Quantitative identification and segmentation repeatability of thoracic spinal muscle morphology

Abstract Objective MRI derived spinal‐muscle morphology measurements have potential diagnostic, prognostic, and therapeutic applications in spinal health. Muscle morphology in the thoracic spine is an important determinant of kyphosis severity in older adults. However, the literature on quantification of spinal muscles to date has been limited to cervical and lumbar regions. Hence, we aim to propose a method to quantitatively identify regions of interest of thoracic spinal muscle in axial MR images and investigate the repeatability of their measurements. Methods Middle (T4‐T5) and lower (T8‐T9) thoracic levels of six healthy volunteers (age 26 ± 6 years) were imaged in an upright open scanner (0.5T MROpen, Paramed, Genoa, Italy). A descriptive methodology for defining the regions of interest of trapezius, erector spinae, and transversospinalis in axial MR images was developed. The guidelines for segmentation are laid out based on the points of origin and insertion, probable size, shape, and the position of the muscle groups relative to other recognizable anatomical landmarks as seen from typical axial MR images. 2D parameters such as muscle cross‐sectional area (CSA) and muscle position (radius and angle) with respect to the vertebral body centroid were computed and 3D muscle geometries were generated. Intra and inter‐rater segmentation repeatability was assessed with intraclass correlation coefficient (ICC (3,1)) for 2D parameters and with dice coefficient (DC) for 3D parameters. Results Intra and inter‐rater repeatability for 2D and 3D parameters for all muscles was generally good/excellent (average ICC (3,1) = 0.9 with ranges of 0.56‐0.98; average DC = 0.92 with ranges from 0.85‐0.95). Conclusion The guidelines proposed are important for reliable MRI‐based measurements and allow meaningful comparisons of muscle morphometry in the thoracic spine across different studies globally. Good segmentation repeatability suggests we can further investigate the effect of posture and spinal curvature on muscle morphology in the thoracic spine.


| INTRODUCTION
Spinal muscles are vital to provide mechanical stability and for effective functioning of the spine. Muscular weakness, 1-3 reduction in cross-sectional area (CSA) and force generating capability, 3 changes in passive elastic modulus, 4 and fatty infiltration 5 are a few musclerelated factors that may contribute to the onset and development of a number of adult spinal deformities. Thoracic hyperkyphosis affects 20% to 40% of adults over the age of 60, 6 and is usually accompanied by degenerative loss of lumbar lordosis. 7 A study on healthy adults (70-79 years) showed that low density of the paraspinal muscles contributed to kyphosis, beyond the effects of age and osteoporosis. 1 Furthermore, another study indicated that older men (>65 years) with the smallest paraspinal muscle volume had the largest Cobb angle compared to those with the largest paraspinal muscle size. 8 Given the rise in aging population (>60 years) from 841 million in 2013 to more than 2 billion in 2050, 9 the prevalence of adult spinal deformities is increasing. Thus, in vivo assessment of spinal musculature becomes very important in this regard and has implications in diagnostic, prognostic, and therapeutic applications in spinal health.
Computed tomography (CT), 8,[10][11][12][13] and magnetic resonance imaging (MRI) [14][15][16][17][18][19][20][21] have demonstrated utility in investigating spinal muscles. Although CT typically has higher spatial resolution and shorter scan times, patients have higher radiation exposure and images have lower soft tissue contrast as compared to MRI. 22 MRI approaches, however, have shown good visualization of skeletal muscle composition 18,23 along with good reliability for manual segmentation of muscles. 21 In order to improve clinical relevance and make meaningful comparison of MRI data across different studies, quantification of MRI measurements is crucial.
Most quantitative MRI studies on the paraspinal muscles have only assessed cervical [24][25][26][27] or lumbar regions. 3,8,17,18,21 Consequently, the published literature and descriptions available for defining the regions of interest (ROI) of paraspinal muscles have also focused only on the cervical 28 and lumbar 23 regions. The paraspinal muscles in the thoracic spine have been understudied to date. Thoracic spinal musculature, however, has considerable clinical relevance. A recent longitudinal study on older men and women (mean age: 61 years) associated larger Cobb angle with smaller CSA and lower muscle to intramuscular fat ratio of the thoracic spine muscles, particularly those situated nearest to the kyphosis curvature. 13 Difficulties in consistently identifying spinal muscles in the thorax has led to poor repeatability 18 and slow clinical translation. Currently, to the best of our knowledge, there are no standardized measurement techniques for the thoracic musculature. Thus, in order to fill this gap in literature we aim (a) to develop a systematic methodology to identify and quantify thoracic spinal muscle morphology from continuous axial MR images of thoracic levels and (b) to assess the repeatability of its measurements.
Both, spinal extensor and flexor muscles are important in stabilizing the spine and its posture. 29 However, spinal extensor muscle strength is shown to be more important for muscular support of the thoracic spine 30 and is presumed to have the greater clinical significance with respect to thoracic spinal health. 23,31 Hence, this work focuses on identification of two paraspinal extensor muscles-erector spinae (ES) and transversospinalis (TS) and one posterior muscle-trapezius (TZ).

