Effect of two eccentric hamstring exercises on muscle architectural characteristics assessed with diffusion tensor MRI

To evaluate the effect of a Nordic hamstring exercise or Diver hamstring exercise intervention on biceps femoris long head, semitendinosus and semimembranosus muscle's fascicle length and orientation through diffusion tensor imaging (DTI) with magnetic resonance imaging.


| INTRODUCTION
Exercise-based interventions are effective in lowering hamstring injury rates. 1 There is strong evidence from a meta-analysis that a 12-week Nordic hamstring exercise intervention can reduce hamstring injury rate by on average 51%. 2 Although effective in daily practice, the underlying preventive mechanism is yet not fully unraveled. 3 A suggested preventive mechanism of the Nordic hamstring exercise intervention are changes in muscle fiber architecture on two-dimensional (2D) ultrasonography. 4 Lengthening of the biceps femoris long head fascicles of up to 2 centimeters is reported. [5][6][7][8][9][10] If the increased fascicle length involves an increased number of sarcomeres in series, it might positively affect flexibility and so reduce the risk of sarcomere overstretching. [11][12][13] A change in pennation angle of the biceps femoris long head might be the other preventive mechanism, but the evidence is conflicting. [6][7][8][9][10] Other reported (functional) results of the intervention is an increase in eccentric hamstring strength, 14,15 and increased anatomical cross-sectional area of the semitendinosus. 5 The semitendinosus is significantly more recruited than the biceps femoris long head during the kneeorientated Nordic hamstring exercise. 16,17 The majority of injuries involve, however, the biceps femoris long head and occur at relative long muscle length, with hip flexed, and knee extended. [18][19][20][21] This has led to the discussion whether the Nordic hamstring exercise is the most appropriate preventive exercise. [22][23][24] Hip-orientated hamstring exercises (e.g., Romanian deadlift, single-leg deadlift and Diver hamstring exercise) are suggested as alternatives, as hip exercises are executed at a longer muscle length with a relatively higher muscle activation pattern for the biceps femoris long head. 19 The Diver hamstring exercise is a simple hip-orientated hamstring exercise that requires no additional equipment, performed over a range of motion at longer muscle lengths, compared with the Nordic hamstring exercise. 19 Effects of the Diver exercise on 2D ultrasound muscle fiber architecture of the biceps femoris long head, semitendinosus and semimembranosus are unknown.
Limitations of 2D ultrasound in Nordic hamstring exercise studies are that findings are predominantly limited to biceps femoris long head, [5][6][7][8][9] based on a single fascicle and these measurements are highly operator dependent. 5,[7][8][9]25 There is furthermore no literature, supporting the generalizability of changes in muscle fiber architecture in the biceps femoris long head to other hamstring muscles. The effect of the Nordic hamstring exercise intervention on the muscle fiber architecture of the semitendinosus and semimembranosus is unknown.
A novel method for evaluating muscle fiber architecture is diffusion tensor imaging (DTI). Diffusion tensor imaging is a non-invasive quantitative magnetic resonance imaging (MRI) technique that measures the motion of water molecules in tissue. Water diffusion in healthy muscle tissue is anisotropic, which means that the majority of water molecules diffuse along the longitudinal axis of the muscle fibers. 26 With custom-built DTI data analysis methods and tractography software, fascicle length and orientation can be reliably determined in lower limb muscles. [27][28][29][30] The aim of this study was to evaluate the effect of the Nordic hamstring exercise and Diver hamstring exercise on muscle fascicle length and orientation in the biceps femoris long head, semitendinosus and semimembranosus through DTI. Secondary aims were to evaluate the effect of these hamstring exercises on strength, flexibility, and individual hamstring muscle volume.

| Study design
The study was a three-arm, single-center, randomized controlled trial performed in a university medical center in the Netherlands. The study was registered in the Netherlands Trial Register (NL7248) on July 13th, 2018, approved by the medical research ethics committee of the Academic Medical Center Amsterdam (NL63496.018.17) and was in accordance with the Declaration of Helsinki. Due to the COVID-19 pandemic, some participants were unable to attend a second measurement session because of shut down of the MRI facility by Dutch governmental sanctions. The study protocol was amended and approved by the medical ethics review committee on April 22nd, 2020. intervention did significantly change the orientation of fascicles of the biceps femoris long head. As both exercises are complementary to each other, the combination is relevant for preventing hamstring injuries.

