It’s a thin line: development and validation of Dixon MRI-based semi-quantitative assessment of stress-related bone marrow edema in the wrists of young gymnasts and non-gymnasts

This prospective observational study was performed according to the Declaration of Helsinki and approved by our internal review board (reference no. 2014_382). Participants between 12 and 18 years old were included from June 2015 until November 2017. Symptomatic gymnasts defined as gymnasts with clinically suspected growth plate stress injury of the wrist were referred by (sports) physicians. Asymptomatic gymnasts and non-gymnast controls were recruited through gymnastics clubs, the bring-a-friend strategy, and via notices within our institution. Written informed consent was obtained from all participants and their parents or guardians prior to participation.

We included 66 participants (mean age, 14.2 years; range, 12.0–17.8 years). These comprised 24 symptomatic gymnasts (mean age, 14.4 years; range, 12.1–16.9 years), as well as 18 asymptomatic gymnasts (mean age, 14.4 years; range, 12.1–17.8 years) and 24 non-gymnasts (mean age, 13.7 years; range, 12.0–17.0 years). Exclusion criteria were history of past fracture, wrist surgery or infection, growth disorder, systemic or oncological disease involving the musculoskeletal system, and fully closed growth plate on hand radiograph. Participants filled out a questionnaire on wrist pain, demographic, and sports characteristics. Gymnasts had performed their sport for at least 1 year and up to 6 months or less prior to study participation.

Posterior-anterior radiographs with a 1.30-m focus detector distance of one (symptomatic) hand and wrist were obtained. MRI of the (symptomatic) wrist was performed in a feet-first, supine position with arms resting alongside the body on a 3.0T MRI scanner (Ingenia, Philips Healthcare) using a dedicated eight-channel receive-only wrist coil. MRI included the qualitative protocol for regular patient care (approximately 20 min) and additional quantitative coronal turbo spin-echo (TSE) T1-weighted and TSE T2-weighted 2-point Dixon series (Table 1).

Bone age was determined on the radiographs using validated software (BoneXpert, v2.0.1.3; Visiana, www.​BoneXpert.​com) [25]. Dixon MRI scans were reconstructed into fat-only and water-only images using the vendor’s standard software, and subsequently blinded and transformed into series of calculated images with every voxel representing the water signal fraction \( \left(\frac{water- only}{\left( water- only+ fat- only\right)}\right) \), using MATLAB (MATLAB and Image Toolbox Release 2017b, the MathWorks, Inc.).

The primary outcome measure was distribution and magnitude of metaphyseal water signal fraction in the radius and ulna. Two musculoskeletal radiology research physicians independently measured metaphyseal water signal fraction on three slices showing the radial maximal width and on the slice showing the ulnar maximal width. Thirteen regions of interest (ROIs) with a 5-mm diameter were drawn using ImageJ (v1.50i, US National Institutes of Health, https://​imagej.​nih.​gov/​ij/​) (Fig. 1). ROIs ranged from the epimetaphysis 0–5 mm proximal to the physis, to the metadiaphysis 20–25 mm proximal to the physis in order to evaluate metaphyseal water signal fraction distribution. After 20 training measurements, both raters performed measurements in 25 participants independently on T1-weighted and T2-weighted images and interrater agreement was determined. In case of good interrater reliability (intraclass correlation coefficient ≥ 0.75), one rater performed measurements in all participants for further analysis.

Intra- and interrater reliability was determined by calculating the intraclass correlation coefficient (ICC) for absolute agreement for each ROI, per sequence, using a two-way random analysis of variance (ANOVA) model (case 2, ICC(2,1)) and the coefficient of variation (CV), defined as the standard deviation of the paired differences divided by the population mean for each ROI. Difference of mean paired differences from zero was assessed using the one-paired t test. Levels of agreement were defined as ICC < 0.5 = poor, ICC 0.5–0.75 = moderate, ICC 0.75–0.9 = good, ICC > 0.9 = excellent [26].

To determine inter-slice reliability, the ICC for absolute agreement between water signal fraction measured on the three middle slices of the radius was calculated using a two-way mixed ANOVA model. Inter-ROI reliability was determined by calculating the ICC for absolute agreement between the mean of ROIs 1, 6, and 8 and ROI 1, and between the mean of ROIs 2, 7, and 9 and ROI 2, using a two-way mixed ANOVA model. Correlation of water signal fraction values between sequences was determined using linear regression.

Ratios were calculated for water signal fraction of the distal, epimetaphyseal ROIs (ROIs 1 and 2 for the radius, ROI 10 for the ulna) compared with the most proximal, diametaphyseal ROI in the same bone (ROI 5 for the radius, ROI 13 for the ulna), which was considered the “reference ROI.” These ratios will further be referred to as “metaphyseal water score.” Between-group differences in participant characteristics and water signal fractions were evaluated using one-way ANOVA, or the Kruskal Wallis test for not normally divided data. In case of significant differences, Tukey or Games-Howell post hoc testing (for normally divided data) or Dunn-Bonferroni post hoc testing (for not normally divided data) was performed to identify differing groups.