| Image acquisition
The participants were scanned within the 56 cm gap of a 0. 5  signed an informed consent. Two thoracic levels, T4 to T5 (ie, the junction between the upper and the middle thorax), and T8 to T9 (ie, the junction between the middle and lower thorax) were imaged in two separate scans. Images were obtained from a stack of continuous, parallel slices with the middle slice aligned to the center of and parallel to the intervertebral discs ( Figure 1A). The number of slices (typically 9 or 11) in a stack was varied in order to cover the entire length of the two vertebral bodies situated on either side of the disc. Towards utilizing the MR scanner to its full capacity and as a groundwork for future studies, the images across both the thoracic levels (T4-T5 and   T8-T9)   Both the vertebral body and the vertebral canal were segmented as circles with constant radii as shown in Figure 1B. Radius (mm) was measured as the distance between the centroids of the muscle and the vertebral body 25 ( Figure 1B). 3D muscle geometry was generated by interpolation of a series of 2D segmented images ( Figure 1C).

| Data analysis
Segmentation was performed individually by three raters after reviewing guidelines outlined in this article. All the data ((2 levels) × (6 volunteers) × (4 postures) = 48 set of images with each set containing either 9 or 11 slices, avg = 480 slices) were segmented twice by one rater (biomedical engineer, first author), with the second repeated segmentation spaced two weeks apart. One third of the data (16 set of images, avg = 160 slices) were segmented once by two additional raters (neuro-radiologist and a neuroradiology fellow, co-authors).
Each rater was initially trained on four data sets (40 slices). The segmentations were reviewed for quality control by the primary rater and those which appeared to be inconsistent and deviated from the guidelines were re-segmented by corresponding raters. While about 20 out of 40 images were re-segmented for the training data set, there were no resegmentations for the actual test data set. The segmentations from training data set were excluded from repeatability assessments.
Intra-rater repeatability was evaluated for two repetitions of the primary rater, while the inter-rater repeatability was evaluated on one segmentation measure of all three raters. For 2D measures like CSA, radius, and angle, segmentation repeatability for every muscle was assessed using intraclass correlation coefficients ((ICC) (3,1)) computed over all data. ICCs were interpreted as: <0.69 poor, 0.70-0.79 fair, 0.80-0.89 good, and 0.90-1.00 excellent. 33 For 3D measure, dice coefficient (DC) was used as a statistical validation metric to evaluate the spatial overlap accuracy between two ratings. The value of a DC ranges from 0 to 1, with 0 indicating no spatial overlap between two sets and 1 indicating complete overlap between them. 34,35 DC >0.70 is considered to be a good overlap in literature. 36

