Sage Journals: Discover world-class research

Abstract

This study evaluated the repeatability and reproducibility of using high-frequency quantitative ultrasound (QUS) measurement of backscatter coefficient (BSC), grayscale analysis, and gray-level co-occurrence matrix (GLCM) textural analysis, to characterize human rotator cuff muscles. The effects of varying scanner settings across two different operators and two US systems were investigated in a healthy volunteer with normal rotator cuff muscles and a patient with chronic massive rotator cuff injury and substantial muscle degeneration. The results suggest that BSC is a promising method for assessing rotator cuff muscles in both control and pathological subjects, even when operators were free to adjust system settings (depth, level of focus, and time-gain compensation). Measurements were repeatable and reproducible across the different operators and ultrasound imaging platforms. In contrast, grayscale and GLCM analyses were found to be less reliable in this setting, with significant measurement variability. Overall, the repeatability and reproducibility measurements of BSC indicate its potential as a diagnostic tool for rotator cuff muscle evaluation.

Keywords

high frequency imaging quantitative ultrasound backscatter coefficient textural characterization quantitative imaging biomarkers alliance rotator cuff muscles

Introduction

Rotator cuff tears are one of the most common causes of shoulder pain, affecting millions of patients globally.^1,2 After rotator cuff tendon injury, the muscle progressively deteriorates with fibrosis and fat deposition, complicating surgical repair and worsening clinical outcomes.^3,4 Medical imaging techniques, in particular, ultrasound (US) and magnetic resonance imaging (MRI), are critical for diagnosing rotator cuff tendon tears and for evaluating the status of the muscle.⁵ Reported performance of US for rotator cuff tendon tears is similar to that of MRI,⁶ though US is only moderately accurate for the diagnosis of rotator cuff fatty atrophy.⁷ However, the added advantages of accessibility, portability, and cost-effectiveness make US particularly appealing.

B-mode image evaluation, the mainstay of US imaging in routine clinical practice, provides subjective rotator cuff assessment. Quantitative ultrasound (QUS) techniques have emerged, which convey more objective information about tissue status.⁸ One group of techniques reduces system-dependent effects by utilizing raw radiofrequency (RF) data and calibrated phantoms with known acoustic properties to derive fundamental tissue parameters, such as backscatter coefficient (BSC).^9,10 The efficacy of the BSC for normal and diseased tissue characterization has been demonstrated by several authors in a variety of organs such as the liver,¹¹ kidney,¹² and prostate,¹³ though utilization in skeletal muscle has been lacking.

For US quantification of muscle tissue, typically post-processed B-mode images have been used, with one of the most common measures being grayscale analysis, referred to as “echo intensity.”^14–17 Some authors have also extracted texture parameters from B-mode images, such as through gray-level co-occurrence matrix (GLCM) analyses.^15,18 Notably, virtually all prior studies using US to quantify muscle tissue maintain constant system settings between participants regardless of size or pathology, as it is generally accepted that B-mode-based analysis is sensitive to varying settings such as beam focus, frequency, transmit and receiver gains, and time-gain compensation (TGC).¹⁹ However, in clinical settings, radiologists and sonographers adjust these settings to optimize image quality and contrast. Variables that motivate the different settings include patient body habitus with resultant varying target tissue depths, differing disease states, and subjective operator preference of tissue contrast.²⁰

There is an increasing need to accelerate clinical translation of QUS. However, there are insufficient studies of repeatability and reproducibility involving muscle, particularly in accordance with guidelines endorsed by the Quantitative Imaging Biomarkers Alliance (QIBA),^21,22 and there is incomplete understanding of the sources of variability (different US machines, operators) that affect QUS for this application. Thus, the purpose of this study was to assess the repeatability and reproducibility of QUS based on analysis of raw RF data (BSC) for in vivo rotator cuff muscle evaluation, and compare these outcomes with those of grayscale and textural analysis (GLCM) applied to routine B-mode images of the rotator cuff. The effects of varying scanner settings across different operators and US machines were investigated in a healthy volunteer and a patient with chronic rotator cuff injury. We hypothesize that normalized BSC-based parameters are repeatable and reproducible outcomes for rotator cuff muscle evaluation, and that these parameters would outperform B-mode image-based analyses in the setting of varying scanner settings.

