Towards defining muscular regions of interest from axial magnetic resonance imaging with anatomical cross-reference: part II - cervical spine musculature

In order to facilitate widespread adoption of agreed and time-efficient techniques for measuring cervical spine muscle quality, a standardised, reliable, and replicable method is urgently required. While there is a general trend toward optimising automated methodologies that quantify muscle composition based on differential tissue signal intensities of paravertebral muscle, even the latest, time-efficient tools require a degree of manual input for defining regions of interest (ROI) [3, 5, 6, 38‐41]. A standardised ROI method is arguably most important for these studies where it has been speculated that difficulties identifying morphology of both the cervical and lumbar musculature results in poorer repeatability [6, 38]. With continued improvements in both the uptake of, and imaging quality from, MRI technology, an agreed analysis plan utilising a common research measurement method for the identification of ROIs could result in meaningful comparisons with a target towards knowledge transfer and clinical translation of muscle imaging. Following on from the recent manuscript detailing a method for determining ROI in the lumbar spine [42], the purpose of this proposed method is to provide a standardised MRI procedure for measuring cervical spine muscle composition. The method also serves to initiate and continue discussion on the analysis of skeletal muscle composition amongst and between the global clinical and scientific communities.

A number of conventional MRI applications (T1, T2, proton-density, Gradient Echo) are available and have been used to qualitatively and quantitatively measure the water and fat species of healthy and diseased soft-aqueous skeletal muscle tissue [1, 3, 41, 43‐49]. Technological advancements have also produced alternatives that can be used to image muscle, such as dual acquisition methods, where frequency is selectively excited to produce a water image [50] and a standard image of fat and water. This, however, produces a challenge when measuring a redundant and anatomically complex set of multi-layered (and small) muscles in the cervical spine. The challenge is further compounded by the advent of higher field scanners (e.g. 3–7 Tesla), where a uniform frequency difference between fat and water content may be difficult, but certainly not impossible, to achieve.

Despite recent technological advances that have permitted further insight into muscle composition, the mechanisms underlying muscle degeneration and their influence on outcomes in neck disorders remain elusive. In addition, the vast majority of symptomatic and asymptomatic population-based studies examining pathoanatomical features (e.g. the intervertebral disc, ligaments, and the skeletal vertebral column) of the cervical spine have used a variety of conventional MRI sequences [1, 2, 12, 13, 26, 30, 34, 51‐55]. Despite the large repository of available works, the data derived from these imaging investigations have not revealed a consistent structural lesion(s), or response to said lesion(s), that have clarified the clinical presentation of traumatic or non-traumatic neck disorders. This has, in our opinion, created a clinical (and research) impasse that we believe is due partly to the heterogeneous methods across a number of high quality studies investigating the usefulness of imaging for understanding spinal pathology. Ultimately, the clinical value of imaging findings of spinal pathology and/or muscle degeneration will be realised if such findings predict important outcomes or help to identify patients likely to respond to specific interventions (e.g. spinal phenotypes).

Research efforts that focus on the consistent assessment of spinal muscle quality with MRI may improve our collective biological understanding of traumatic and non-traumatic neck disorders and why some, but not others, recover spontaneously. Accordingly, a robust and easily-replicated platform for acquiring, assessing, measuring, analysing, and interpreting imaging data on muscle composition and morphology is needed. Currently a wide variety of methods are used to describe the composition and morphology of cervical spine muscles (see Table 1 for a non-exhaustive summary). This represents a key challenge for both producing consistent regions of interest of cervical spine muscles and allowing comparison between research studies.

The muscles spanning the mid-to-lower cervical spine that are typically examined include: multifidus, semispinalis cervicis, semispinalis capitis, splenius capitis, scalenes, levator scapulae, sternocleidomastoid, and longus capitis and longus colli. We do not describe muscles of the suboccipital region (rectus capitis posterior major and minor, and the superior and inferior obliquus muscles [56]) as it is not possible to accurately measure a clinically useful ROI of the suboccipital muscles from the typically employed transverse images used for assessing cervical musculature. This is because no suboccipital muscle has a long axis close to perpendicular to the transverse plane, thus making measurement of useful cross-sectional ROI impractical. Further, fan shaped muscles such as both rectus capitii muscles require special consideration in order to validate useful measures, given a single cross-sectional measurement along the length of either muscle would pose difficulty for determining whole muscle volume. Future work should include developing imaging protocols for the suboccipital muscles as they require more nuanced imaging methods and measures with careful consideration around the highest resolution possible within a reasonable scan time.

The anatomical study we use and recommend for reference are those detailed in Au et al. [57]. They have provided a comprehensive series of labelled axial MR images from one individual to serve as a reference atlas of the cervical spine musculature to guide clinicians and researchers in the accurate identification of these muscles on MR imaging. We have further reinforced by cross-referencing with the E-12 plastinates that have previously been used to assist morphological studies [42, 58].

Similar to that reported for the lumbar spine, [42] a standard scout image from the sagittal localiser or conventional T2-weighted scan can be used to cross-reference and discern cervical level from axial MR. Users will also find it useful to scroll between the adjacent axial slices to accurately landmark anatomical structures when producing ROIs. The method is applicable to studies examining paravertebral ROIs for single (cross-sectional) or multiple (volumetric) slices. Previous work from the lumbar spine suggests a randomised approach for starting with either the left or right side, and/or separate muscles can influence repeatability when creating ROIs [38, 66]. The same randomised approach is suggested for the cervical spine.