Participant demographics are reported in Table 2. Although skeletal and calendar ages did not differ between groups, the ratio of skeletal age compared with calendar age differed significantly between girls in both gymnast groups compared with the non-gymnast control group (Table 2).

For water signal fraction measurements on T1-weighted Dixon, interrater agreement was good to excellent (ICC, 0.79–0.99; CV, 2.6–27.7%) and intrarater agreement was excellent (ICC, 0.91–1.0; CV, 2.1–10.9%) for individual ROIs in the radius and ulna (Supplementary Table 1). Interrater agreement was good to excellent (ICC, 0.88–1.0; CV, 1.9–19.7%) and intrarater agreement was good to excellent (ICC, 0.88–1.0; CV, 3.8–18.7%) on T2-weighted Dixon as well (Supplementary Table 2). Since these outcomes indicated good to excellent interrater reliability for the majority of ROIs, measurements of one observer were used for further analysis.

The ICC for inter-slice agreement between ROIs on the three middle slices of the radius was moderate to excellent on T1-weighted Dixon (ICC, 0.55–0.94) and T2-weighted Dixon (ICC, 0.70–0.96) (Supplementary Table 3).

Inter-ROI agreement for ROI 1 and the mean of ROIs 1, 6, and 8 was excellent on T1-weighted and T2-weighted Dixon (respective ICCs, 0.97 and 0.96). Inter-ROI agreement between ROI 2 and the mean of ROIs 2, 7, and 9 was also excellent on T1-weighted and T2-weighted Dixon (ICCs, 0.95 and 0.97, respectively) (Supplementary Table 4). As the agreement between these ROIs was excellent, we decided to focus on ROIs 1–5 for the radius.

For all ROIs, measurements on T1-weighted and T2-weighted Dixon showed a linear correlation with slopes significantly different from one, with consistently higher water signal fractions on T2-weighted Dixon compared with T1-weighted Dixon (Supplementary Fig. 1). ROIs on T1-weighted images demonstrated smaller standard deviations and therefore a slightly higher effect size.

Water signal fraction was significantly higher in symptomatic gymnasts compared with asymptomatic gymnasts in 3 distal radial ROIs (ROIs 2–4) on T1-weighted or T2-weighted Dixon images, but not in the metadiaphyseal reference area (ROI5) (Table 3). In all radial ROIs on T1-weighted Dixon images, and in the ulnar epimetaphyseal ROI 10, water signal fraction was significantly higher in symptomatic gymnasts compared with non-gymnasts on both sequences. No significant differences were observed in any of the ROIs between asymptomatic and non-gymnasts. Scatterplots of absolute water signal fraction measurements in ROIs 1, 3, and 5 on T2-weiged Dixon images in the three study groups can be found in Fig. 2.

In the radius, the metaphyseal water score (the ratio of the water signal fraction of epimetaphyseal ROI 2 versus the water signal fraction of metadiaphyseal ROI 5) was significantly higher in symptomatic gymnasts compared with asymptomatic gymnasts on both sequences (Table 4, Fig. 3). On T1-weighted Dixon images, this ratio was also higher in symptomatic gymnasts compared with non-gymnasts. In the ulna, the metaphyseal water score of the epimetaphyseal ROI 10 versus the metadiaphyseal ROI 13 was significantly higher in symptomatic gymnasts compared with non-gymnasts on T1-weighted Dixon images (Table 4). No significant differences were observed in ratios between asymptomatic and non-gymnastics on T1- and T2-weighted Dixon images in both the radius and the ulna.

We found increased radial and ulnar metaphyseal water signal fractions in symptomatic gymnasts compared with asymptomatic gymnasts and non-gymnasts, using a reliable semi-quantitative Dixon MRI-based method. The ratio of radial metaphyseal water signal fraction in areas 5–10 mm versus 20–25 mm proximal to the physis was higher in symptomatic gymnasts compared with asymptomatic gymnasts, while their gymnastics level, training hours, and water signal fraction 20–25 mm proximal to the physis did not differ significantly.

Inter- and intrarater agreement was comparable for T1-weighted and T2-weighted Dixon sequences. The consistent discrepancy of water signal fraction on T1- and T2-weighted images most likely originates from the sequences’ difference in T1- and T2-weighting. Although the effect size of the T1-weighted sequence was slightly larger, significant differences were present in both T1- and T2-weighted images, and as T2-weighted Dixon allows both morphological and semi-quantitative image evaluation, we recommend its use for this measurement method to minimize scan time.