Materials and Methods

Study Design and Ultrasound Acquisition

Our institutional review board approved this study and written informed consent was obtained. Two participants were recruited, including one healthy volunteer without shoulder complaints (healthy participant, 36-year-old man, BMI: 18.6) and one patient with known, chronic massive rotator cuff tearing in both shoulders (injured participant, 73-year-old man, BMI: 28.2). Magnetic resonance imaging exams interpreted by a musculoskeletal radiologist demonstrated that the healthy participant had intact bilateral rotator cuffs and normal supraspinatus and infraspinatus muscles (all Goutallier grade 0), whereas the injured participant had bilateral massive rotator cuff tears and diseased supraspinatus (both Goutallier grade 3) and infraspinatus (both Goutallier grade 2) muscles.²³ The participants were selected based on the fact that they had bilateral, symmetric muscles that were representative of clinically important grades. Goutallier grade 0 is entirely normal and grades 2 and 3 are the most important grades in the determination of potential surgical treatment because those are the grades where significantly higher failure rates occur.^24,25 These two participants represent clinically significantly different conditions; a larger study of participants across all clinical grades was purposely not performed to minimize the introduction of other patient-specific variables in order to focus on scanner and operator effects.

Two clinical US machines with linear probes were utilized (14L5, S2000, Siemens Healthineers, Erlangen, Germany and UHF22, Vevo MD, Fujifilm Visualsonics, Toronto, ON, Canada) and beam-formed RF signals and B mode images were both acquired. Imaging was performed by two experienced operators (E.Y.C, a musculoskeletal radiologist with 12 years of US experience and L.T.S, a general radiologist with 9 years of US experience). The supraspinatus muscle was imaged in short axis approximately 1-inch medial to the acromion and the infraspinatus muscle was imaged in a similar plane inferior to the scapular spine as shown in Figure 1. Presets were restored to default values before each muscle acquisition and operators were instructed to adjust depth, focus, TGC and receiver gain (but not frequency) to optimize the image quality as would be performed in routine clinical practice. After each muscle RF acquisition, data from a commercially available homogeneous tissue-mimicking phantom (calibrated from 1 to 10 MHz) containing 117GU Zerdine formulation (Sun Nuclear, Norfolk, Virginia) was obtained. After the phantom image acquisition, the presets were restored again and a separate B-mode image was acquired by each operator. Three acquisitions of muscle RF, phantom RF, and B mode data were acquired bilaterally by each operator on the two participants, with three repeated measurements for each acquisition.

Figure 1.

Imaging location of ultrasound (US) exam with corresponding MR image. (A) Photograph of a volunteer shows the imaging locations for the supraspinatus and infraspinatus muscles (red lines). The curved blue line outlines the scapular spine and acromion. (B) The sagittal T1-weighted fast-spin echo image from the healthy volunteer shows the normal supraspinatus and infraspinatus muscles without fatty infiltration (Goutallier grade 0). (C) Flow chart of study design. There are two operators, two machines, two participants, two sides, two muscle groups, 3 different acquisitions (RF muscle & phantom, B-mode images) with three repeated measurements.

In total, there were two operators (E.Y.C and L.T.S), two machines (S2000, 14L5 and Vevo MD UHF 22), two participants (healthy and injured), two sides (right and left), two muscle groups (supraspinatus and infraspinatus), three repeated measurements per acquisition, comprising 96 elements of both BSC and B-mode based measurements in the dataset of this study shown in Figure 1(C)).

Analytical Method

QUS Analysis Methods

BSC ( $c m^{- 1} S r^{- 1}$ ) is a parameter used to assess the scattering properties (“echogenicity”) of tissue, defined as the differential backscatter cross-section per unit volume (i.e., the backscattered power received by the transducer divided by the incident intensity, normalized to the scattering volume). The BSC is a quantitative measure of the intrinsic ability of the tissue to redirect US energy to the transducer in the backscattering direction. Accurate measurement of the BSC must account for factors that affect the scattered signal, such as instrumentation, transmission, and attenuation.

