Population
Thirty-seven patients affected by LBP were enrolled prospectively in this study, in which the primary outcome was to study contrast diffusion in intervertebral discs [17]. The selected population included male and female subjects, with an age of 42.5 ± 9.1 years (mean ± standard deviation) and an age range from 18 to 60 years. Each patient received detailed information regarding the study protocol and gave her/his consent. Exclusion criteria were: age under 18 or over 60 years, contrast agent allergy, reduced renal function, and contraindications to MRI. The study was approved by the local ethical committee.
Radiological evaluation
MRI of the lumbar spine was performed with a 1.5-T scanner (Avanto; Siemens, Erlangen, Germany) with a phased-array back coil. Standard examinations included routine sagittal and axial T1-weighted (repetition time = 500 ms, echo time = 13 ms) and T2-weighted (repetition time = 4180 ms, echo time = 104 ms) turbo spin-echo sequences as well as axial T2-weighted sequences. In addition, ProHance® (gadoteridol; Bracco Diagnostics, Princeton, RI, USA), a paramagnetic macrocyclic non-ionic contrast agent, was injected at a dose of 0.2 mmol/kg and a second T1-weighted image was taken approximately 5 minutes after contrast injection. A higher dose with respect to standard clinical applications was used following the indications of a previous study about diffusion in the intervertebral disc [18].
A musculoskeletal radiologist with more than 30 years of experience noted the presence and the type of MCs in the endplates from T12–L1 to L5–S1. Only the MCs with a vertical height of more than 5 mm were considered for the semi-quantitative evaluation, but the presence of smaller MCs was noted [13].
Semi-quantitative measure of Modic changes
From the scans showing at least one MC, the unenhanced slice where each MC had the greatest depth was selected; the same slice was selected for the contrast-enhanced series. Unenhanced and contrast-enhanced T1-weighted scans were co-registered to ensure alignment using Elastix, a registration toolkit based on the National Library of Medicine Insight Segmentation and Registration Toolkit (ITK) [19]. To this aim, two-dimensional affine registrations (six degrees of freedom) were performed.
In order to obtain semi-quantitative data, software allowing for manual selection of a polygonal region of interest (ROI) on the unenhanced image (Figs. 1 and 2) and pixel-based calculation of the signal intensity of the selected area was developed in Matlab® (MathWorks, Natick, MA, USA).
For each MC under consideration, the developed protocol asked the operator to select a ROI corresponding to the zone of altered intensity on the unenhanced image and two ROIs as controls, because a reference for the “healthy” bone marrow in at least one control ROI was necessary. The ideal control ROI would be close enough to the MC to minimise the influence of local field fluctuation [20], but also wide enough and free of signal alteration, which in some cases would be possible only in another site. These considerations lead to an investigation of two different kinds of control ROI:
-
1.
a ROI in the same vertebra affected with the MC (same vertebra [SV]);
-
2.
a ROI corresponding to the section of the closest upper vertebra without MC—in this case one ROI was used as reference for all the MCs in the same image (other vertebra [OV]).
For each ROI, the following three indexes were calculated:
-
1.
mean value of the pixels encompassed in the ROI (PRE):
$$ P R E= mean\left( RO{I}_{pre}\right) $$
-
2.
mean value of the difference between post and pre contrast signal intensity (DIFF):
$$ \mathrm{DIFF} = \mathrm{mean}\ \left({\mathrm{ROI}}_{\mathrm{post}}-{\mathrm{ROI}}_{\mathrm{pre}}\right) $$
-
3.
ratio between PRE and DIFF, multiplied by 100 (NORM.DIFF):
$$ NORM. DIFF=\left(\frac{DIFF}{PRE}\right)*100 $$
The three indexes calculated per ROI encompassing an MC were normalised with respect to the relevant control ROIs, to calculate the normalised signal intensity (NSI), as follows:
$$ {\mathrm{NSI}}_{\mathrm{PRE}}=\frac{\left({\mathrm{PRE}}_{\mathrm{MC}}-{\mathrm{PRE}}_{\mathrm{CONTROL}}\right)}{{\mathrm{PRE}}_{\mathrm{CONTROL}}}*100 $$
NSI was extracted for PRE, DIFF and NORM.DIFF values and for each control ROI (SV, OV), for a total of six indexes.
In order to analyse inter-rater and intra-rater reliability, this procedure was repeated by the same operator 5 months after the first evaluation. A second operator, a resident in radiology with 3 years of experience not directly involved in the research, received a brief explanation about the software and the aim of the study before rating all data in one session.
Statistical analysis
Inter-rater and intra-rater agreement was analysed with the interclass correlation coefficient (ICC) (two-way mixed model, type absolute agreement), taking into consideration that an ICC of 0–0.2 represents slight agreement, 0.21–0.4 fair agreement, 0.41–0.6 moderate agreement, 0.61–0.8 substantial agreement and 0.81–1 excellent agreement [21].
The existence of significant differences in the described indexes among MC I, MC II and MC I/II was evaluated with a rank-sum test or Kruskal–Wallis one-way analysis of variance. The choice of a non-parametric test was justified by the non-normal distribution of data as confirmed by the Shapiro–Wilk test. If a statistical difference among groups was found, a multiple comparison procedure with Dunn’s method was performed to establish the existence of difference among pairs. One-sample t tests were carried out to test whether data had a mean significantly different with respect to zero. The presence of correlation between data was studied by Spearman rank-order correlation. Differences were considered significant when p < 0.050.