Even with good inter-ROI and interslice reliability, water signal fraction measurement in single ROIs may be difficult to reproduce and compare in clinical practice, as reference values have not yet been established and will differ among MRI scanners and protocols. We therefore evaluated the clinical utility of a “metaphyseal water score”: the ratio of the epimetaphyseal area 5–10 mm proximal to the radial physis (ROI2) versus a within-person reference area 20–25 mm proximal to the physis (ROI5). This ratio also showed between-group differences, and is presumably less sensitive to scanner-dependent bias than single-ROI-based water signal fractions when reproduced at other institutions using off-the-shelf Dixon MRI sequences. However, effect size of the metaphyseal water score was smaller compared with absolute water signal fractions, and therefore, its applicability for evaluation of injury severity, prognosis, and relationship with gymnastics training intensity needs further evaluation in larger athlete and non-athlete populations.

The overall increased metaphyseal water signal fractions in symptomatic gymnasts compared with both other groups suggest that higher water signal fraction is indicative of physeal stress injury. However, intensive sports performance, stress injury, and maturation likely all contribute to edema-like changes, and therefore, the line is thin between injury-related edema and physiological increase in metaphyseal water content.

Residual red bone marrow in asymptomatic active children can cause signal intensity changes [19, 20], like high signal intensity on T2-weighted MRI, easily mistaken for abnormalities [27]. Heterogeneous red bone marrow is a common MRI finding when small areas have not yet undergone the physiologic conversion to yellow bone marrow that starts distally in the bone during late childhood [28]. The marrow’s subsequent increase in fat content and decrease in water content [24] likely affect water signal fraction [28]. In this study, participants were therefore matched on skeletal age prior to inclusion to minimize potential interference of maturation with the study’s results.

Additionally, focal periphyseal edema can be seen adjacent to physes in response to growth-induced biomechanical stress in the area of initial physeal closure [29]. The ROIs in this epimetaphyseal region showed highest water signal fractions in all groups, and largest inter- and intrarater variability, suggesting more proximal areas like ROI 2 to be more suitable for identifying stress injury.

Young gymnasts often show attenuated growth and delayed menarche compared with non-gymnasts [30, 31], and in our study, despite the absence of significant differences in absolute skeletal or calendar age between the study groups, skeletal in relation to calendar age was significantly younger in female gymnasts compared with non-gymnasts. Although the exact relationship between maturation status and changes in bone marrow composition is unclear, we postulate that delayed maturation in—especially female—gymnasts may cause a stress-induced marrow shift delay, with relatively higher percentages of red marrow contributing to increased metaphyseal water signal fractions compared with non-gymnasts. In line with this, we found that water signal fraction in the radial reference area (ROI 5) was significantly higher in symptomatic gymnasts compared with non-gymnasts, but not to asymptomatic gymnasts. However, contradictory to our expectations, the increases in water signal fraction of the radial reference ROI in asymptomatic gymnasts compared with non-gymnasts were not significantly different. Future studies with larger sample sizes should explore if the absence of this difference between asymptomatic and non-gymnasts is to be attributed to the relative small study population in this explorative study.

Finally, distal radial bone mineral content and bone mineral density can increase after wrist-loading sports performance during youth [32]. Bone composition changes may influence the bone’s water and fat distributions and MRI signal derived from these components. Effects on ulnar mineralization status have not been documented, but as physeal stress injury reportedly occurs mainly in the radius because of its major weight-bearing function in the pediatric wrist [7], these load-induced changes may be more distinct in the radius.

As the metadiaphyseal ROI 5 showed higher water signal fractions in symptomatic gymnasts compared with non-gymnasts—but not asymptomatic gymnasts—increased metaphyseal water content more proximal to the physis may (partly) result from gymnastics practice. However, asymptomatic gymnasts showed no increased water signal fractions compared with non-gymnasts in any ROIs in the radius and ulna. Symptomatic and asymptomatic gymnasts had a similar training intensity, and therefore, we assume that the significant increase in water signal fractions in symptomatic gymnasts compared with non-gymnasts in nearly all radial ROIs, and one epimetaphyseal ulnar ROI, cannot be merely attributed to (physiological) changes in bone mineral content and density.

Considering these findings and potential gymnastics-induced bone composition changes, we recommend comparing metaphyseal edema scores of symptomatic gymnasts with those of asymptomatic gymnasts instead of non-gymnasts. This comparison is also relevant for clinical purposes; as in daily practice, the method is aimed to confirm or exclude the presence of early stress-related changes in gymnasts. The absence of between-group differences in skeletal age and training intensity suggests that the observed differences in absolute water signal fraction in several ROIs and metaphyseal water score between symptomatic and asymptomatic gymnasts are the result of physeal stress injury. This indicates that the proposed method can aid in detecting early stress-related edematous changes that indicate the presence of physeal stress injury. Further studies should explore the method’s feasibility for assessment of injury severity and for using it for following up stress injuries.