In vivo MRI features of spinal muscles in the ovine model

Abstract

Muscle fatty infiltration (MFI) has been identified in patients with spinal pain using magnetic resonance imaging (MRI). Even though sheep are a commonly used animal model for the human spine, comparative sheep MFI data from MRI is not available. Determining MFI in sheep spinal muscles using acquisition protocols commonly used in man will identify the applicability of this approach in future sheep model studies, such that the effects of spinal interventions on muscle can be assessed prior to their use in a human (clinical) population. Therefore, the purpose of this study was to quantify ovine lumbar spine MFI using 3D 2-point Dixon and T1-weighted sequences. T1-weighted and Dixon lumbar spine axial sequences were collected in 14 healthy Austrian mountain sheep using a 1.5T MRI. At each vertebrae, the region of interest of psoas major and minor (PS), multifidus (M) and longissimus (L) were identified. To determine MFI from the T1-weighted images, mean pixel intensity (MPI) was calculated as a percentage of subcutaneous or intermuscular fat. For the Dixon images, fat sequence MPI was calculated as a percentage of the summed fat and water sequence MPIs. Spinal degeneration was graded and correlated to MFI. Dixon MFI was compared to T1-weighted MFI obtained from subcutaneous and intermuscular fat. For every muscle, T1-weighted MFI calculated using subcutaneous fat scored significantly lower than Dixon MFI and T1-weighted MFI calculated using intermuscular fat (p<0.001). There were no significant MFI differences between T1-weighted images calculated using intermuscular fat and Dixon images for M and L (p>0.05), although significant differences were found for PS. In sheep, Dixon sequences provide an acceptable comparison to T1-weighted sequences for lumbar extensor MFI based on intermuscular fat. However, compared to human literature, ovine lumbar musculature contains greater MFI, making interspecies comparisons more complex.

Introduction

Magnetic Resonance Imaging (MRI) is considered the imaging gold standard for quantifying muscle fatty infiltrate (MFI) [1]. Muscle composition has often been evaluated in patients with low back pain [2–6] and neck pain [7–11], as an increase in MFI in spinal muscles can been seen in these patients. While T1-weighted images have been found reliable for MFI quantification in healthy females without neck pain [12], there are several more rapid imaging sequences that can be used to quantify [13] and correlate MFI with histology [1,14,15]. Sequences such as the Dixon method, where MFI data are collected when water and fat are in-and opposed-phase, have been used in healthy controls [16] and in patients with low back pain [17]. It has also been shown to be comparable to MFI obtained from T1-weighted imaging in the cervical multifidus in healthy adults [16].

The ovine is frequently used as an animal model for the investigation of spinal disorders and surgical interventions due to similarities in homo sapiens skeletal anatomy [18,19]. In vivo ovine studies investigating spinal implants are primarily performed to collect in vitro data on implant stability after the implant has been in situ, although other relevant in vivo data are not available [20]. Due to the importance of the muscular system in spinal health [21,22], obtaining in vivo muscle parameters from animal models would be important to determine the validity of translational research between ovine and man. Although the amount of intramuscular fat in the ovine longissimus dorsi muscle has been published previously [23], these data were obtained at slaughter. To our knowledge, MFI in ovine spinal musculature has not been reported or compared to MRI quantification methods used in man. Therefore, the aim of this study was to quantify and compare MFI in the ovine lumbar spine musculature using a T1-weighted and 3D 2-point Dixon sequence, as these sequences are commonly used to quantify muscle fat in man. Additionally, two locations for fat calculation from the T1-weighted sequences were evaluated, as both methods have been used in previously published research [12,24].

Materials and Methods

Study Population

Seventeen healthy male and female Austrian Mountain sheep were included in the study. A range of young and older sheep were included in order to identify the effect of age on MFI. The female sheep underwent an ultrasound investigation to exclude pregnancy. All sheep were assessed by an experienced orthopaedic veterinarian and were deemed medically fit to undergo anaesthesia for the imaging procedure of computed tomography (CT) and MRI. Ethical approval was obtained from the Austrian Federal Ministry of Science and Research (13/10/97/2011) and the guidelines for animal care and use were followed.

Data collection

Latero-lateral radiographs (Computed Radiography, Imaging plate Fuji) of the thoracic and lumbar spine were obtained (70 kV, 2.1 mAs; Super 100 CP, Philips, Eindhoven, Netherlands) in unsedated sheep to identify the presence of any spinal pathology.