K E Y W O R D S
diffusion tensor imaging, fascicle length, hamstring injury, pennation angle The COVID-19 amendment voluntarily extended the intervention period of 12 weeks for the already enrolled participants to a maximum of 24 weeks. The effect of an extended intervention period on the primary outcome measures was expected to be marginal as the intervention effect reaches a certain ceiling with time. 7 Patients or the public were not involved in the design, or conduct of this study. At the start of this study, no comparable studies using DTI were available. Therefore, sample size calculation was based on an ultrasonographically determined 16% (19 mm) increase in fascicle length of the biceps femoris long head, following eccentric training. 31 With an approximate effect size of 1.2, power of 80% and an alpha of 0.05, a sample size of 24 participants per group were needed.

| Participants
Participants were recruited from both recreational active and elite basketball teams, via promotion at sports and hospital facilities as well as social media platforms. Detailed information about the study procedure, participant rights, and contact information for further questioning was sent via email or handed over in paper-form before screening for inclusion. Inclusion criteria were 16 years of age or older, basketball player and capable of doing an active exercise program. Exclusion criteria were a hamstring injury within the past year and contraindication for the MRI device (e.g., claustrophobia and pacemaker). All participants gave written consent before the start of data collection.

| Procedure & randomization
All participants were registered in a data management program by the coordinating researcher (JS) (CastorEDC, CIWIT B.V., Amsterdam, the Netherlands). The participants were randomized by a computer-generated scheme (Microsoft Excel ASELECT function) and allocated to a Nordic-group, Diver-group or Control-group by two researchers (GR or JS). The allocation of participants of the same team was made in cluster randomization and allocation of the individual participants were randomized as single entities. All participants took part in two measurement sessions, a baseline measurement in the week preceding the start of the intervention and a follow-up measurement within 7 days after the 12-week intervention period. 7 Participants were included for analysis if they attended both baseline and follow-up measurement sessions. Blinding of participants to their exercise and to the presence of another groups was not guaranteed an feasible when they were part of the same club.

| Intervention & compliance
For both the Nordic and Diver-group, a 12-week intervention protocol was prescribed, translated from proveneffective preventive intervention protocols. 32,33 The number of sessions is described in Table 1. Protocol details were provided via mail, including a step-by-step description and link to a videotaped example. 34 The intervention was unsupervised and participants were advised to execute the exercise at the end of a training yet free to choose appropriate moments. The Control-group allocated participants were asked to continue their usual training regime. In all groups, participants were asked to postpone the start of additional hamstring strengthening exercises to after the follow-up measurement session. In case of participants were already performing hamstring exercises, they were asked to continue their training regime. Compliance with the intervention was evaluated through online questionnaires. A more detailed description of the online compliance questionnaire is included in Appendix S1.

| Hamstring strength and flexibility
At baseline and 12 -week follow-up, hamstring strength was assessed in a prone position, with the knee to 15° and 90° of knee flexion, as determined by visual estimation. The tester placed a handheld dynamometer (MicroFET2, HOGGAN Scientific) at the heel of the participant and applied force to the heel, gradually increasing in 3-5 s. Participants were instructed to resist the applied force (break test). The highest peak force of three attempts in both knee flexion angles was documented for both positions in Newton. Hamstring flexibility was acquired with the passive leg raise and the active knee extension test in degrees. 35 T A B L E 1 Exercise protocol for Nordic hamstring exercise and Diver hamstring exercise intervention.
Week no.