Definitions for ROI measures from MRI for the multifidus, semispinalis cervicis, semispinalis capitis, longissimus capitis, splenius capitis and cervicis, levator scapulae, longus colli, longus capitis, scalenus and sternocleidomastoid are included, describing the anatomical borders (cross referenced to Figs. 1, 2, 3). ROI definitions are detailed with particular reference to cervical levels C2/3, C5/6, and C7/T1. Technical notes are also provided where identifying the guided ROI on MRI may be difficult.

It is worth noting that an anatomical plane that passes laterally and posteriorly in an arc from the anterior aspect of the vertebral body presents a reliable reference point for identifying the anterior aspect of all anterior muscles apart from the sternocleidomastoid (Fig. 2).

The type, quality, and output of images acquired from MR scans are highly influenced by many factors including, but not limited to, user-prescribed parameters. Similar to our previous paper covering the lumbar spine, [42] we endorse consistency in the adoption of MR imaging parameters to facilitate standardised operational procedures that allow intra-study/−institutional comparison and future pooling of results for meta-analyses.

The parameters listed here are based on those widely utilised in literature (refer to Table 1), and are adapted from those published in a previous paper on ROI for lumbar spine muscles [42]. The parenthetical values given with each parameter are not definitive or unique to a cervical spine study; rather they are displayed as an example of the consistent reporting style we propose. At a minimum, we believe the following information should be reported in all submitted manuscripts: Field strength (e.g. 3 Tesla); sequence type (e.g. 2-point DIXON (3D fast-field echo T1) whole body); repetition time (e.g. TR 4.2 ms); echo time (e.g. TE 1.2 and 3.1 ms); flip angle (e.g. 5°); field of view (e.g. FOV 560 × 352 mm); acquired voxel dimensions (e.g. 2.0 × 2.0 × 4.0 mm); reconstructed voxel dimensions (e.g. 1.0 × 1.0 × 2.0 mm); bandwidth (e.g. 240 Hz/Px), acquisition time (e.g. TA 5 min 22 s) and slice thickness (e.g. 4.0 mm). Additionally, the description should include axial slice alignment (e.g. aligned parallel to C2–3 intervertebral disc), slice selection (e.g. measurements taken at most cephalad slice per vertebral level), and subject body position including any support materials that may influence cervical spine posture/curvature (e.g. subjects positioned supine with arms at sides and 2 in. foam cushion under head).

Measures of muscle size are frequently reported in MRI and other imaging-based studies (e.g. ultrasound). In both the lumbar and cervical regions, methods employing a single cross-sectional MR slice are time efficient for determining muscle size and fat proportion within an ROI. However, a CSA measure from a single-slice should not be taken to constitute a whole muscle size or fat measure [15, 68]. Accordingly, volumetric measures, may be more appropriate [15, 69, 70]. We therefore recommend a multi-slice approach that derives muscle size and fat content based on a three-dimensional volume across the levels of interest. In going forward, such measures should be accurately categorised as a 3-dimensional volume of the entire muscle as 3D acquisition methods with MRI have evolved and are not as sensitive to the radio frequency slice profile as is 2D imaging [15].

It is of course acknowledged that acquiring such data with both semi-automated or automated programmes for both the lumbar [42] and cervical spines is time-consuming. However, with the evolution of higher-resolution imaging techniques a more time-efficient capture of cervical muscle volumes from a single vertebral level may correspond to a representative marker of MFI across the entire cervical column. While this has been demonstrated in the healthy lumbar spine [29] where the fat content at L4 best represents that of the entire lumbar region, future research should continue to systematically include the entire cervical spine in healthy and symptomatic cohorts to build a stronger body of evidence regarding age-aggregated cervical paravertebral muscle composition.

Another issue with longitudinal designs, where muscle measures are produced over time, remains a general lack of reporting on how the MRI slices are aligned in plane. A failure to do so could potentially result in registration discrepancies depending how the angle through each muscle was performed. Using some standard anatomical reference (e.g. vertebral bone) that is not expected to appreciably change over time could control for this. Errors of this type can be further minimised by reporting muscle volume over the full length of the muscle (from origin to insertion), as suggested above, rather than a single-slice CSA.

The demonstration of neck muscle fatty infiltrates on T1- weighted imaging in acute [2, 3] and chronic traumatic neck pain [1, 5, 8, 30] has been reported in cross-sectional and longitudinal fashion and across three countries (Australia, [2] Sweden, [5] and the United States [3, 6]). Such findings are not present to the same magnitude for those with chronic idiopathic neck pain [7] and it has been postulated that these muscle changes represent one neurophysiologic basis for the transition to chronic pain in this population [71]. A variety of newer and more rapid high resolution MRI techniques (3D Fat/Water Separation and Proton-Density Fat Fraction, Fat suppression) [65, 72‐77] and analyses (FCSA/CSA, Fat Signal Fraction, MFI %) could help better visualise and quantify physiologic changes at the level of the muscle cell or other disease processes when compared to other conventional clinical imaging sequences (e.g. T1- and T2-weighted). However, such variety across methods and techniques also complicates comparison among studies. Accordingly, we call for all authors to clearly detail their fat infiltration measurements to ensure that future pooling of data efforts is possible. Further, with the number of proprietary semi-automated or automated methods appearing in the literature, and of which descriptions are limited due to commercial sensitivity, we contend it will be helpful for authors to include enough technical detail for comparisons to the fundamental literature to be made.

It is our recommendation that participants should lie supine inside the magnet with a foam pad under their knees and foam padding placed on the right and left of the head to minimise head movement. A neutral position, visually determined by ensuring that a horizontal position of the forehead to the chin is parallel to the MRI table, is also recommended.