Prior to MRI, CT (Siemens Somatom Emotion 16, Erlangen, Germany) was performed to ensure the sheep did not have any metal in their bodies (i.e. ingested foreign objects) that could move when introduce to the magnetic field environment. The sheep were anaesthetised and placed in dorsal recumbency for CT and MRI data collection. Premedication for anaesthesia consisted of intravenous butorphanol (0,1mg/kg) and xylazine (0,15mg/kg). After obvious sedation was achieved, general anaesthesia was induced with ketamine (5 mg/ml) and maintained with inhalation of vaporized sevoflurane. For CT, dorsoventral images (80 kV, 25 mA, tilt 0.0, TI 10.6s, slice thickness 0.6mm, DFOV 153.3 x 153.3 cm) were obtained to determine the number of lumbar vertebrae and identify any potential foreign objects, and transverse images (130 kV, 200 mA, tilt 0.0m TI 1.0s, slice thickness 1.2mm, DFOV 25.0 x 25.0 cm) were obtained to assess the presence of spinal pathology. For MRI data collection, lumbar spine T1-weighted and Dixon sequences were collected using a 1.5 Tesla MRI (Siemens Magnetom Esprit, Erlangen, Germany). T1-weighted images were obtained using a 4mm slice thickness, 2mm gap space, DFOV 450 x 450mm, TR 448 ms, and TE 11 ms. Dixon images were obtained using a 1.5mm slice thickness, 1.5mm gap space, DFOV 450 x 450mm, TR 7.2ms, TE 2.4ms (oppose-phase), and TE 4.8 (in-phase). Axial images of the entire lumbar spine were obtained in up to three acquisitions. It was attempted to obtain the entire lumbar spine in one acquisition, but in some sheep this was not possible due to a risk of temperature increase because of a proportionally large fluid-filled rumen. In these cases, the lumbar spine was imaged in either two acquisitions (cranial and caudal lumbar spine) or in three acquisitions (cranial, middle and caudal lumbar spine). A ventral to dorsal phase direction was used and where this was not possible due to substantial movement artefacts, sequences were run in a left to right phase direction.

At the end of the imaging procedure, sheep were placed in lateral recumbency until they showed active breathing movements, at which time they were extubated and the orogastric tube was removed. Following extubation, sheep were placed in sternal recumbency in a clean stall with shavings and were continuously observed until they were stable at walk and stance.

Data analysis

For each lumbar vertebral body, an axial image closest in location to the mid vertebral body (based on absence of intervertebral discs or endplates and the transverse processes being well demarcated) was selected from the T1-weighted images. Dixon axial images which most closely matched the selected T1-weighted images were then selected. The last lumbar vertebra was defined as L6, regardless of whether sheep had 6 or 7 lumbar vertebrae. The last investigated lumbar vertebra in all sheep was defined as L1, even though in sheep with 7 lumbar vertebrae this would have been L2. From each axial T1-weighted and opposed-phase Dixon image, one assessor (SV) traced the region of interest (ROI) around the left and right sides of psoas major and minor (PS), multifidus (M) and longissimus (L) (Figure 1) using AnalyzeDirect software (V 11.0, Mayo Clinic, Rochester MN). For the Dixon images, the in-phase ROIs were directly applied to the opposed-phase images. For the T1-weighted images, mean pixel intensity (MPI) was obtained from each ROI and the average MPI from the left and right sides of each muscle per vertebral level calculated. A ROI of intermuscular fat between PS and L from the left and right sides, and a ROI of subcutaneous fat from the right side of the body as close to the midline as possible were taken (Figure 2), and the MPI calculated for these ROIs. An axial image from the last lumbar vertebrae where fat was most visible was chosen for these measures. T1-weighted MFI was determined relative to intermuscular fat and to subcutaneous fat by dividing the muscle MPI by that of the subcutaneous or intermuscular fat, and multiplying by 100 to obtain a percentage. To determine fat from the Dixon images, MPI from the fat sequence was divided by the summed fat and water MPI and multiplied by 100 to obtain a percentage.

Figure 1.

Region of interest (ROI) of psoas major and minor (PS), Multifidus (M) and Longisimus (L) on both sides of L4 in (a) T1-weighted (b) Dixon in-phase (c) Dixon oppose-phase axial images

Figure 2.