| Magnetic resonance imaging
Magnetic resonance imaging datasets were acquired on a 3 Tesla MRI scanner (Ingenia, Philips, Best). All participants were examined in supine position and feet first using a 16-channel receive coil and 12-channel receive coils located in the scanner table. An elastic band around the feet and sandbags at the lateral side of the lower legs were used to minimize movement and reproduce scanner placement between measurement sessions. The data were acquired in a multi-stack protocol covering the full upper leg region. The scan protocol consisted of four axial proton density weighted Dixon scans for complete anatomical mapping of the hamstring muscles from proximal ischial origin to the most distal insertion on the tibia. Three axial diffusion-weighted (DWI) and axial noise scans corresponded with the three most proximal Dixon sequences and covered all of the hamstring's muscle tissue. The noise scans were for the purpose of data-quality assessment.
The acquisition parameters are provided in Appendix S1.

| Primary outcome measures
The two primary outcome measures were change in fascicle length (e.g., fiber tract lengths representing fascicle length in millimeters) and fascicle orientation (e.g., angle between DTI parameter eigenvector 1 and a reference line representing the intramuscular tendon in degrees) of the biceps femoris long head, semitendinosus and the semimembranosus over 12 weeks, obtained with DTI and fiber tractography. 36

| Secondary outcome measures
Secondary outcome measures were change in isometric hamstring strength (15° and 90° knee flexion), hamstring flexibility (passive leg raise and active knee extension) and individual hamstring muscle volume (in cubic centimeters) over 12 weeks.

| Magnetic resonance imaging & processing
Magnetic resonance imaging datasets were processed using QMRITools (www.qmrit ools.com) for Mathematica (Wolfram Research, Inc., Mathematica, Version 12.1, Champaign). 37 The DWI data were de-noised using a principal component analysis noise algorithm and corrected for motion and Eddy currents using affine registration. 38,39 The diffusion data were then registered to the anatomical space using sequential rigid and B-spline registration to correct for EPI-distortions. The individual stacks of the Dixon and the diffusion scans were merged with an overlap of five slices without motion correction. The diffusion tensor was calculated using an iterative weighted linear least squares (iWLLS) algorithm. Each dataset was manually segmented in ITK-SNAP on the axial scanner reconstructed Dixon water image, on a slice by slice basis for the biceps femoris long head, semitendinosus and the semimembranosus by two examiners, JS and KH ( Figure 1A,B). 40 The proximal and distal borders were manually determined by identifying the first and last slice containing a pixel of the segmented muscles by one examiner, JS. The individual longitudinal muscle length was obtained from the sum of number of slices. The volumes of individual muscle segmentations in ITK-SNAP were stored as secondary outcome measure in cubic centimeters (cm3). An average signal-to-noise ratio was calculated per muscle segmentation with the noise scan as DTI signal divided by the noise.

| Primary outcome measure: fascicle length
Fiber tractography of the individual segmented hamstring muscles was performed using software (vIST/e; Eindhoven University of Technology, Eindhoven, the Netherlands). Fiber tracts were created through an automatically generated axial region of interest at 50% longitudinal muscle length within each muscle segmentation ( Figure 1C-E) and for each muscle, the number of tracts were stored for analysis. The longitudinal center was defined per individual muscle at 50% of the individual longitudinal length. The endpoints of fiber tracts were defined by the borders of segmentation or by stopping criteria of the tractography software (step size of 0.05 per voxel, maximum turning angle per step = 5°, minimal FA = 0.1). 41 These criteria were chosen conservatively to guarantee that the majority of the reconstructed fibers span the full length of intramuscular tissue from proximal to distal. For each participant's measurement session and muscle, the resulting fiber tracts distribution was fitted with a skewed-normal distribution to obtain a mean tract length in millimeters, representing the mean fascicle length of the specific hamstring muscle in millimeters. Fiber tractography and assessment of legitimacy of fiber tracts was carried out by one of the examiners (GS) who was blinded for group randomization. Datasets for which the stitching areas of the stacks caused distinct non-anatomically plausible irregularities in tracts were excluded.

