Diffusion tensor imaging and quantitative T2 mapping to monitor muscle recovery following hamstring injury

MRI examinations are accurate for diagnosing sports‐related acute hamstring injuries. However, sensitive imaging methods for assessing recovery of these injuries are lacking. Diffusion tensor imaging (DTI) and quantitative T2 (qT2) mapping have both shown promise for assessing recovery of muscle micro trauma and exercise effects. The purpose of this study was to explore the potential of DTI and qT2 mapping for monitoring the muscle recovery processes after acute hamstring injury. In this prospective study, athletes with an acute hamstring injury underwent a 3‐T MRI examination of the injured and contralateral hamstrings including DTI and qT2 measurements at three time points: (1) within 1 week after sustaining the injury, (2) 2 weeks after time point 1, and (3) return to play (RTP). A linear mixed model was used for time‐effect analysis and paired t‐tests for the detection of differences between injured and uninjured muscles. Forty‐one athletes (age 27.8 ± 7 years; two females and 39 males) were included. Mean RTP time was 50 (range 12–169) days. A significant time effect was found for mean diffusivity, radial diffusivity, and the second and third eigenvalues (p ≤ 0.001) in the injured muscles. Fractional anisotropy (p = 0.40), first eigenvalue (p = 0.02), and qT2 (p = 0.61) showed no significant time effect. All DTI indices, except for fractional anisotropy, were significantly elevated compared with control muscles right after the injury (p < 0.001). Values normalized during the recovery period, with no significant differences between control and injured muscles at RTP (p values ranged from 0.08 to 0.51). Mean qT2 relaxation times in injured muscles were not significantly elevated compared with control muscles at any time point (p > 0.04). In conclusion, DTI can be used to monitor recovery after an acute hamstring injury. Future work should explore the potential of DTI indices to predict RTP and recovery times in athletes after an acute strain injury.


| INTRODUCTION
Hamstring injuries are the most prevalent injuries in major sports such as football (soccer) and rugby, comprising 12% of all injuries 1 and 37% of all muscle injuries. 2 These injuries have a high re-injury rate of 12%-33%, 2,3 suggesting that there is inadequate assessment of the recovery process, with athletes returning too early to full sports participation. These re-injuries are usually more severe and with a longer absence of play compared with the initial injury. 2 Over the past decade, MRI examinations have been increasingly used for the assessment and follow-up of hamstring injuries. [4][5][6] Primarily, conventional T2-weighted MRI is used to depict edema associated with the injury, and has been shown to be accurate and sensitive for the purpose of diagnosing muscle injuries. 7,8 Edema, detected on these fluid-sensitive sequences, is secondary to the injury itself, develops postinjury, and generally spreads out in a distal direction of the injury location because of gravity. Currently, sparse evidence exists that these techniques are also sensitive for monitoring and predicting recovery. [9][10][11] More specifically, Reurink and colleagues showed that only the absence of hyperintense signal on fluid-sensitive sequences was moderately associated with a shorter time to return to play (RTP) and that proximal tendon involvement was associated with a longer time to RTP. 9 Furthermore, recent work showed that edema is often still present upon RTP and is therefore not an appropriate outcome measure to indicate recovery or to predict the recovery time. 12 Therefore, there is a clinical need to explore advanced, quantitative methods to characterize muscle tissue recovery after severe injury.
Recently, quantitative MRI techniques, such as diffusion tensor imaging (DTI) and quantitative T2 (qT2), have emerged for characterization of muscle micro trauma and exercise effects. [13][14][15] These previous studies 14, 16 showed that DTI is able to detect micro trauma that remained undetected with conventional fat-suppressed T2-weighted MRI, indicating that DTI facilitates sensitive detection of small but distinct changes in muscle microstructure. Furthermore, these studies also showed that DTI can map the recovery of these distinct changes over time. 14,16 Similarly, qT2 MRI has been relevant in the assessment of inflammation in neuromuscular disorders [17][18][19] and monitoring exercise effects in skeletal muscle. 15,16,20,21 We therefore think that both DTI and qT2 mapping could be good candidate imaging techniques to monitor muscle injury recovery and may perhaps even have prognostic capabilities with respect to recovery time. Additionally, both these quantitative techniques have been shown to be sensitive to a multitude of injury-related processes, such as inflammation, water content, membrane damage, membrane permeability, and fibrosis. [22][23][24] Consequently, the combination of the two techniques could also provide more insight into these underlying pathophysiological processes in muscle injury and recovery. Therefore, the aim of this study was to explore the potential of DTI indices and qT2 relaxation times as outcome measures for monitoring the recovery process of the muscle tissue after an acute hamstring injury. Hence, we placed a region of interest (ROI) just above the area of edema at the suspected core location of the injury to characterize the injury itself and its recovery.