Region of interest (ROI) of (a) intermuscular fat and (b) subcutaneous fat in an axial T1 weighted image at L6

For intra-rater reliability, the ROI of the investigated muscles at each vertebral level in all sheep was traced a second time by the same assessor for the T1-weighted images and the cross-sectional areas (CSA) compared to the first assessment. For the second assessment, the assessor was blinded to the results from the first assessment.

To quantify pathological changes in the lumbar spine on CT, a grading system similar to that used in the canine species [25] was applied. Grading was performed by two assessors (AE and TL) by evaluation of the radiographs and CT images. Decisions were based on consensus opinion. At each vertebral level of the lumbar spine, degenerative changes were classified as either grade 0 = no pathological changes; 1 = minimal pathological changes i.e. small osteophytes; 2 = moderate pathological changes present i.e. large osteophytes without fusion to adjacent vertebra; 3 = considerable pathological changes i.e. large osteophytes creating a fusion with adjacent vertebra. For each sheep, the highest pathological grade observed was reported.

Statistical Analysis

Statistical analysis was performed with SPSS (IBM, version 22). Normal distribution of the data was assessed using the Shapiro-Wilk test and by investigation of the frequency and Q-Q plots. For each muscle and spinal level, differences between MFI obtained from the intermuscular fat T1-weighted method, the subcutaneous fat T1-weighted method, and the Dixon method were evaluated using a one-way ANOVA with post-hoc Bonferroni correction or Friedmans ANOVA with pair-wise comparisons as the non-parametric equivalent. An independent t-test or Mann-Whitney-U test was used to determine significant differences between male and female sheep for each MFI quantification technique, muscle and spinal level. To determine intra-rater reliability, an intraclass correlation (ICC_3,1) was performed on the first and second CSA ROI measures of all muscles and vertebral levels. Spearman’s correlations were used to investigate the relationship between pathology grade and MFI, and between age and MFI for each MFI quantification technique, muscle and spinal level.

Results

From the 17 sheep, 3 were excluded; 2 due to lack of measurable subcutaneous fat and 1 due to only a part Dixon sequence being available. Of the remaining 14 sheep, eight sheep were male. Mean age and body mass were 2.88 years (± 3.06; range 9 months to 8.5 years) and 66.75 kg (± 20.34) respectively. The majority of sheep (n=9) had 6 lumbar vertebrae. No peri-or post- anaesthesia complications occurred and recovery was uneventful in all sheep except one, which required a tracheotomy after biting off parts of the intubation tube on sudden awakening after anaesthesia. Further treatment for this sheep consisted of daily wound cleaning, antibiotics and anti-inflammatories (carprofen 1.5mg/kg per day for two days and cefquinome 1mg/kg per day for two days). Further recovery of this sheep was uneventful.

Intra-rater reliability for ROI CSA of all muscles at all vertebral levels was high (ICC 0.930 to 0.997; Table 1). Table 2 shows the MFI values for the intermuscular fat T1-weighted method, the subcutaneous fat T1-weighted method, and the Dixon method. For all levels of all muscles, MFI determined using the subcutaneous fat T1-weighted method scored significantly lower than both the Dixon and intermuscular fat T1-weighted methods (all p<0.01). For PS, the Dixon method produced significantly greater MFI than both the intermuscular- and subcutaneous fat T1-weighted methods (all p<0.001), although no significant differences were found between the Dixon method and the intermuscular fat T1-weighted method for any level of M and L. No significant differences were found between male and female sheep for any muscle, vertebral level or MFI quantification method.

Table 1.

Intra-rater intra-class correlations (ICC) for the cross sectional areas from the regions of interest of the muscles psoas major and minor (PS), multifidus (M) and longissimus (L) for L6 to L1. The upper and lower confidence intervals (CI) are shown.

Vertebra	Muscle	ICC	CI
			lower	upper
L6	PS	0.982	0.943	0.994
	M	0.930	0.779	0.978
	L	0.985	0.953	0.995
L5	PS	0.992	0.975	0.998
	M	0.985	0.955	0.995
	L	0.997	0.988	0.999
L4	PS	0.997	0.988	0.999
	M	0.994	0.981	0.998
	L	0.996	0.984	0.999
L3	PS	0.997	0.992	0.999
	M	0.985	0.956	0.995
	L	0.997	0.991	0.999
L2	PS	0.991	0.973	0.997
	M	0.950	0.847	0.984
	L	0.996	0.970	0.999
L1	PS	0.985	0.952	0.995
	M	0.994	0.982	0.998
	L	0.996	0.982	0.999

Table 2.