| Primary outcome measure: fascicle orientation
Fascicle orientation maps were determined for the volume of the biceps femoris long head, semitendinosus and semimembranosus individually. 30 The orientation maps consisted of one angle per voxel within the hamstring muscles, with respect to a reference line, placed on the proximal intramuscular tendon of the individual hamstring muscle. To create a reference line, two xyz-coordinates were manually placed in the middle of the intramuscular tendon by one examiner (JS) in the axial in-phase Dixon image, at 40% and 60% of the longitudinal muscle length ( Figure 1G). The fascicle orientation between the DTI eigenvector 1 and the reference line were calculated in each voxel ( Figure 1H) by one of the examiners (LS) who was blinded for group randomization. Because of skewed distributions in fascicle orientations in all three muscles, median fascicle orientations were determined in whole muscle volumes for statistical analysis while excluding the stitching areas of the stacks. These stitching areas sometimes contained distinct erroneous fascicle orientations due to the stitching algorithm.

| Reliability
We refer to Appendix S1 for the reliability assessment of the manual segmentation, longitudinal muscle length and placement of the reference lines.

| Statistical analysis
Statistical analyses were performed using IBM SPSS Statistics (IBM SPSS Statistics for Windows, Version 27.0. Armonk, NY: IBM Corp). Descriptive statistics were used to describe participant characteristics at baseline, which included: age, height and mass. Adjustments were made for baseline variables that influenced the primary outcome with p < 0.10. For analysis of the primary outcome measures fascicle length and orientation, a general linear model for repeated measures was used (repeated withinsubjects factors: Muscle, non-repeated between-subjects factor: Intervention). In a per-protocol sensitivity analysis of the primary outcome measures, we included the Control-group and compliant participants of the Nordic and Diver-group. For analysis of the secondary outcome measures hamstring strength and flexibility, a univariate general linear model was used (between-subjects factor: Intervention). For analysis of muscle volume, a linear model for repeated measures was used with the same factors as the analysis of the primary outcome measures. Absolute changes from baseline to follow-up were included in all tests to test for the effect of the intervention on the between-group differences over 12 weeks. For all repeated measures tests, when sphericity was violated (Mauchly's test p < 0.05), the Greenhouse-Geisser correction was used as an adjustment method. The level of significance was set at an alpha of 0.05 for the main and interaction effects. When appropriate, the level of significance was Bonferroni-adjusted in pairwise post-hoc comparisons.

| Data collection
Between September 2018 and March 2020, 100 nonprofessional competitive basketball players showed interest in participating, from which 72 participants met the inclusion criteria and were allocated to a Nordicgroup (n = 24), Diver-group (n = 24), and Control-group (n = 24) (Figure 2). One participant in the Diver-group withdrew before the baseline measurement session. Due to the COVID-19 lockdown, 20 participants (Nordicgroup, n = 3; Diver-group, n = 6; Control-group, n = 11) could not attend the 12-week follow-up measurement session as planned. They were instructed to voluntarily proceed with their exercise protocol, or regular training regime in the Control-group. The median intervention time of these participants was 12.5 weeks (range: [12][13][14][15][16][17]. Two participants withdrew their interest in participation because continuation of the intervention was not desired (Diver-group, n = 1; Control-group, n = 1). Fifty-three participants (Nordic-group, n = 19; Diver-group, n = 16; Control-group, n = 18) completed a measurement session at baseline and follow-up. Thirty-nine datasets were used for primary outcome measure fascicle length (Nordic-group, n = 13; Diver-group, n = 13; Controlgroup, n = 13) as 14 datasets were excluded because of distinct non-anatomical irregularities in fascicle tracts in the stitching areas. Example of irregularities is provided in Appendix S1. All 53 datasets were used for the primary outcome fascicle orientation. Detailed results about the reliability tests are presented in Appendix S1, and for detailed results about the scan quality and number of tracts for the fascicle length calculation we refer to Appendix S1.

| Baseline characteristics
Baseline characteristics are illustrated in Table 2. No adjustments of baseline variables were made for the primary outcome analysis, as these variables were not statistically associated with fascicle length and fascicle orientation per individual muscle (p > 0.10).

