Skip to main content

Table 1 Representative literature summary of methods in investigations describing paravertebral muscle analysis using magentic resonance imaging (MRI)

From: Manually defining regions of interest when quantifying paravertebral muscles fatty infiltration from axial magnetic resonance imaging: a proposed method for the lumbar spine with anatomical cross-reference

Citation Reliability MRI Sequence Slice Selection Muscles of Interest ROI Selection Fat Detection Measure
Antony et al., 2016 [52] No T2 FRFSE 1 slice per level at IVD L3-S1 MF, ES Semi-Automated interactive Segmentation (intelligent scissors) Semi-Automated User set pixel intensity threshold FCSA =
(CSAFat)/(CSATotal)
Battaglia et al. 2014 [53] 2 raters (n = 25); Qualitative: Intra-rater weighted kappa = 0.71–0.93, Inter-rater kappa = 0.76–0.85; Quantitative: Inter-rater ICC = 0.73–0.90 T1 1 slice per level at IVD L4-5, L5-S1 MF Qualitative: visual inspection;
Quantitative ImageJ manual trace, exclude fascial boder between MF and ES
Qualitative: Goutallier Classification (0,1,2,3,4); Quantitative: pixel intensity range set by user selected fat ROI within MF % fat =
(# of fat pixels)/(total # of pixels)
Beneck et al., 2012 [12] 1 rater; intra-rater: ICC (3,1) = 0.961 for CSA T1 All slices spanning L4, L5-S1, and S2-S3 vertebral bodies MF, ES Slice-o-matic manual trace Muscle divided into 6–9 regions. Gray-scale signal threshold semi-automatically determined for each region based on user selected small region of muscle Muscle volume =
total volume – fat volume
Bhadresha et al., 2016 [19] 2 raters (n = 20); inter-rater ICC = 0.33–0.76, Cronbach’s alpha = 0.59–0.91 T2 FSE I slice per level at IVD L3-4, L4-5, L5-S1 MF, ES, PS Visual inspection 18×27 (1 cm apart) grid applied to image. By visual inspection: # of grid points touching fat vs. muscle counted Muscle to Fat Ratio=
Muscle/#fat pixels
Crawford et al., 2016 [5] No 2-point DIXON (3D fast-field echo T1) whole body Every 3rd slice L1-L5 (with interpolation to full volume) MF, ES Semiautomatic segmentation with linear interpolation (Myrian Intrasense, Paris, France) Semi-automated ROI from water to fat image Fat signal fraction (FSF) =
(Signalfat/[Signalwater + Signalfat])*100
D’Hooge et al., 2012 [10] No T1 FSE 1 slice per level at L3 superior endplate, L4 superior endplate, L4 inferior endplate MF, ES, PS ImageJ manual trace Automated based on pixel intensity gray-scale threshold MFI =
(signal intensity of “lean muscle”)/(signal intensity of user defined fat ROI)
Fortin et al., 2014 [13] 1 rater; intra-rater: ICC = 0.90 –0.96) T2 1 slice per level at IVD L3-4 and L5-S1 MF, ES, MF + ES ImageJ manual trace following fascial borders User selected pixel intensity gray-scale threshold selected from 4–6 sample ROIs within visible ‘lean muscle’ FCSA =
(CSAFat)/(CSATotal);
Proportion estimate of muscle fat content = Average signal intensity of Total ROI
Hebert et al., 2014 [18] 1 rater; intra-rater (n = 30) ICC (3,1) = 0.93, Bias = −0.70, 95% LOA = −8.11–6.72 T1 SE (0.2 Tesla) 1 slice per level at L4 and L5 MF Manual trace Custom MatLab script; threshold set from midpoint between histogram peaks for ‘fat and muscle’ % IMAT =
CSAFat/CSATotal
(IMAT = intramuscular adipose tissue)
Hu et al., 2011 [54] 3 raters (n = 29); intra-rater ICC (3,1) for FCSA = 0.832–0.847, for T2 SI = 0.926–0.957; inter-rater CC for FCSA = 0.858–0.894, for T2 SI = 0.891–0.923 T2 FSE 1 slice per level at IVD L3-4, L4-5, L5-S1 MF, ES PACS workstation manual trace: lean muscle CSA
Drawn avoiding visible fat;
Fat% - following outer perimeter
PACS embedded ROI and gray-scale histogram software calculated from mean T2 signal intensity Lean muscle FCSA =
Manually traced Muscle CSA
Fat Infiltration =
T2 signal intensity
Kjaer et al., 2007 [8] 2 raters (n = 50); intra-rater k = 0.86;
inter-rater k = 0.58
T1 SE (0.2 Tesla) 1 slice per level at IVD L3-4, L4-5, L5-S1 MF Visual Inspection Visual categorization:
Normal 0–10%, slight 10–50%, severe > 50% fat
Fat infiltration =
(0,1,2)
Paalanne et al., 2011 [28] 2 raters (n = 35);
intra-rater ICC = 0.86–0.88,
inter-rater ICC = 0.85–0.87
T1 FSPGR (In-Phase and Opposed-Phase) 1 slice at upper endplate of L4 MF, ES neaView Radiology; manual trace Average Signal Intensity Relative signal loss =
(IP SI – OP SI)/(IP SI)*100%
Pezolato et al., 2012 [22] 2 raters (n = 10);
intra-rater ICC = 0.90–0.94,
inter-rater ICC = 0.86–0.83
T2 FSE 2 slices per level at upper and lower endplates of L1-L5 MF, MF + ES ImageJ manual trace Grayscale threshold Fat infiltrate =
TotalCSA – FCSA
Ranson et al., 2005 [55] 1 rater (n = 6) × 3:
average intra-rater ICC = 0.97
T2 1 slice per level at lower vertebral endplate of L1-L5; upper vertebral end plate of L5-S1 MF, ES, PS, QL ImageJ manual trace following fascial borders Grey scale pixel intensity range for muscle, fat, and bone were determined from histogram of manual ROIs of “lean paraspinal muscle”, inter-muscular fat, and vertebral body for either: method 1: global grey scale range (0–120); Method 2: slice specific grey scale range FCSA =
TotalCSA of pixels within grey-scale range for fat
Shahidi et al., 2016 [48] No T2 1 slice at L4
(to standardize CSA across individuals)
MF, ES Quantitative manual trace using MatLab Pixels identified as either fat or muscle by fitting a two term Gaussian model to the pixel intensities histogram of from segmented ROIs; identified intersection where pixel values above classified as fat and pixels below classified as muscle. Cross Sectional Area
Fat signal fraction =
npixels fat/npixels fat + npixels muscle
Valentin et al., 2015 [7] 1 rater (n = 24); intra-rater ICC = 0.88–0.99 T1 All slices spanning lower endplate of L5 to upper endplate of L1
(10 mm slice thickness)
MF, ES, PS, RA Analyze Direct
Manual trace
Average SI MFI =
(Muscle SI)/(Subcutaneous Fat SI)
  1. PS Psoas Major, MF Multifidus, ES Erector Spinae, QL Quadratus Lumborum, RA Rectus Abdominus, FCSA Functional Cross Sectional Area, SI Signal Intensity, IP In-Phase (Water), OP Opposed-Phase (Fat), FSPGR Fast-Spoiled Gradient Echo, ROI Region of Interest