| Participants and study set-up
In this prospective study, athletes with an acute hamstring injury underwent three MRI examinations at three different time points: within 1 week after sustaining the injury (time point 1), 2 weeks after the first MR examination (time point 2), and within 10 days after the athlete resumed full training at preinjury intensity (RTP) (time point 3). For some athletes RTP coincided with time point 2. This means that their time point 2 data were also used for RTP analysis and the third MRI examination was not performed. In summary: all athletes were measured at time points 1 and 2, and some at time point 3, while RTP analysis consisted of time point 3 or time point 2 data.
Athletes, both amateurs and professionals from any sport, were recruited online or referred through their sports club physician/physical therapist. Inclusion criteria were new (< 7 days old) acute hamstring injury and age older than 16 years. The clinical diagnosis of an acute hamstring injury is defined as: anamnestic acute injury; anamnestic pain in posterior thigh; localized pain during palpation of the hamstring muscle or during straight leg raising and increasing pain during isometric contraction. Exclusion criteria for this study were: hamstring injury of the affected leg in the prior 2 months, extrinsic trauma on the posterior thigh as cause of the hamstring injury, another concurrent injury inhibiting rehabilitation, and MRI contraindications. Athletes with complete proximal tendon avulsions (grade 3) were also excluded as these athletes did not follow a regular rehabilitation trajectory and it was unknown when and if they returned to full sports activity at preinjury level. For this specific time course substudy, we defined additional exclusion criteria: only one MR examination time point (incomplete protocol), nonmatching field of view (FOV), and insufficient signal-to-noise ratio (SNR) of the DTI data (SNR < 20). Overview of exclusions is available in Table 1. This study was approved by the IRB of Amsterdam University Medical Center. All participants provided written informed consent prior to participating in this study.

| MRI examination
MRI datasets were acquired in both upper legs with a 3-T Philips Ingenia MRI scanner (Philips, Best, the Netherlands) using an anterior 16-channel receive coil and a posterior 10-channel receive coil. Subjects were placed supine in the scanner. To minimize B0 and B1 inhomogeneities, the injured hamstring was always placed on the same side in the scanner (i.e., subjects with right hamstring injury were positioned feet first, whereas those with left hamstring injury were positioned head first). The geometrical planning of the second and third scanning sessions was carefully matched to the first session using anatomical landmarks and screenshots of the planning of the first session. The MRI protocol consisted of an axial DTI sequence, 20 22 an axial and a coronal T2-weighted sequence for injury grading, and an axial proton density (PD) sequence for anatomical reference. For a detailed overview of the remaining sequence parameters, see Table 2.

| Data analysis
MR data were processed using QMRITools (https://github.com/mfroeling/QMRITools) for Wolfram Mathematica 11.3. 25,26 Diffusion data were denoised using a principle component analysis and spatially registered to correct for motion, eddy currents, and EPI distortions using an open-source registration tool (http://elastix.isi.uu.nl). 27 An iterative Weighted Linear Least Squares (iWLLS) algorithm 28 was used for the estimation of the DTI parameters. The diffusion parameters were corrected for the effects of pseudodiffusion (perfusion) based on the Intra Voxel Incoherent Motion (IVIM) model. 29 In order to assess DTI data quality, SNR per muscle was estimated, based on the average signal on the b = 0 image and the sigma of the background noise. Injured muscles with SNR less than 20 were excluded from analysis along with their control muscle. [30][31][32] If only the control muscle had low SNR, the injured muscle was included. An extended phase graph fit was used to calculate qT2 maps, accounting for different relaxation times for the water and fat signal components. 33

| ROI analysis
For each dataset, manual segmentation of the muscles was performed in ITK-snap (www.itksnap.org) on the out-of-phase PD-weighted DIXON image on a slice-by-slice basis, carefully excluding the subcutaneous fat. Injured muscle ROIs consisted of seven slices (35 mm) overlaying the origin of the injury ( Figure 1). In this study we are solely interested in detecting changes in DTI parameters resulting from architectural changes caused by the injury, and not edema, as these might be more insightful with respect to characterization of the recovery of the injury. For this purpose, the location of the primary injury was determined on coronal T2-weighted images, in consultation between two observers (MM, 25 years of experience in evaluation of musculoskeletal MRI; JM, 3 years of experience in evaluation of musculoskeletal MRI) by identifying the slice with the most anomalies in the muscle tissue. The corresponding slice on the axial T2-weighted image and the out-of-phase DIXON was identified and three slices were added cranially and caudally to the main injury slice. Segmentations were performed by one experienced observer (JM, 3 years of experience in evaluation of musculoskeletal MRI). The control muscle ROI consisted of seven closely matching slices in the uninjured contralateral muscle. For the subsequent visits, we visually identified/matched the same location on the coronal and axial T2-weighted images using anatomical landmarks and screenshots of the planning of the first visit. The selected slice and the six adjoining slices were used for our quantitative analysis. The ROIs were projected afterwards onto the DTI and qT2 data to report the mean values and standard deviations per muscle.

