Quantitative ultrasound tissue characterization in shoulder and thigh muscles – a new approach

The right and left supraspinatus muscles, and the right vastus lateralis muscle were scanned in nine participants (7 females and 2 males) with the mean age of 35.8 (23–47) years. All participants except one were right-handed and all were without musculoskeletal disorders. None of them had performed physical exercise prior to the measurements. In addition, the same muscles were scanned in one right-handed female laboratory technician (56 years old) with severe work-related musculoskeletal symptoms in the right shoulder region related to the supraspinatus muscle. The local ethical committee approved the study, and all participants were fully informed about the protocol and procedures, before giving their consent.

Ultrasonography was performed with an ultrasound scanner (Diasonics VST Masters Series System, GE-Medical, Denmark: "Small parts") fitted with a 5 (3.5–7.0) MHz transducer (5.0 MHz 60R CLA Curved Linear Array HI-Density Probe, GE-Medical, Denmark). The digital recordings were stored on discs. The settings of the ultrasound system was standardized for all the participants and kept constant during all measurements to avoid potential changes in the images due to different setttings. We used a depth setting of 5 cm. Depth-gain compensation was built-in in the used scanner, meaning that the gain automatically was increased, while the depth increased to compensate for the loss through the tissue scanned. The acoustic signal received by the ultrasound transducer was converted into digital 8 bit intensity values making the ultrasound image. Each image consisted of pixels with a size of 0.13 mm × 0.13 mm = 0.017 mm². As the approximate speed of sound through an average tissue is 1540 m/s and the transducer used was 5 MHz, the wavelength was approximately 0.3 mm [2, 3, 18]. During the ultrasound scanning, the participants were sitting upright in a chair with the upper arms relaxed and kept in a vertical position and with the forearms resting on the thighs. Contact gel was used for acoustic coupling and care was taken not to exert undue pressure on the imaged tissue, since hard pressure from the transducer may influence the grey scale intensity of the pixels in the images. The transducer was placed perpendicular to the surface of the muscles under examination and adjusted to get the brightest image.

M. supraspinatus: to determine the sites for measuring, the distance between the lateral margin of the acromion and the medial margin of the scapulae at the level of the spina scapulae was measured with a ruler, and a line was drawn parallel to and 2 cm cranial to the spina scapulae. Three measuring sites were marked. The middle site was defined as the midpoint of the drawn line. The medial and lateral sites were located on the line 2 cm medially and laterally to the middle measuring site, respectively, both on the right and left supraspinatus muscles. For each of the supraspinatus muscles, the scanning session consisted of measurements at each of the three measuring sites, longitudinally as well as transversely [19].

M. vastus lateralis: five measuring sites were marked on the right m. vastus lateralis. The middle measuring site was marked in the longitudinal direction by selecting the midpoint between trochanter major and epicondylus lateralis, and in the transverse direction by selecting the midpoint based on muscle palpation. The measuring sites proximal, distal, medial, and lateral, were located 2 cm proximally, distally, medially, and laterally to the middle measuring site.

Each image consisted of 512 × 512 pixels with grey-scale values ranging from 0 to 255. The lowest values corresponded to the darkest, echo-poor areas in the images, while the highest values corresponded to the brightest, high-intensity areas (Figure 1). The region of interest (ROI) for calculations was defined and marked on each image by visual inspection. The criteria for ROI selection were to include as much of the muscle as possible, and to exclude the surrounding bone and the fascia. Furthermore, any areas indicating poor contact between the probe and the skin were excluded. The range of the ROI size was 10085 to 56893 pixels for the supraspinatus muscles, and 25058 to 93957 pixels for the vastus lateralis muscles mainly depending on the scanning direction.

To evaluate the ultrasound images, two types of computerized image texture analyses programmed in Interactive Data Language (IDL) were used (Interactive Data Language, Research Systems, Inc., Boulder, Colorado, USA). The two analyses derive properties of the image content that are related to intensity, and spatial structure (blob analysis). The blob analysis was developed for the present study. The two types of analyses reflect our hypothesis that the muscle tissue composition can be described by a combination of intensity and structure measures.

The intensity analysis consisted of a computation of the frequency distribution of the grey-scale values of the pixels inside the ROI. All images were filtered by the use of a median filter (kernel size = 3 × 3 pixels) to remove single pixel noise. The reproducibility of this analysis has been described earlier [4].

The higher-order grey-scale statistical blob analysis consisted of a computation of the number of structures in the image and their size distribution. A simple method of extracting or segmenting the structures in the image is based on thresholding. Thresholding is the transformation of an input grey-level image to an (segmented) output binary image where input pixels with grey-values below the threshold T have been set to 0, and pixels with grey-value above T have been set to 1. In the resulting binary blob image, connected regions consisting of at least three pixels (corresponding to an area of 0.05 mm²) with pixel value equal to 1, are referred to as blobs. Parameters like the number of blobs and the size distribution of blobs in the blob image may be computed by counting connected pixels with pixel value 1. Since the structures under interest cannot be segmented by applying a single threshold, we computed blob parameters at several (m) thresholds T. (Figure 2 illustrates the procedure). Two ranges of thresholds were selected, the first range corresponding to the grey-level values of muscle tissue, and the second range corresponding to non-contractile tissue; both previously found in healthy participants [4]. The threshold values were defined on the basis of the grey-scale values in the 25%–75% quartiles interval for the mean grey-scale intensity for the healthy group. The threshold values for muscle tissue were 28–38 and 47–59 for the supraspinatus and the vastus lateralis muscle, respectively. Threshold values for non-contractile tissue were 90–106 in grey-scale intensity [4]. The results presented are the calculated average values of the number and the mean area of the blobs in these intervals.