Muscle Fatty Infiltrate (MFI) of psoas major and minor (PS), multifidus (M) and longissimus (L) for each vertebral level (L6-L1), obtained from T1weighted sequences calculated relative to intermuscular fat (T1 int) and subcutaneous fat (T1 subcut), and from the Dixon sequences. The mean, standard deviations (stdev) and median are shown. Matching superscripts indicate non-significant differences between MFI quantification methods.

Muscle	Vertebra	T1 intermuscular MFI		T1 subcutaneous MFI		Dixon MFI
		Mean (stdev)	median	mean	median	mean	median
PS	L6	37.65 (5.38)	37.15	22.38 (4.85)	22.83	56.40 (2.03)	55.94
	L5	35.79 (7.24)	35.25	21.05 (4.44)	19.46	54.86 (1.73)	54.60
	L4	31.85 (7.13)	31.00	18.63 (3.83)	17.55	54.43 (1.82)	54.16
	L3	30.15 (7.31)	30.95	17.55 (3.60)	17.98	53.97 (1.71)	53.99
	L2	31.58 (7.65)	32.06	18.58 (4.80)	19.17	54.29 (2.03)	54.08
	L1	34.08 (8.18)	36.32	20.14 (5.53)	19.85	56.19 (1.72)	56.52
M	L6	58.97 (12.55)	55.11a	34.67 (7.49)	32.58	58.03 (3.00)	56.84a
	L5	56.83 (14.62)	50.63b	33.18 (7.44)	30.57	57.41 (3.90)	56.29 b
	L4	52.64 (13.18) c	48.61	30.81 (6.90)	29.25	55.98 (3.26) c	55.05
	L3	51.82 (13.35) d	47.50	30.14 (6.20)	28.35	55.83 (3.84) d	54.82
	L2	54.13 (13.90) e	47.27	31.57 (6.84)	29.90	55.91(3.81) e	55.05
	L1	55.54 (13.97) f	52.21	32.47 (7.23)	31.45	57.04 (4.41) f	56.02
L	L6	54.45 (12.88)	50.72 g	31.81 (6.44)	30.46	57.67 (5.54)	55.64 g
	L5	48.42 (8.61)	45.28 h	28.67 (6.39)	26.85	51.93 (2.56)	50.97 h
	L4	45.32 (8.09)	44.76 i	26.88 (6.22)	26.17	51.31 (2.58)	50.23 i
	L3	45.50 (9.64)	42.64 j	26.57 (4.55)	25.56	51.20 (2.38)	50.16 j
	L2	45.88 (9.18)	44.92 k	27.13 (6.39)	26.79	51.15 (2.06)	50.33 k
	L1	48.54 (7.80)	49.54 l	28.96 (6.90)	29.73	51.32 (2.12)	50.73 l

The pathology grade ranged from 0 to 2 (median 0.5). Seven sheep scored grade 0, four sheep scored grade 1, and three sheep scored grade 2. The youngest sheep which scored a grade 2 was three years old. The correlations between MFI and age and MFI and pathology grade are shown in Tables 3 and 4.

Table 3.

Correlations (corr) between age and muscle fatty infiltrate (MFI) using three different MFI quantification methods; T1 intermuscular, T1 subcutaneous, and Dixon. Significant (signf) findings are indicated by an asterisk (*p<0.05, **p<0.01), and the p-values are shown.

Spinal level	Muscle	Correlations of MFI with age
		T1 intermuscular		T1 subcutaneous		Dixon
		corr (r)	signf	corr (r)	signf	corr (r)	signf
L6	P	0.050	0.864	-.607*	0.021	0.624*	0.017
	M	0.635*	0.015	-0.156	0.595	0.429	0.126
	L	0.853**	0.000	0.115	0.696	0.690**	0.006
L5	P	0.147	0.617	-0.610*	0.021	0.296	0.305
	M	0.612*	0.020	-0.021	0.944	0.472	0.088
	L	0.415	0.140	-0.410	0.145	0.481	0.082
L4	P	0.089	0.761	-0.633*	0.015	0.282	0.329
	M	0.328	0.252	-0.330	0.249	0.424	0.131
	L	0.216	0.459	-0.596*	0.024	0.608*	0.021
L3	P	0.151	0.606	-0.722**	0.004	0.388	0.171
	M	0.408	0.147	-0.183	0.530	0.268	0.354
	L	0.209	0.474	-0.596*	0.024	0.489	0.076
L2	P	-0.053	0.858	-0.706**	0.005	0.342	0.232
	M	0.374	0.188	-0.021	0.944	0.275	0.341
	L	0.131	0.656	-0.578*	0.030	0.731**	0.003
L1	P	-0.199	0.494	-0.713**	0.004	0.000	1.000
	M	0.305	0.289	-0.165	0.573	0.362	0.203
	L	-0.197	0.499	-0.702**	0.005	0.527	0.053

Table 4.