| Fascicle length of the biceps femoris long head, semitendinosus and semimembranosus
There was a significant interaction effect for type of intervention and muscle on between-group differences over 12 weeks for fascicle length (F(4, 70) = 2.6, p = 0.047). Detailed results are presented in Table 3 and illustrated in Figures 3 and 4

| Fascicle orientation of the biceps femoris long head, semitendinosus and semimembranosus
There was a significant interaction effect on the type of intervention and muscle on the between-group differences over 12 weeks for fascicle orientation (F(4, 100) = 3.1, p = 0.019). Detailed results are presented in Table 3  significant interaction effect on the type of intervention and muscle on the between-group differences over 12 weeks for fascicle length (F(4, 58) = 1.7, p = 0.147). There was a significant interaction effect on the type of intervention and muscle on the between-group differences over 12 weeks for fascicle orientation (F(4, 84) = 3.1, p = 0.011). Detailed results of participant's individual intervention compliance and per-protocol analysis are presented in Appendix S1.

| Secondary outcome measures
There were no significant between-group changes over 12 weeks for the secondary outcome measures. Detailed results are presented in Table 3.

| DISCUSSION
In this three-arm randomized controlled trial among amateur basketball players, a 12-week Nordic hamstring exercise intervention, compared with usual training, did significantly increase the fascicle length of the semitendinosus by 14% (2 cm). A 12-week Diver hamstring exercise, compared with usual training, did significantly decrease the angle of fascicle orientation of the biceps femoris long head 9% (2.4°). Both outcome parameters were obtained with DTI of the upper legs.

| Compared to literature
Since no studies reported the effect of exercise interventions on hamstring muscle fiber architecture with the use of quantitative DTI, comparisons with existing literature is limited to 2D ultrasound studies. For fascicle length, previous 2D ultrasound studies described an increment of up to 2 cm of the biceps femoris long head following a Nordic hamstring exercise intervention. 5,[7][8][9] This observation in the biceps femoris long head was not confirmed by our results. We did show a significant increment of fascicle length of 2 cm in the semitendinosus, compared to the Control-group. So far, no literature exists on the evaluation of fascicle length of the semitendinosus and semimembranosus to evaluate the effect of hamstring exercises. The calculated fascicle lengths for biceps femoris long head and semitendinosus are in the same range as mean fascicle lengths from human cadaveric specimens. 42 The main difference between ultrasound and DTI studies is that with ultrasound, the fascicle lengths are predominantly calculated on only one fascicle of interest in a 2D plane in the central part of the muscle. 5,7-9 A recent ultrasound study reported adaptation of fascicle length in response to Nordic hamstring exercise training in the distal but not in the central portion of the biceps femoris long head. 43 In our study, fascicle lengths were calculated on 100-150 fiber tracts per muscle. Diffusion tensor imaging provides a more comprehensive overview of the geometry of the fascicles in 3D, taking into account the T A B L E 2 Baseline characteristics. curvature of fascicles. Simplification to a 2D plane in conventional ultrasound might underestimate fascicle length calculations.