The BSC was determined using the reference phantom method that explicitly accounts for experimental factors affecting the ultrasound signal.²⁶ This method involves a comparison at the same depth of the power spectra from tissue with the power spectra from a reference phantom whose BSC and attenuation coefficient are known:

B S C_{t i s s u e} (f) = B S C_{r e f} (f) \frac{P_{t i s s u e} (f)}{P_{r e f} (f)} e^{4 d [α_{t i s s u e} (f) - α_{r e f} (f)]}

(1)

where f is the frequency of the acoustic waves, $B S C_{r e f} (f)$ is the BSC of the reference phantom material measured in a water bath with a single element transducer, $α_{t i s s u e} (f)$ is the attenuation of the tissue, $α_{r e f} (f)$ is the attenuation of the reference phantom, $P_{r e f} (f)$ is power spectrum recorded from the phantom (single frame) and $d$ is the distance between the transducer and the region of interest (ROI). All BSCs reported in this study were logarithmically transformed, where $B S C (d B) = 10 l o g_{10} (B S C_{t i s s u e} / B S C_{0})$ , with $B S C_{0} = 1 c m^{- 1} S r^{- 1}$ . The integral BSC (iBSC) was extracted by normalizing the integrated BSCs between 7 and 10 MHz for the 14L5 transducer. For UHF 22, although the central frequency range is 6 to 15 MHz, 8 to 10 MHz range was selected because the reference phantom is only calibrated to 10 MHz. By using the iBSC, the signal-to-noise ratios in the central bandwidth were optimized.²⁷ Attenuation compensation for the heterogeneous tissue types overlying the rotator cuff muscles was performed in a similar manner to previous work.²⁸ For each layer of tissue skin,²⁹ muscle,³⁰ and fat³¹ visible in the image, we estimated the combined attenuation coefficient using the following equation:

a_{t o t} (f) = \sum_{j = 1}^{N - 1} a_{j} (f) \frac{z_{j}}{z_{T}}

(2)

with N the number of different tissue layers, $a_{j}$ the attenuation coefficient of the jth layer, ∆z_j,the corresponding layer thickness, and $a_{t o t} (f)$ as the effective attenuation of all the tissue layers overlying the ROI for a total thickness of $z_{T}$ .

Image GLCM Texture Analysis

The B-mode images recorded from the two US machines were analyzed using the GLCM algorithm.³² The GLCM is defined as a histogram of co-occurring grayscale intensity pairs in corresponding pixels of an image:

C_{Δ x, Δ y} (m, n) = \sum_{x = 0}^{n - 1} \sum_{y = 0}^{m - 1} {\begin{matrix} 1, i f I (x, y) = m a n d \\ I (x + Δ x, y + Δ y) = n \\ 0, o t h e r w i s e \end{matrix}

(3)

The B-mode US image is treated as a matrix of grayscale pixel intensities, $I (x, y)$ , at the xth and yth matrix location. $(m, n)$ are the grayscale pixel intensity levels. Eight different grayscale pixel intensity levels were considered in this study. The offset ∆χ is equal to 0 in the horizontal direction and ∆y is equal to $λ / 8$ , with $λ$ representing the wavelength of the acoustic wave propagating through the tissue. Contrast values was extracted from each GLCM. The contrast feature measures the local variations of grayscale intensity present in the B-mode image. All these features were extracted using MATLAB (v2020b, The Math Works, Natick, MA).

Grayscale Analysis

0 to 255 levels of the grayscale B mode images intensity were considered. Grayscale mean and standard deviation values were assessed on the same ROI as the GLCM analysis by utilizing “regionprops” function in MATLAB (v2020b, The Math Works, Natick, MA).

ROI Selection

RF and B-mode datasets were arbitrarily assigned to two scientists experienced with QUS processing. The beamformed RF data analysis was standardized using an estimator graphical user interface (GUI) for offline processing. The supraspinatus (Figure 2(A), (B), (E), and (F)) and infraspinatus (Figure 2(C), (D), (G), and (H)) muscle ROIs and overlying tissues for both RF and B-mode data were outlined by the two scientists under the guidance of a musculoskeletal radiologist. Specifically, the ROIs of the cuff muscles were delineated by locating the overlying trapezius and deltoid muscles and then enclosing the epimysium of the supraspinatus and infraspinatus muscles. The GUI selected sub-ROIs within the given ROI with 75% overlap. The sub-ROI dimensions were $= 15 λ$ in both axial and lateral directions.

Figure 2.

Representative B-mode ultrasound (US) images of rotator cuff muscles with orange dash lines as the region of interest. (A) Image of the supraspinatus muscle in a healthy volunteer obtained with the 14L5 transducer. (B) Image of the supraspinatus muscle in a patient with chronic, massive rotator cuff tearing obtained with the 14L5 transducer. (C) The image of the infraspinatus muscle in a healthy participant was obtained with the 14L5 transducer. (D) The image of the infraspinatus muscle in an injured participant was obtained with the 14L5 transducer. (E–H) B-mode images were obtained in the same manner as (A–D) except with the UHF22 transducer. Note that the echogenicity of the muscles is not comparable between images as depth, focus, and gain have been independently adjusted.