Correlations (corr) between pathology grade and muscle fatty infiltrate (MFI) using three different MFI quantification methods; T1 intermuscular, T1 subcutaneous, and Dixon. Significant (signf) findings are indicated by an asterisk (*p<0.05, **p<0.01), and the p-values are shown.

Spinal level	Muscle	Correlations of MFI with age
		T1 intermuscular		T1 subcutaneous		Dixon
		corr (r)	signf	corr (r)	signf	corr (r)	signf
L6	P	-0.060	0.839	-0.593*	0.025	0.397	0.160
	M	0.444	0.111	-0.277	0.337	0.494	0.073
	L	0.702**	0.005	-0.043	0.884	0.573*	0.032
L5	P	0.060	0.839	-0.702**	0.005	0.253	0.382
	M	0.148	0.613	-0.425	0.130	0.444	0.111
	L	0.148	0.613	-0.552*	0.041	0.263	0.363
L4	P	0.201	0.491	-0.485	0.079	0.079	0.789
	M	0.210	0.471	-0.409	0.147	0.375	0.186
	L	0.306	0.288	-0.466	0.093	0.401	0.155
L3	P	0.158	0.590	-0.597*	0.024	0.229	0.430
	M	0.220	0.450	-0.327	0.253	0.375	0.186
	L	0.253	0.382	-0.440	0.116	0.306	0.287
L2	P	-0.189	0.518	-0.693**	0.006	0.229	0.430
	M	0.184	0.529	-0.320	0.264	0.392	0.166
	L	-0.007	0.981	-0.683**	0.007	0.640*	0.014
L1	P	-0.100	0.733	-0.518	0.058	-0.124	0.672
	M	0.299	0.300	-0.222	0.445	0.444	0.111
	L	-0.014	0.961	-0.545*	0.044	0.349	0.222

Discussion

The presence of increased spinal muscle fat in patients with back and neck pain has been identified using MRI [4,5,7,10]. Even though the ovine is commonly used for experimental spinal investigations [19], ovine MFI had not been investigated using similar outcomes reported in man. This study has quantified MFI in ovine lumbar spinal muscles using two different, but widely available, MRI techniques. The results showed that MFI outcomes from Dixon and T1-weighted sequences are comparable for the dorsal muscles M and L, if intermuscular fat rather than subcutaneous fat is used for T1-weighted images. In the human cervical spine, multifidus has also shown comparable muscle-fat findings between T1-weighted and Dixon sequences [16], where mean signal intensity of C2 intermuscular fat was used. The reason why PS was significantly different between all methods compared in the present study whereas M and L were similar between the Dixon and T1-weighted intermuscular fat method, may be due to differences in muscle morphology such as fibre typing or internal structure complexity, although this would require further investigation using a histological approach. In T1-weighted images, fatty intramuscular septa are clearly identified but smaller proportions of homogeneous intramuscular fat cannot be defined [28] and may even be invisible [13].

The subcutaneous fat T1-weighted method to determine MFI resulted in significantly lower MFI outcomes compared to the Dixon method for all muscles investigated. Although it is not known whether histological differences in fat composition exist between the intermuscular and subcutaneous fat locations used in the present study, it is most likely that the differences between methods found occurred due to B0 inhomogeneity, which can influence signal intensities in MRI [26] and is more pronounced further away from the isocentre of the magnet [27]. Further, a gradient echo approach can experience smearing of signals from tissues with high fat content to adjacent musculature with low fat content, as it is sensitive to susceptibility effects [28]. These factors may also explain the differences seen in the present study for the correlations between MFI and pathology grade using the different MFI quantification methods. Both the intermuscular fat T1-weighted and Dixon method produced positive correlations with pathology grade and MFI, although the subcutaneous fat T1-weighted method produced only negative correlations. Previous studies have also shown positive associations in spinal pathology and muscle fat [29,30]. Therefore, using subcutaneous fat as a measure to determine MFI in sheep may not be appropriate based on the findings of the present study due to the potential limitations described above. Similarly, the negative associations found between age and MFI for the T1 weighted subcutaneous fat method should be interpreted with caution, as ageing has been reported to either have no influence on MFI of spinal muscles [31] or increase in some but not all spinal muscles [24]. For this reason, the negative correlations for the T1 weighted subcutaneous fat method should be interpreted with caution and the site of fat used to determine MFI in MRI studies should be considered carefully.