| Image scoring
The T2-weighted images with fat suppression were graded by one musculoskeletal radiologist (> 25 years' experience in evaluation of musculoskeletal MRI) for acute muscle injury using a modified Peetrons grading. 35 Signs of muscle injury were graded as 0, 1, 2, or 3, with grade 0 = no abnormalities, grade 1 = edema without architectural distortions, grade 2 = edema with architectural distortions, and grade 3 = complete tear at time point 1. Injuries with grade 3 were excluded from further analysis as these were outside the scope of this study.

| Statistical analysis
All statistical analyses were performed using IBM SPSS Statistics 25. Differences in DTI parameters and qT2 values over time were assessed using a linear mixed model per parameter. Post hoc analysis was performed to determine which of the time points contributed to this overall time effect. Paired samples t-tests were used to determine differences between injured and control muscles at each time point and at RTP. Statistical significance level was corrected for multiple comparisons (7) and set at p less than 0.007 (0.05/7) for all analyses.

| Subject characteristics
The inclusion flowchart is shown in Figure 2. Briefly, 66 subjects were screened for inclusion. One subject was excluded for an MRI contraindica-

| DTI and qT2 relaxation times
Representative DTI (b-value = 0 s/mm 2 ) images, mean diffusivity (MD) maps, and qT2 maps of three athletes who received three MRI examinations are shown in Figure 3. Table 3 lists mean values and standard deviations of all DTI indices and qT2 relaxation times for the three time points and RTP for both the injured and control muscles. Mean DTI and qT2 values and individual datapoints are also plotted in Figure 4 and all the p values are reported in Table 4.  Table 4). The mean values of DTI indices (except for FA) were highest directly after the injury (compared with time points 2 and 3) and reduced during the recovery period (Table 3). There were no overall time effects for qT2 relaxation times in the injured muscles and in the control muscles (Table 4).

|
Post hoc analysis showed that the overall time effect found for MD in the injured muscles was caused by differences between time points 1 and 2, and between time points 1 and 3 ( Table 4). The difference between time points 2 and 3 was not significant (Table 4). A similar pattern was observed for RD, λ 2 , and λ 3 (Table 4) .

| RTP analysis
Time effects were found between time point 1 and RTP (n = 24) for λ 1 (p = 0.002) and MD (p = 0.005) in the injured muscles. There were no time effects for λ 2 , λ 3 , RD, and FA in the injured muscles. There were no time effects between time point 1 and RTP in the control muscles for any of the DTI indices. Similarly, there were no time effects between time point 1 and RTP for qT2 relaxation time in injured and control muscles (Table 4).

| Injured muscles versus control muscles per time point
Mean values of MD (p < 0.001), RD (p < 0.001), λ 1 (p < 0.001), λ 2 (p < 0.001), and λ 3 (p < 0.001) were significantly higher for the injured muscles compared with the control muscles at time point 1 ( Figure 5 and Table 4). There was no statistical difference between the injured and control muscles for FA at time point 1. At time point 2, the mean values of MD (p = 0.002), λ 1 (p = 0.007), and λ 2 (p = 0.003) were significantly higher for the injured muscles, whereas FA, RD, and λ 3 showed no differences between control and injured muscles at time point 2 (Table 4). Lastly, there were no significant differences between the mean values of injured and control muscles for any of the DTI indices at time point 3 and at RTP (n = 24) ( Table 4). Mean qT2 relaxation times of the injured muscles were not significantly different compared with the mean qT2 relaxation times of the control muscles at any of the time points (Figure 4 and Table 4).

