Animal preparation
The study was approved by the Local Animal Ethics Committee and was performed in accordance with Spanish (RD 53/2013) and European (2010/63/UE) legislation. Wistar male adult rats (n = 27), weighing approximately 425 g each (425.4 ± 43 g), were housed at 22–24 °C, maintained on a 12-h light/dark cycle, and used for the experiments. Water and food were given ad libitum during the experiments. Anaesthesia was induced with 4% isoflurane (Aerrane, Baxter Healthcare, Deerfield, IL, USA) and maintained with 1.5–2.0% isoflurane in O2 at 1 L/min during the whole study. A 22-gauge intravenous cannula (Introcan Safety® IV Catheter, 22G, B. Braun, Melsungen, Germany) was placed in the tail vein for iopamidol (iopamiro 370 mg/mL, Bracco Imaging, Milan, Italy) administration as contrast agent (CA) by means of an infusion pump (Harvard apparatus PHD 2000 Infusion, Holliston, MA, USA). Post-surgically, 250 mL of water with 10% of buprenorphine (buprecare 0.3 mg/mL, Divasa Farmavic, Barcelona, Spain) was given to all operated animals.
Contrast agent dose optimisation
Optimal CA dose to distinguish between injury and neighbouring tissue was first determined by performing a contrast enhancement study in a separate set of experiments including five rats. These were placed in the micro-CT bed, and five different infusion rates (100, 200, 300, 400, and 500 μL/min) were tested for 20 min to obtain different contrast-to-noise ratios (CNR) in acquired images. Imaging parameters were as follows: field of view 40 mm, voxel size 80 μm (isotropic), kVp 90, tube current 200 μA and exposure time 120 s. Images were acquired approximately every 2.5 min, resulting in total of 9 images per rat. Contrast enhancement was assessed in the femoral artery and neighbouring tissue by delineating circular volumes of interest of 1 mm in diameter, obtaining their mean density and standard deviation values. From these, we calculated CNR using following formula [11]:
$$ CNR=\frac{\left|{\mu}_{\mathrm{A}}-{\mu}_{\mathrm{T}}\right|}{\sqrt{\sigma_{\mathrm{A}}^2+{\sigma}_{\mathrm{T}}^2}}, $$
where μA and μT represent mean density values of volumes of interest in femoral artery and tissue area, respectively, and σA and σT represent their standard deviations. Optimal infusion rate providing adequate image contrast was set following the Rose criterion (CNR > 5), as previously described [12].
After CA administration, the pump was stopped, and micro-CT images were acquired. One millilitre of saline solution was injected after image acquisition was finished.
Skeletal muscle injury model
For the injury model, we used a recently developed surgically induced SM injury rat model proposed by Contreras-Muñoz et al [6]. Briefly, 23 anaesthetised rats were immobilised by the fixation of tail and extremities with adhesive strips to a styrofoam surface exposing the ventral side of the right crus. Skeletal traumatic muscle injuries were induced in the rat medial gastrocnemius muscle by a 18-gauge biopsy needle (Bard® Monopty® Disposable Core Biopsy Instrument, Bard Biopsy Systems, Tempe, USA) with a 0.84-mm inner diameter. Transversal biopsy procedure was performed at the muscle-tendon junction level of the left leg medial gastrocnemius muscle (3 mm from the start of muscle-tendon junction and 2 mm in depth). Immediately after muscle injury, a cannula (Introcan Safety® IV Catheter, 22G, B. Braun, Melsungen, Germany) was introduced within the injury in order to proceed with the micro-CT imaging protocol.
Micro-CT
Micro-CT studies were performed using a Quantum FX micro-CT scanner (PerkinElmer, Hopkinton, MA, USA). Rats (n = 23) were positioned on their right side in a bed, and their left leg was immobilised to minimise possible involuntary movement and motion-related artefacts. During the scans, rats were kept under anaesthesia, and CA was administered for 20 min as described above. Two images were acquired at the same day when muscle injury was induced: first an image of the injury with the cannula at 14 min after the beginning of CA infusion for location purposes, and then, after removing the cannula from the injury, a second image was acquired at 20 min right after stopping the CA injection. Again, 1 mL of saline solution was injected after image acquisition was finished to ease contrast clearance.
To image lesion recovery at different time points, 20 rats were sorted into 5 different groups (n = 4 per group) according to the follow-up day at 2, 4, 7, 10, or 14 days after injury, respectively. Additional 3 rats were imaged at all mentioned follow-up days as validation. High-resolution images were acquired so that the centre of the field of view was aligned with the middle of the fibula in sagittal and coronal planes, thus covering the maximal region of the left limb containing the injury. Experimental imaging parameters were field of view 40 mm, voxel size 80 μm (isotropic), kVp 90, tube current 200 μA and exposure time 270 s. All images were reconstructed using a filtered back-projection approach with a Ram-Lak filter, including a ring reduction algorithm [13].
Image processing
Greyscale values between image datasets were normalised by an histogram matching algorithm implemented in the Insight Toolkit [14]. All image datasets in the study were normalised by matching their histograms to the reference image histogram, which was chosen to be the injury image of rat 1 at day 0. Adapted normalised image datasets were then filtered in Fiji [15] using a non-local means denoising filter [16, 17]. Noise standard deviation was set to 2 with smoothing factor set to 1. Contrast-enhanced anatomy, together with high-intensity anatomy like bones, was segmented from denoised images using thresholding based on Renyi entropy [18] with min and max values set to 77 and 255, respectively. These values were experimentally derived from denoised images to ease the injury segmentation process. All segmented contrast-enhanced anatomy images were then manually processed in three-dimensional slicer (version 4.9.0) [19]. The injury was further segmented from contrast-enhanced anatomy mask, excluding surrounding adipose tissue and vasculature. Once the refined SM injury mask was obtained, injury volume was calculated as the mask voxel number multiplied by voxel volume.
Lesion recovery prediction model
Single follow-up day cohort data (n = 20) were used to feed an exponential model of SM injury recovery. The generalised exponential model used was Vt = V0 × exp (b × t), where t is the corresponding post-injury time (days), V0 is the initial injury size (mm3) and b is the healing rate (days−1). Model was trained using the least absolute residual method available in Matlab (The Mathworks Inc., Natick, MA, USA), quantifying both R2 and root mean squared error. Afterwards, injury volumes were predicted in the validation cohort (n = 3) at all monitored post-injury time points.
Statistical analysis
Data are expressed as mean and standard deviation. Injury volumes and model predictions were compared using paired t test and Bland-Altman analysis after testing for normality. All statistical analyses were performed using the Graphpad Prism software (Graphpad Software Inc., San Diego, CA, USA). A p value below 0.05 was considered as statistically significant.