The quantification of MFI from MRI in the ovine lumbar spine has not been investigated therefore the results from the present study cannot be directly compared. However, based on near-infrared spectroscopy on samples of 100g homogenous muscle after slaughter [32], thoracolumbar longissimus dorsi fat content in Merino cross breed sheep has been determined [33,34]. The studies showed that intramuscular fat ranged from 3.82% to 4.84% across the breeds and age groups investigated. This is considerably lower than the findings from the present study and this may be due to structural and fibre type differences between thoracic longissmus and the longissimus, or simply due to differences in breed and methodological approaches. To resolve this would require histological analyses.

CT is another imaging technique that can be used to identify fat content. In sheep carcasses, Gardner et al [35] used CT to determine whole carcass fat, although the fat content within muscle specifically was not reported. Others have investigated the relationship between muscle fat from infrared spectroscopy and pixel density on CT images from sheep carcasses [36] but only a moderate inverse correlation was found for the longissmus lumborum muscle. Lambe et al [37] obtained CT images in lambs in vivo, but this was also to determine internal fat rather than muscle fat content. Further in vivo CT imaging of sheep carried out by these authors revealed that muscle density determined by Hounsfield Units (HU) from CT was significantly associated with meat palatability [38]. This shows that CT may be a suitable tool for the assessment of fat content in muscle however the CT studies reported here used diverging HU boundaries for muscle and fat. Until it can be shown which fat and muscle HUs on CT corresponds to muscle fat from histology, absolute sheep muscle fat determined from CT may be questionable. Furthermore, the boundary cut-offs for muscle, fat and bone are influenced by the voltage used in CT acquisition [39]. The tube potential used in several sheep CT studies used higher values than that used in the present study.

In comparison to the existing literature on human spinal muscle MFI from multiple echo images, the apparently healthy ovine lumbar multifidus contains considerably more MFI. For example, multifidus MFI obtained from Dixon sequences in the present study ranged from 55 to 58% depending on vertebral level. On the contrary, cervical multifidus MFI from healthy human participants has been shown to range from 15 to 24% from Dixon sequences [16]. Direct comparisons to the human lumbar multifidus is limited due to different methodological and analytical approaches in other studies, for example categorical MFI grading by visual inspection [4] or by reporting differences between in-phase and opposed-phase signal intensities [5,17]. No other studies were identified which investigated lumbar MFI in humans using multi-echo images, however using 1H proton MR spectroscopy, human lumbar multifidus was shown to contain 15% MFI in healthy participants and significantly greater MFI (24%) in patients with low back pain [3]. These values were reported as 26% and 29% in the longissmus muscle. Therefore, spinal muscle MFI in healthy sheep appears to exceed that of MFI found in healthy participants and even in patients with back pain, based on MRI findings. The implications of these assumptions might suggest that a relative smaller proportion of functional muscle tissue may be available in the ovine compared to man. This is less likely to be due to pathology and probably due to the genetics underlying meat production, as intramuscular fat is a positive factor for meat palatability [34]. Fat proportion differences may however also have biomechanical implications, which at present have not been investigated in the ovine in-vivo. Muscle volume has been shown to have an influence on spinal stiffness in vitro [40], however a greater relative amount of fat in a given muscle volume may negatively influence muscle functionality [8].