Statistical Analysis

Statistical analysis was performed with IBM SPSS Statistics for Windows version 28.0 (IBM, Armonk, NY, USA). Descriptive statistics were summarized with mean ± standard deviation. QIBA endorsed guidelines were followed and a summary of the metrics used in this study is shown in Table 1.^21,22

Table 1.

Summary of All QIBA Repeatability and Reproducibility Statistical Metrics.

Repeatability and Reproducibility Metric	Between-operator SD	Between-participant SD	Repeated measurement SD	Reproducibility SD ( $σ_{R D C}$ )	RC	RDC	ICC(1,1)
Model representation	$σ_{O}$	$σ_{P}$	$σ_{e}$	$\sqrt{σ_{O}^{2} + σ_{O P}^{2} + σ_{e}^{2}}$	$2.77 σ_{e}$	$2.77 σ_{R D C}$	$\frac{σ_{P}^{2}}{σ_{P}^{2} + σ_{e}^{2}}$

RC = repeatability coefficient; RDC = reproducibility coefficient; ICC = intraclass correlation coefficient; SD = standard deviation.

To test the between-image repeatability, the dataset was divided into subgroups since factors (i.e., operators, left/right side, machines) may affect repeatability. The repeatability was assessed using a one-way random-effects model:

Y_{i j} = μ + μ_{i} + ε_{i j}, (i = 1, 2, j = 1, 2, 3)

(4)

where $Y_{i j}$ is the jth repeated measurement on the ith participant. µ is the overall mean, and $μ_{i} ~ N (0, σ_{P}^{2})$ , $ε_{i j} ~ N (0, σ_{E}^{2})$ are the independent random variables of different participants and repeated measurements. $N (0, σ^{2})$ denotes normal distribution with a mean equal to 0 and standard deviation of $σ$ . Two forms of intraclass correlation coefficient (ICC) were also assessed in this study, ICC(1,1) and ICC(1,3), representing the ICC calculated from a single measurement and from an average of three repeated measurements, respectively. ICC was interpreted as follows: poor reliability at <0.5, moderate reliability between 0.5 and 0.75, good reliability between 0.75 and 0.9, and excellent reliability at >0.9.³³

The reproducibility of operators/machines was assessed using a two-way random effects model:

Y_{i j k} = μ + α_{i} + β_{j} + α β_{i j} + ε_{i j k}, (i = 1, 2, j = 1, 2, k = 1, 2, 3)

(5)

where $Y_{i j k}$ is the kth repeated measure from the ith participant with the jth operator and $α_{i} ~ N (0, σ_{P}^{2})$ , $β_{j} ~ N (0, σ_{O}^{2})$ , $α β_{i j} ~ N (0, σ_{P O}^{2})$ and $ε_{i j k} ~ N (0, σ_{E}^{2})$ are the independent random variables of different participants, different operators, participants by operator interaction and repeated measurements, correspondingly.

Results

In this study, we evaluated three effects of variability inter-images, inter-operator, and inter-machine on three different outcome parameters: iBSC, B mode grayscale intensity, and GLCM contrast under the QIBA guidelines for technical performance assessments.

B-Mode, BSC, and GLCM Textural Measurement Results

Representative B-mode images from the participants are shown in Figure 2 with orange dashed lines highlighting the ROIs. The left column of images showed the supraspinatus and infraspinatus muscles acquired from the healthy volunteer (healthy participant). The right column of images was acquired from the patient with chronic, massive rotator cuff tearing (injured participant). B-mode images showed injured participant to have a deeper targeted muscle compared with healthy participant. Both radiologists independently observed that the rotator cuff muscles in the healthy participant were generally more hypoechoic with higher contrast compared to surrounding fat, whereas the internal architecture of the degenerated rotator cuff muscles in injured participant was generally effaced.

Twelve single-image BSCs (three each for two sides by two operators) were computed for each participant. Figure 3(A) and (B) display the BSC(f) curves from the 14L5 probe on healthy and injured participants. The injured participant (red) demonstrated increased BSC values compared to the healthy participant (blue) for both supraspinatus and infraspinatus muscle groups. Figure 3(C) and (D) show BSC(f) curves from the UHF22 probe for both participants with a statistically significant difference between healthy and injured participants. Figure 3(E) and (F) summarize the iBSC outcomes by including two different operators, left and right sides of the rotator cuff, and three repeated measurements. Injured participant demonstrated increased iBSC value compared with healthy participant for both infraspinatus and supraspinatus muscle for both probes.

