturation exchange between the two (FIGURE 3; black dashed line). This magnetization transfer (MT) effect has demonstrated sensitivity to the macromolecular concentration and surface chemistry,^{18; 131; 132} and can be measured relative to a baseline condition (FIGURE 3, black dashed line relative to the blue line). This provides a semi-quantitative map of neuronal integrity of white matter pathways in the spinal cord. While this demonstrates potential prognostic value in traumatic neck pain,^36,115 larger-scaled longitudinal work is required for translation to standard diagnostic practice.

Figure 3 –

Example of Magnetization Transfer Contrast

Ongoing research aims to standardize and automate MT measures throughout the cervical spinal cord, and to apply these methods to a large prospective dataset of individuals with WAD (ClinicalTrials.gov Identifier: NCT02157038). This approach may assist in and contribute to early prognosis of individuals at risk of poor functional recovery, and in particular, patients exposed to and presumably injured from a non-catastrophic MVC (eg, WAD), where evidence of salient tissue damage is rarely observed with conventional scans.

DCM is characterized as a narrowing of the cervical spinal canal leading to neurological impairment and spinal pain presumably due to damage of the spinal cord.^{96; 119} In individuals with DCM, abnormal levels of spinal cord metabolites were shown with magnetic resonance spectroscopy compared to matched controls.^{60; 107} One study revealed elevated lactate peaks, indicative of DCM-related spinal cord ischemia.[60] Abnormal spinal cord metabolites were associated with the modified Japanese Orthopedic Association score, highlighting a radiographical biomarker for the assessment (and informed management) of patients with advanced DCM.^{60; 107}

Diffusion tensor imaging (DTI),^{68; 130} MT, and semi-automated atlas-based analyses are 3 other approaches permitting more in-depth investigation of the spinal cord and its pathways in DCM (FIGURE 4). In DCM, the cervical spinal cord tends to have reduced total density of tracts,¹³⁰ yet elevated tract density at the site of cord compression.³⁰ Spinal cord DTI holds potential as a prognostic/diagnostic imaging biomarker. Magnetization transfer ratios have also shown diagnostic promise in identifying cord abnormalities in individuals with DCM. While a trend towards decreased MT ratios was found at the cervical spinal cord cross section compared to controls, significant differences were revealed when analyzing the anterior portion of the cord.¹⁷ Decreased MT ratios were correlated with both clinical outcome scores^{17; 118} and a quantitative measure of upper extremity hyperreflexia.¹⁷

Figure 4-.

Example structural, diffusion, magnetization transfer, and functional axial cervical spinal cord images are shown. The structural image was acquired using a multi-echo gradient-echo sequence, which provides high white matter to gray matter contrast. In the diffusion image, the principle direction of diffusion in the spinal cord is through the axial plane as indicated by the colored lines. Magnetization transfer imaging can be used to calculate the magnetization transfer ratio (MTR), which provides a measure of tissue macromolecule content. The functional images show group average activation from an acute thermal pain stimulus applied to the right ventral forearm and group average connectivity to the C7 right anterior horn (light blue). Green = spinal cord gray matter.

Emerging evidence suggests spinal cord MRI data can be examined for structural abnormalities in specific gray matter regions and in the ascending and descending white matter tracts. As an example, reductions in spinal cord tract volumes and diameters were predictive of worsening clinical outcomes in individuals with DCM.^{62; 63} Atlas-based semi-automated approaches are also being applied, and potentially inform treatment schemes of traumatic SCI.¹¹⁶

Future directions for spinal cord imaging

Although currently pre-clinical, functional MRI (fMRI) shows promise for the study of neural activation. By measuring the blood oxygen level-dependent (BOLD) signal, an indirect measure of intrinsic neural activity can be obtained.