Participant characteristics at baseline
For ultrasonographically determined fascicle orientation in 2D (e.g., pennation angle), there is conflicting evidence of an effect of the Nordic hamstring exercise intervention. [6][7][8][9][10] Both an increase of and decrease of approximately 1-2° have been reported. 6-10 We did not find any significant effect of the Nordic hamstring exercise intervention on the angle of fascicle orientation in any of the three hamstring muscles. For the Diver hamstring exercise intervention, our finding was a decrease in angle of the biceps femoris long head fascicle orientation and this has not been studied before.
The reliability of measuring the fascicle length in hamstring muscles with DTI and tractography software was reported once. 44 These measurements were limited to the biceps femoris long head. They reported good reliability and relative large MDC of 3.5 cm, using a simple and non-modifiable inbuilt manufacturer MRI tractography software. 44 In our analysis, predefined stopping criteria per muscle, manually selected endpoints (e.g., segmentations of the hamstring muscles) and tractography software specially developed for research purposes were used, resulting in MDCs for respectively fascicle length and orientation of <1 cm and 1°. 28 The semitendinosus muscle is affected by the Nordic hamstring exercise and the biceps femoris long head by the Diver hamstring exercise. The Diver hamstring exercise is part of a proven-effective rehabilitation protocol after an acute hamstring injury, executed at longer muscle lengths. 19 Absolute training volume of the intervention protocol in the Diver-group was matched to a proven-effective Nordic hamstring exercise intervention. 33 Besides eccentric loading, the hamstring muscles are also acting as a stabilizer across the Diver hamstring exercise. 45 Difference in relative muscle load is not specified in literature. Which exercise(s) an athlete should perform might depend on the desired goal and the (un)injured state of the athlete. Since the biceps femoris long head is most frequently injured, strength training specifically for this muscle would be a rational decision. 18 The Nordic hamstring exercises intervention has, however, shown its preventive potential in RCTs, increasing eccentric hamstring strength. 2 Training the semitendinosus is possibly behind the success of the Nordic hamstring exercise, protecting the more vulnerable biceps femoris long head. 46 As both exercises are complementary to each other, the combination might be relevant for preventing hamstring injuries. This study has several strengths. First, our DTI approach allows for the assessment of full muscle architecture in 3D in the upper leg muscles. Muscle imaging with 2D ultrasound presents a simplified interpretation of the complex geometry of human muscles fascicles. Second, muscle fiber architecture of all three hamstring muscles were evaluated at the same time. Third, the RCT study design minimalized risk of bias.
This study has several limitations. First, compliance to the intervention protocol in both the Nordic-, and Diver-group and participant drop-out did affect the outcome measures which was not taken into account in the sample size calculation. However, compliance rates were in a comparable range as in previous RCTs reporting preventive effects. 47 Our reported compliance was sufficient to detect an effect on muscle fiber architecture. Second, although training volume was matched between intervention groups, exercise intensity and consequently training stimuli is considered different between exercises. Lower levels of hamstring muscle activity is measured for hip-orientated exercises in general compared to the Nordic hamstring exercise. 16,45,48 Third, the observed nonanatomical irregularities in fascicle tracts in stitching areas between MRI datasets was the reason for excluding 11 datasets in primary analysis. Together with the exclusion of non-compliant participants, the low number of fascicle length assessments is a risk for false negative outcome of the results (type II error) in the per-protocol analysis. Fourth, compression of the hamstring muscles in a supine position in the MRI scanner might potentially affect fascicle orientation. Although there might be a general effect of scanner position on absolute fascicle length and orientation, it should not influence the measured difference over time. Fifth, inclusion was restricted to uninjured basketball players, which might limit generalizability to other sports and injured athletes. Furthermore, if participants were already performing hamstring exercises at start of the intervention period, they were no not asked to pause their training temporarily. Sixth, blinding to the allocated exercise intervention for participants was not possible due to the nature of the content of the exercise. Seventh, isometric strength testing is possibly lacking test specificity. An increase in isokinetic eccentric strength would be expected, however the use of a measurement device for such parameter was not feasible within this study.

| Future directions
A combination of Diver hamstring exercises with Nordic hamstring exercises might potentially increase the preventive effect. A three-armed randomized controlled trial (with a combination group: Nordic + Diver hamstring exercise, Nordic hamstring exercise only and a control group) should assess a possible additional value. Analyzing muscle fiber architecture with DTI has the potential for evaluating changes following exercises directed to performance enhancement, prevention and muscle injury recovery in the hamstring and other muscles.

| CONCLUSION
The Nordic hamstring exercise intervention increases the fascicle length of the semitendinosus. The Diver hamstring exercise intervention decreased the angle of fascicle orientation of the biceps femoris long head. No effect was found for fascicle length and orientation of the semimembranosus. The combination of exercises might be relevant for preventing hamstring injuries.

| PERSPECTIVE
The effect of exercises on hamstring muscle fiber architecture is predominantly evaluated for the most frequently injured biceps femoris long head. A comprehensive overview of the effect of the Nordic and Diver hamstring exercises on the hamstring muscle's three-dimensional fiber architecture through diffusion tensor imaging is described. Our findings suggest that the success behind the Nordic hamstring exercise could be more of a complementary nature; training the semitendinosus protects the more vulnerable biceps femoris long head. By showing that both interventions had a different effect on the muscle's individual fiber architecture, we recognized that exercises have a heterogeneous effect on hamstring muscles. As both exercises are complementary, applying them in combination could become relevant for preventing hamstring injuries.