| Defining the ROI
We developed specific guidelines for segmenting ES, TS and TZ from axial MR images using anatomical descriptions of these muscles from the literature. Typical origin and insertions of the showing the orientation of the parallel slice stack for thoracic levels T4 to T5 (9 slices) and T8 to T9 (11 slices). B, Image analysis measurements of 2D muscle CSA and position (radius and angle). The brown colored circle represents the vertebral body, the white colored circle represents the vertebral canal. C, 3D geometry obtained from series of 2D muscle segmentations. ES, erector spinae; TS, transversospinalis, TZ, trapezius individual muscles comprising the three muscle groups are listed in the Table 1. The individual muscles comprising of ES and TS functional groups are epaxial (developed to form the post and paravertebral muscles) and are not encapsulated by an independent layer of epimysium (a factor that helps to discretely delineate a skeletal muscle 23,37 ). It is therefore challenging to identify and interpret where the individual fascicles originate and insert, and they are not distinguishable from one another within each muscle group. Thus, the ES and TS are individually identified as single region of interest.

| Trapezius
TZ is a flat, triangular muscle that extends over the back of the neck and upper thorax. It is the most superficial muscle that can be located immediately anterior to the subcutaneous fat layer. It can be visualized as a long-bread or a cigar shaped mass on either side with an increasing volume from the T1 to T3 or T4 levels and decreasing caudally from T4 ( Figure 2). The variability in existence and visualization of the TZ in the lower thoracic levels (T8-T12) is high for different individuals, as the vertebral attachment begins to terminate anywhere between levels T8 and T12. 38 Furthermore, the shape and size of the TZ in almost all individuals is not expected to be bilaterally symmetric, especially at levels T8 to T9 ( Figure 3C). In most transverse slice images, the anterior and posterior borders of the TZ can usually be identified as the longer dimensions of the muscle mass.
1 The anterior border is defined by the fascial line (epimysium) separating TZ and rhomboideus (RH) laterally and TZ and ES group more medially for levels T4 to T5 ( Figure 2). For levels T8 to T9, it is the fascial line separating TZ and latissimus dorsi laterally and TZ and ES group more medially (Figure 3).

T A B L E 1
Anatomical attachment sites of trapezius (TZ), erector spinae (ES), and transversospinalis (TS) muscle groups   Figure 2C) or with TZ more medially and with rhomboideus major laterally (left ES in Figure 2C and both ES in Figure 2E). At lower thoracic levels (T8-T9), however, the anat-

| Technical considerations and segmentation criteria
1 The ROI is segmented as a smooth, continuous island.  (Table 2). For radius and T A B L E 2 Intra and inter-rater repeatability of cross-sectional area (CSA), radius and angle with three raters using intraclass correlation coefficients, ICC (3,1) trapezius (TZ), erector spinae (ES), and transversospinalis (TS)

| 3-D measurements
The average intra-rater Dice coefficient of volume for TZ, ES, and TS

| DISCUSSION
We developed a systematic methodology to identify and quantify thoracic der at all along T4 to T5 (right TZ in Figure 2C). 2 The chances of misidentifying the medial part of latissimus dorsi as TZ, especially caudally along T9 (where TZ has already terminated) is very high as shown in Figure 4B,C. Craniocaudal segmentation is thus recommended to visualize the termination of TZ midway along T8 to T9. When confusion persists in identifying the region of interest, it must be recalled that the volume of TZ always decreases going from level T4/T5 to T12 and hence the region of interest represented by dotted border in Figure 4E cannot be correct.

| Erector spinae
1 The spinalis, the most medial muscle in the ES group tends to diminish in size as one moves inferiorly, as this muscle extends only in the thoracic region and not in the lumbar (  Figure 8B). In order to avoid this error, the medial border of ES is initially identified (including spinalis if visible) at superior-most slice and the same border is tracked in subsequent inferior slices ( Figure 8C).
2 At levels T8 to T9, for some individuals, it is sometimes observed that the ES appears to be more laterally placed at the spinous process and along the supraspinous ligament ( Figure 4) as opposed to the ES seen in Figure 3, which has a more medial position.

| Transversospinalis
Sometimes, where the interspinalis muscle and/or the interspinous ligament is evidently distinct with a slightly irregular and brightened edge ( Figure 9B), the lateral contour of these must be followed instead of the spinous process in defining the medial border of TS ( Figure 9D)

| Limitations of the study
Interpretation of the data presented in this study has limitations. The