Recently, increased BOLD signal was found in the ipsilateral spinal cord corresponding to upper extremity muscle activation,¹²⁹ while a similar specific BOLD signal fluctuation corresponding to increasing levels of thermal stimulus applied to the upper extremity was also reported.¹²⁸ Early reports have shown promise that fMRI may be sensitive to differences in resting state spinal cord networks in DCM and other conditions afflicting spinal cord physiology, such as fibromyalgia and multiple sclerosis.^{19; 79; 85}

Other exciting pre-clinical advanced MRI approaches include the use of high field strengths (7 Tesla) for mapping the spinal cord,⁸⁶ and using machine-learning algorithms to both automatically segment white and gray matter of the cord⁹⁹ and objectively detect focal and gross pathology.⁵⁴ The future of advanced spinal cord imaging will likely involve combining sequences,⁸² such as DTI, MT, and atlas-based applications in conjunction with higher field strengths and machine-learning technology, to comprehensively characterize the spinal cord across a number of spinal disorders, such as WAD, DCM, and SCI.

BRAIN IMAGING:

Pain is acknowledged to be as much a psychological phenomenon as physiological; wherein, a complex integration of multiple physiological, and cognitive/emotional processes, as well as sociocultural exposures, are shaped by individual context, past experiences, perceived sense of self, and one’s expectation for recovery.¹⁰⁴ Over the last 2 decades, brain neuroimaging has resulted in a shift towards characterizing the neural processing of a patient’s pain experience and how we consider the effect of pain, or the experience thereof,³⁴ on the brain itself.⁸³

fMRI has demonstrated that nociceptive processing and the subjective perception of pain is not encoded by a single brain region but distributed across a network of multiple brain regions, each with specific roles in the sensory and affective dimensions of the pain experience.⁴ To this end, MRI has been leveraged to study brain plasticity in clinical pain conditions towards identifying unique patterns of brain reorganization (ie, structural, diffusion, and functional) in patients with chronic pain; distinct from acute pain (FIGURE 5).⁸⁴

Figure 5-.

Example structural, diffusion, and functional brain images are shown. The axial structural image was acquired using 3D MPRAGE T1-weighted gradient-echo sequence and can provide morphometric properties of the gray matter. The diffusion example shows a 3D tractography map using the right ventral posterolateral nucleus of the thalamus as a seed. The axial functional images show average group activation from an acute thermal pain stimulus applied to the lower back and group average connectivity to the bilateral posterior cingulate cortices (light blue).

Of the spinal pain conditions, chronic LBP has been the most studied using MRI to demonstrate differences in brain responses to experimentally evoked painful mechanical^{74; 80} and thermal⁷ stimuli. Baliki et al studied brain activity related to the spontaneous fluctuations in chronic LBP intensity and identified increased medial prefrontal cortex (mPFC) activity during periods of sustained LBP where the extent of mPFC activity correlated to symptom intensity.⁶

Excitingly, functional and structural MRI has demonstrated the brain structure is more plastic than previously understood. People with chronic LBP are shown to have reduced prefrontal and thalamic gray matter density,⁵ distinct spatial and temporal gray matter reorganization,⁸ and altered functional connectivity between the periaqueductal gray and the ventral medial prefrontal and rostral anterior cingulate cortices (rACC).¹³⁸ Reduced functional connectivity between primary somatosensory cortex and the rest of the brain following provocative exercise has also been demonstrated.⁷⁵ Collectively, these cross-sectional studies demonstrate co-localization of functional and structural brain changes, providing empirical evidence of abnormal cortical pain modulation in chronic LBP.

In longitudinal studies, Vachon-Presseau et al demonstrated that increased white matter and resting state functional connectivity within the corticolimbic circuitry can predict persistence of LBP at 3 years.¹²¹ Seminowicz et al reported that abnormalities in cortical thickness and brain activity in individuals with chronic LBP could be reversed following effective treatment, surgery or facet joint injections in this example.¹¹¹ Determining which treatments are effective, and for whom, needs to be explored and established on a patient-by-patient basis.

Future directions for Brain imaging:

New directives for brain imaging are to determine the pathophysiological relevance of these changes and how they relate to the clinical presentation of both acute and chronic pain; that is, do these changes result from an acute injury or pain onset; do they have prognostic value; and/or do they reflect a vulnerable-phenotype at risk for chronicity, or a resilient-phenotype likely to recover spontaneously, and are the reported functional and structural brain abnormalities across common musculoskeletal conditions reversible with less invasive, non-surgical interventions?

WHAT TO DO WITH ALL OF THIS IMAGING (MUSCLE, SPINAL CORD, BRAIN) DATA?