F I G U R E 4
Boxplots showing the median, IQR, and Min-Max values for all DTI indices and qT2 relaxation times per time point for both the injured (silver) and control (gold) muscles. Also, the individual injured and control muscles are depicted in the graph as dots (black). Significant differences between the control and injured muscles are indicated with an asterisk (*). DTI, diffusion tensor imaging; FA, fractional anisotropy; MD, mean diffusivity; qT2, quantitative T2; RD, radial diffusivity; λ 1 , first eigenvalue; λ 2 , second eigenvalue; λ 3 , third eigenvalue The observed higher λ 2 and λ 3 directly after the injury could be attributed to increased diffusivity from cell swelling, increased membrane permeability, or membrane damage, disrupting the boundaries of the cell. The normalization of these diffusion indices, early and later on in the rehabilitation process, could indicate regeneration of the myofiber microstructure or a reduction in cell swelling and membrane permeability. We also observed a higher λ 1 value in the injury. However, because fiber length is much larger than the diffusion distance covered, the abovementioned factors and processes probably do not play a role here. The increase in λ 1 (and also in λ 2 and λ 3 ) could be explained by increased contribution of free-water diffusion in the extracellular space. The absence of significant changes in qT2 in the main injury location suggests no prominent inflammatory infiltration in this specific area. However, the slightly higher qT2 values compared with the uninjured leg could indicate some slight cell swelling or increased extracellular space. The main inflammatory response is likely located distally to the main injury location.
Although it is difficult to pinpoint the precise underlying pathophysiological processes responsible for the observed changes in diffusion indices, the significant time effects and normalization of these indices underline the relevance and potential of DTI for mapping muscle recovery processes in skeletal muscle after acute strain injuries.
Our observations are also in line with previous studies that detected similar changes in diffusion indices, that is, the eigenvalues and MD directly increased after acute thigh and calf muscle injuries. 13,38,39 However, in these studies, longer qT2 relaxation times and a reduced FA were also measured. The authors of those works associated the changes in DTI indices with modifications in muscle microstructure, while elevated qT2 relaxation times were attributed to the inflammatory processes. We observed no qT2 or FA changes in the injury, in apparent contradiction to previous studies. 13,38,39 For the qT2 values, we believe this is because we focused on the core of the injury rather than the region characterized by edema. Because we are specifically interested in the microstructural changes at the primary location of injury, we confined our ROI-based analyses to a few slices covering the injury. Not finding changes in FA may be caused by the absence of a pronounced inflammatory effect or because of simultaneous elevation of all three eigenvalues, resulting in a similar FA. In these previous works imaging was also restricted to only one or two time points, which limits insight into the rehabilitation process after injury and restricts comparisons with our data. Only the study by Biglands and colleagues measured at two time points, directly after the injury and at RTP, and found normalization of the diffusion indices and qT2 relaxation times that converge with our data. The additional time point we measured provided enhanced insights into the recovery process Two other studies did measure qT2 and DTI at multiple time points to evaluate muscle micro trauma in marathon runners. 16,20 Small but distinct changes were found for the majority of DTI indices but not for qT2 after running a marathon, with values returning to baseline after 3 weeks. DTI and qT2 were also used to evaluate muscle exercise effects. 15,40 Similar or more pronounced changes in DTI and qT2 have been reported after strenuous types of eccentric exercise that are known to cause muscle damage. Interestingly, these studies also showed that DTI indices and qT2 relaxation times correlate with creatine kinase concentrations, a blood marker for muscle damage, emphasizing the sensitivity of these measures with respect to muscle damage and recovery. 16,40 The different patterns found for qT2 and DTI in our study are in agreement with previous exercise and micro trauma studies 15,16,37 and show the complexity of muscle injuries on a microstructural level. This also illustrates the importance of multiparametric approaches when assessing muscle injuries.
Taken combined, our study explored quantitative MR measures to assess the muscle recovery process over time after an acute muscle injury.
On a group level basis, we showed that DTI indices (except FA) increased after acute strain injury and eventually returned to baseline after full muscle recovery. Specifically, we identified promising diffusion indices (i.e., λ 1 and MD) for mapping muscle recovery with respect to RTP. Looking forward, this also suggests the predictive potential of these indices with respect to time to RTP. Therefore, the next steps should aim at exploring the prognostic potential of the diffusion indices, for instance by narrowing down essential time points for detecting full recovery and by establishing correlations with time to RTP and muscle re-injury rates.
Some limitations of the study should be acknowledged. The overall statistical power of the study decreased towards RTP because of loss of follow-up (17 missing RTP examinations), nonetheless we identified promising DTI indices to map recovery processes after acute strain injury, which can be explored even further in future research. This study consisted of only three time points with at least 2 weeks in between. This could have resulted in missing important data between time points and the final stages of the recovery process that could have been beneficial for the interpretation of the recovery process. A comparatively small ROI comprising 3.5 cm was chosen, which means that even although we took the utmost care to place the ROI to contain the main injury side, we may have missed data by analyzing such a small area. The data in their current form are not corrected for variations in severity and injury location (myotendinous junction, tendon, myofascial). Doing so could provide more insights into the recovery process and may help to identify more personalized outcome measures.

| CONCLUSION
Our DTI protocol proved to be sensitive to detect changes related to the recovery process in muscles following a hamstring injury, while qT2 values did not. The distinct changes in muscle DTI can be used to monitor the recovery of hamstring muscle injuries. Future work should aim to explore the potential of these DTI indices, to predict RTP and recovery times in athletes after an acute strain injury.