Mean blob size, s, in the blob image at threshold T (1); Mean number of blobs, N, for a set of blob images computed at a range of thresholds (2); and Mean blob size over a range of thresholds, S, (3) was calculated as follows:

where

p _j is the size of blob number j in the blob image,

n _i is the number of blobs in the blob image at threshold i,

s _i is the mean blob size at threshold i.

Data are presented as group mean of 9 participants. In the tables average values, (range), and [quartiles] are used for the average values of the 9 participants, whereas the mean, the variance, the skewness, the kurtosis, and the median are used to describe the distribution of the pixels in the individual ROI-images.

Data was tested for normality, and non-parametric statistics in terms of the Friedman and Wilcoxon-tests were used. The level of significance was selected at p < 0.05 [20]. The Friedman test was used when comparing different measuring sites, and the Wilcoxon test was used when comparing muscles. The Minitab Statistical Software – Release 14 (Minitab Inc, State College, PA, USA).

Significant differences in the grey-scale intensities were found between the vastus lateralis muscle and the supraspinatus muscles, as the vastus lateralis was brighter than the supraspinatus (Tables 1 and 2). The measuring sites of the supraspinatus muscles differed significantly and systematically (Table 1), but no significant differences were found with respect to the sites in the vastus lateralis.

The intensities for the supraspinatus muscle from the shoulder patient appeared higher than from the healthy participants, i.e. exceeding the 25–75% quartiles. Similar deviations were also found for the vastus lateralis muscle (Tables 1 and 2).

For the supraspinatus muscle, the number and size of blobs differed significantly between the measuring sites (Figures 3, 4, 5, 6). These differences were more marked in the dominant side. There were no statistically significant differences between the measuring sites for vastus lateralis either at the low-intensity level or at the high-intensity level. The number of blobs at the high-intensity level corresponding to non-contractile tissue differed significantly between the vastus lateralis and the supraspinatus muscles (Figure 4), as the vastus lateralis had more blobs than the supraspinatus. Also the blob size tended to be larger (0.05 < p < 0.1) in the vastus lateralis than in the supraspinatus at the high-intensity level corresponding to non-contractile tissue (Figure 6). No significant differences between muscles were observed for blobs at the low-intensity level corresponding to muscle tissue.

Comparing the shoulder patient with the healthy participants the supraspinatus muscle from the patient had generally more and larger blobs, i.e. exceeding the 25–75% quartiles (Figures 3, 4, 5, 6).

The grey-scale intensity of the vastus lateralis muscle was higher than the intensity of the supraspinatus muscle, indicating that vastus lateralis contained more non-contractile components than supraspinatus. This result is in accordance with the general anatomy of the two muscles [26]. The vastus lateralis is a long muscle forming the major part of the largest muscle in the body, the quadriceps femoris, and has two large aponeuroses and a tendon. In contrast the supraspinatus muscle is small and triangular with a short tendinous structure. However, in other studies [14, 15, 27] using a similar method an increased echo intensity of the vastus lateralis muscle was assumed to reflect an increased relative proportion of non-contractile tissue. In the liver Layer et al. [28] demonstrated that connective tissue only lead to a weak increase in the grey-level intensity, whereas the addition of connective tissue to a given tissue lipid might reduce the image brightness. Thus the intensity is so far not able to distinguish between different textures of fat and connective tissue and whether an increased brightness of the images is associated with fat or connective tissue [4].

We found differences between the measuring sites of the supraspinatus muscle, indicating inhomogeneity of the muscle. This is in accordance with previous anatomical findings [29]. Therefore more than one measuring site is needed to describe the muscle as a whole. No differences were found between measuring sites in vastus lateralis, but this fact does not necessarily indicate that the muscle is homogeneous as a whole. The thigh muscle is far larger than the shoulder muscles, and the five measuring sites were concentrated on the middle of the muscle.

Since the first-order statistics expresses the distribution of grey-scale values without taking their spatial arrangement into account, muscles with identical echogenicity may have very different tissue structure. Therefore, to improve the fidelity of the image analysis we developed the higher-order grey-scale statistics blob analysis, reflecting the structures in the images, to obtain a more complete description of the scanned muscles. Theoretically, a muscle with a significant tendon structure will result in few and large blobs, whereas a muscle with a diffuse structure will result in many small blobs. This means that the blob analysis may quantitatively separate scans of different muscle composition, even though the mean grey-scale intensity is the same.

The number of the blobs was higher and the size of the blobs tended to be larger in the vastus lateralis muscle compared to the supraspinatus muscle at the level corresponding to non-contractile tissue. These findings indicate as expected from anatomical descriptions that the structures in the vastus lateralis muscle are coarser and larger (Figure 1).