Regardless of the differences between the human and sheep spine and spinal muscle morphology, sheep are commonly used as an animal model for the investigation of new spinal interventions prior to them being used in a clinical (human) population. Unfortunately, data from the ovine model are generally obtained ex vivo. This study has shown that MFI of the paraspinal muscles can successfully be determined from sheep in vivo using similar MRI acquisition methods as those used in man. The benefits of future ovine model studies using such an approach are numerous: The paraspinal muscles are widely investigated in patients with low back pain, however muscles in animal models have received very little attention, probably because of the ex vivo nature of animal model studies. Therefore, using the same MRI acquisition methods commonly used in man for animal studies will allow comparable (in vivo) data to be obtained. This should provide a better insight into the possible changes in muscle tissue which may be seen as a result of spinal interventions tested pre-clinically. This would allow more rigorous testing and potentially provide better predictors on clinical efficacy of interventions. Furthermore, careful recovery based studies using animals can be utilised over several time points, which could provide for a better indicator of the clinical course of changes observed in paraspinal musculature.

There are several limitations in the present study which are to be acknowledged. First of all, for 2 out of the 14 sheep, sequences were obtained in a left to right phase direction due to considerable movement artefacts seen in the anterior to posterior phase direction. This may have had an effect on signal intensity causing left to right side differences, but as the mean pixel intensity of the two sides were taken, it was anticipated this effect was minimal. In illustration, when the MFI data of all individual sheep for each vertebral level and muscle were ranked by increasing value, the mean rank of these two sheep were 7 and 10 out of 14 for the T1-weighted images, therefore placing them in the middle of the data set and not raising cause for concern. For the Dixon sequences these findings were 2 and 5 respectively. Further, due to the large proportion of rumen fluid to body tissues, movement artefacts were seen in most sheep although the quality of the images obtained was deemed sufficient for data analysis. Additionally, the Dixon approach is advantageous in cases where inhomogeneity artefacts may be a problem [41]. There was also a difference in slice thickness between the T1-weighted and Dixon sequences, and while this effect is thought to be minimal, caution is warranted when interpreting such data as they could contain partial volumes. Accordingly, image registration between T1 and Dixon images was not possible but the multi-echo Dixons were co-registered. Attempts to co-register the T1 and Dixon images should be considered in future studies. In order to minimise bias, the single assessor in this study conducted the ROI twice with high intra-rater reliability. Future studies should include multiple assessors of sheep muscle ROI to determine inter-rater reliability and enhance the robustness of the study findings. Previous work has shown acceptable inter-rater reliability of lumbar muscle ROI from T1-weighted MRI images in man, particularly for the muscles PS and L [42], therefore it is anticipated that similar reliability outcomes would be observed in sheep using these same MRI evaluation methods.

Future ovine studies should compare muscle fat findings obtained from MRI with fat values determined using muscle histology in the same sheep. Unfortunately this approach was out-with the scope of the present study, and to our knowledge, a comparison of MRI- and histology-determined fat has, as yet, not been reported. However, this method has been carried out in a rabbit and pig model, which showed good agreement between MFI from Dixon sequences both at 1.5T and 3.0T and histology [15]. An additional aspect that should be quantified in similar future sheep studies is the composition of the fat (i.e. brown or white adipose tissue) from the fat sample sites. A significant difference in signal intensity between brown and white fat has been shown in man using chemical-shift water-fat MRI [43], therefore differences in the quantity of brown and white fat may be a possible source of error in muscle fat quantification. This may also explain the differences in muscle fat values calculated relative to intermuscular and subcutaneous fat. However, it has been shown that brown fat in sheep declines from birth and at the age of 30 days, only white fat is discernible on histology of perirenal-abdominal adipose tissue [44], therefore it is anticipated that this effect on signal intensity on the sheep in the present study is minimal.

Lastly, investigating different sheep breeds and sheep with higher grades of spinal pathology would also determine which sheep would be most suited to represent human muscle morphometric parameters of MFI and muscle volume. Gaining such information would be an important contribution to determining the appropriateness of the ovine as a human spinal model, as debate remains whether such animal models are appropriate in translational spinal research [45].

In conclusion, the Dixon multi-echo fat/water sequence provides an acceptable comparison to T1-weighted sequences in the ovine lumbar multifidus and longissimus muscles if T1-weighted MFI is determined relative to intermuscular fat rather than subcutaneous fat. This is not the case for the psoas muscle, where both fat sources used to determine MFI from T1-weighted yielded significantly lower MFI values compared to the Dixon method. It was evident that MRI-determined MFI of the ovine lumbar muscles found in this study is greater than MFI in human lumbar muscles reported in published literature. This may have implications for sheep to human spine translational outcomes, due to the reduced biomechanical efficacy of muscle with increased MFI. Future studies should compare ovine MRI-obtained MFI with the gold-standard biopsy.