Figure 3.

Quantitative ultrasound outcomes from rotator cuff muscles. (A) Raw BSC data from two operators on the supraspinatus muscles from two participants using the 14L5 probe (healthy participant in blue, injured participant in red). (B) Raw BSC data from two operators on the infraspinatus muscles from two participants using the 14L5 probe (healthy participant in blue, injured participant in red). (C and D) Raw BSC data for supraspinatus and infraspinatus muscles using the UHF22 probe. (E and F) Bar plot of iBSC data for supraspinatus and infraspinatus muscles using 14L5 and UFH 22probe.

Repeatability of Single-Image Measures, Between-Operator and Between-Machine Reproducibility of iBSC Parameter

The between-image repeatability was evaluated under specific conditions (two operators using two machines on each side of the two participants for two muscle groups), which resulted in six measurements in each individual group (3 measurements each on healthy participant and injured participant). The descriptive statistics, between-participant standard deviation (SD), between-image SD, repeatability coefficient (RC), ICC(1,1), and ICC(1,3) for iBSC are summarized in Table 2. By examining the descriptive statistics, the data acquired using the 14L5 (5.21 ± 18.73) and UHF22 (2.95 ± 7.93) transducers show a similar order of magnitude of iBSC within the central frequency bandwidths (7–10 MHz for 14L5, 8–10 MHz for UHF 22).

Table 2.

Between Images Repeatability Estimate for the iBSC Measures Under Various Conditions on Human Shoulder Muscles (Supraspinatus and Infraspinatus.

Conditions					Summary statistics of iBSC (dB)			ICC Estimates of iBSC(dB)
Operator group	Side (left/right)	Transducer	Muscle type	Mean	Between-participants SD	Between-image SD	RC	ICC (1,1)	ICC (1,3)
1	Left	14L5	Supraspinatus	5.75 ± 26.50	34.08	2.31	6.40	1.00	1.00
	Left	14L5	Infraspinatus	0.34 ± 8.85	10.45	3.57	9.90	0.90	0.96
2	Left	14L5	Supraspinatus	11.61 ± 12.22	11.51	8.36	23.15	0.65	0.85
	Left	14L5	Infraspinatus	–0.41 ± 25.89	33.14	3.37	9.32	0.99	1.00
1	Right	14L5	Supraspinatus	3.13 ± 19.07	21.43	9.38	25.98	0.84	0.94
	Right	14L5	Infraspinatus	5.69 ± 16.91	20.66	5.46	15.12	0.94	0.98
2	Right	14L5	Supraspinatus	12.06 ± 22.94	26.87	10.66	29.54	0.86	0.96
	Right	14L5	Infraspinatus	3.01 ± 16.29	17.90	8.55	23.68	0.81	0.93
1	Left	UHF22	Supraspinatus	1.97 ± 2.79	2.63	1.91	5.29	0.65	0.85
	Left	UHF22	Infraspinatus	2.17 ± 8.20	9.53	3.58	9.92	0.88	0.96
2	Left	UHF22	Supraspinatus	4.24 ± 4.43	4.41	2.82	7.81	0.71	0.88
	Left	UHF22	Infraspinatus	5.99 ± 7.23	6.64	5.08	14.01	0.63	0.84
1	Right	UHF22	Supraspinatus	–1.00 ± 4.48	10.45	4.48	12.41	0.84	0.93
	Right	UHF22	Infraspinatus	5.80 ± 13.53	17.29	1.88	5.21	0.99	1.00
2	Right	UHF22	Supraspinatus	0.30 ± 4.44	5.63	0.83	2.301	0.98	0.99
	Right	UHF22	Infraspinatus	6.57 ± 7.97	9.17	3.62	10.03	0.87	0.95

In this scenario, each group contains six repeated measurements (from healthy and injured participants). iBSC = integrated backscatter coefficient; RC = repeatability coefficient; ICC = intraclass correlation coefficient; SD = standard deviation.

As shown in Table 2, the between-participant SD values are larger than the between-image SD values, indicating repeatability of the iBSC measurement. Our results also demonstrate generally high reliability (ICC values in six cases were classified excellent, six cases were good, and four cases were moderate). No iBSC measurements were classified as poor reliability. Moreover, ICC(1,3) for the three-image measures had 12 cases classified as excellent and four cases as good reliability. Overall excellent to good reliability was demonstrated for the BSC-based measurements for both US systems.