Here, we will briefly highlight different data types that could be collected (eg, genomics, phenomics, physical activity, past medical history, previous experience with pain, pre-injury emotional status, or psychopathology, such as depression or related mental health disorder, even early life adversity) and how they may interact to affect the health of not only various soft-tissues, but in making the individual more or less resilient (or vulnerable) to a good (or adverse) outcome.^{125; 126}

A growing body of evidence suggests various genetic influences on resilience or vulnerability to persistence of pain or heightened stress.^{9; 10; 77; 78; 88; 89} Relevant to our discussion are the common types of genetic variations called single nucleotide polymorphisms, or “SNPs”; some of which have demonstrated associations with the magnitude and persistence of pain complaints and distress following trauma exposure.^{10; 89} While exciting, establishing the cause-and-effect influence of genetic vulnerabilities based on polymorphisms and how this may/may not relate to findings from advanced imaging applications is complex and more work is needed in this area.

While the reader is cautioned when interpreting such results, it is difficult not to be excited by the potential value in combining quantitative data from radi-omics, phen-omics, gen-omics and, arguably the most common collection of clinical data; patient self-report outcome measures. Self-report variables could include cultural beliefs about the experience and expression of pain,^{11; 39; 95} the medicolegal context within which their experience is being scrutinized,^{15; 98} the amount of social support they receive,^[101] and other stressors along the life-course.¹³ Importantly, the extent to which each of these factors are either resiliencies or vulnerabilities likely varies by context – a factor that may prove protective under one condition may be a vulnerability under different conditions.² The interpretation of radiological abnormalities and the manner in which they are reported to and shared with the patient could also be influential.

Clearly, these situations cannot be ignored and may help provide some rationale behind recovery – or failure to recover. However, when combined with various data types, associations could be explored to help 1) define precision medicine, 2) produce more informed plans of care, and 3) improve our patient’s experience and hopefully, outcome.

CONCLUSION

We present observable and measurable changes in soft aqueous tissues of the spine, spinal cord, and brain that may contribute towards explaining persistent spinal disorders and its sequelae. This clinical commentary is not exhaustive; instead, we provide readers with a snapshot of what we know, what we don’t know, and what we need to know about the imaging of and characterization for common, but enigmatic neuromusculoskeletal pain conditions involving the spine. The long-term intent of our own research and that of our global colleagues in this area of imaging study should remain a continued line of focused interdisciplinary research towards exploring and establishing causal relationships between muscle, spinal cord, and brain morphometry and poor functional recovery for patients with spinal disorders (traumatic and non-traumatic). However, only an exploration of multiple sources of data can rationalize the dynamic pathways underlying recovery on a patient-by-patient basis. We assert that big data-driven identification of meaningful associations within and between disparate biopsychosocial data sources may unlock keys to understanding an individual’s pain experience like never before.

Synopsis:

The development of persistent spinal (traumatic and non-traumatic) pain is common and contributes to high societal and personal costs, globally. There is an acknowledged urgency for new and interdisciplinary approaches to the problem, and soft tissues including skeletal muscles, the spinal cord, and brain are rightly receiving increased attention as important biological contributors. In reaction to recent suspicion of and questioned value for imaging-based findings, this paper serves to recognize the promise that the technological evolution of imaging techniques, and particularly magnetic resonance imaging (MRI), is allowing in characterizing previously less visible morphology. We emphasize the value for quantification and data analysis of several contributors in the biopsychosocial model for understanding spinal pain. Further, we highlight emerging evidence regarding the pathobiology of changes to muscle composition (eg, atrophy, fatty infiltration) as well as advancements in neuro- and musculoskeletal imaging techniques (eg, fat/water imaging, functional MRI, diffusion imaging, magnetization transfer imaging) of these important soft tissues. These non-invasive and objective data sources may complement known prognostic factors of poor recovery, patient self-report, diagnostic tests, and the -omics fields. When combined, advanced ‘big-data’ analyses may assist in identifying associations previously not considered. Our clinical commentary is supported by empirical findings that may orient future efforts towards collaborative conversation and hypotheses-generation, interdisciplinary research, translating across a number of health fields. Our emphasis is that MRI technologies and research are crucial to the advancement of our understanding of the complexities of spinal conditions.