The between-operator reproducibility was assessed using two-way random effect models with operators and participants as the main random effects. The iBSC statistical metrics, between-participant SD, between-operator SD, QIBA reproducibility SD, and between-image SD values without dividing by different operators, are summarized in Table 3. For each muscle, each side, and each machine (eight different groups), the between-operator and QIBA reproducibility SD values are smaller than between-participant SD values, indicating that less variability was introduced by the operators, interaction, and error terms. As a result, there is excellent reproducibility by the two operators for iBSC measurements.

Table 3.

Operators Reproducibility Estimate for Single-iBSC Measures Under Various Conditions on Human Shoulder Muscles (Supraspinatus and Infraspinatus), Calculated Using the Two-Random Effect Model.

Conditions				Summary statistics of iBSC (dB)
Side (left/right)	Muscle type	Transducer	Mean	Range	Between-participants SD	Between-operators SD	QIBA-reproducibility SD	Between-image SD	RDC
Left	Supraspinatus	14L5	8.34 ± 20.31	[–19.82,33.20]	21.41	0*	12.46	7.27	34.52
Left	Infraspinatus	14L5	–0.03 ± 18.45	[–26.45,25.71]	18.80	0*	11.63	3.47	32.22
Right	Supraspinatus	14L5	7.94 ± 20.89	[–17.14,40.55]	24.65	3.42	9.82	10.11	27.21
Right	Infraspinatus	14L5	4.35 ± 15.89	[–23.41,26.01]	19.70	1.50	6.26	7.17	17.35
Left	Supraspinatus	UHF22	1.55 ± 1.86	[–1.21,4.40]	1.83	0.66	1.35	1.21	3.74
Left	Infraspinatus	UHF22	2.04 ± 3.82	[–4.42,7.27]	4.21	1.20	2.33	2.20	6.45
Right	Supraspinatus	UHF22	–0.91 ± 3.54	[–6.61,4.27]	6.39	0*	4.90	4.18	13.58
Right	Infraspinatus	UHF22	3.09 ± 5.30	[–4.68,9.45]	6.38	0*	2.32	1.44	6.43

iBSC = integrated backscatter coefficient; RDC = reproducibility coefficient; SD = standard deviation.

Means for the two-random effect method, a negative or zero component estimate occurs.

S2000 exhibits higher overall mean iBSC (5.21 ± 18.73) than that of Vevo MD (2.95 ± 7.93) potentially due to different central frequency range of the two transducers. Table 4 demonstrates that iBSC values showed excellent reproducibility between machines since the between-machine SD values are lower than the between-participant SD values for every condition. QIBA-reproducibility SD values are also lower than the between-participant SD values for four cases. In comparison, the mean RDC for the between-operator test was 17.69 ± 12.25, which was lower than the mean RDC for the between-machine test at 29.72 ± 18.32. Overall, iBSC parameter acquires a good repeatability and reproducibility of detecting the healthy versus injured rotator cuff muscles.

Table 4.

Machine Reproducibility Estimate for Single-BSC Measures Under Various Conditions on Human Shoulder Muscles (Supra and Infra), Calculated Using the Two-Random Effect Model.

Conditions				Summary statistics of iBSC (dB)
Side (left/right)	Muscle type	Operator	Mean	Range	Between-participants SD	Between-machines SD	QIBA-Reproducibility SD	Between-image SD	RDC
Left	Supraspinatus	1	3.86 ± 18.07	[–19.82,33.20]	9.86	0*	15.95	2.12	44.19
Left	Infraspinatus	1	1.26 ± 8.19	[–8.84,12.93]	10.19	1.21	3.22	3.57	8.93
Right	Supraspinatus	1	–2.53 ± 9.08	[–13.22,8.54]	15.40	0*	8.80	7.08	24.38
Right	Infraspinatus	1	5.80 ± 13.53	[–9.36,18.90]	19.03	0*	3.78	4.08	10.47
Left	Supraspinatus	2	7.58 ± 10.34	[–14.77,21.95]	7.95	1.80	8.81	7.36	24.40
Left	Infraspinatus	2	2.79 ± 18.43	[–26.45,25.71]	15.53	12.15	22.28	4.31	61.71
Right	Supraspinatus	2	–1.00 ± 4.48	[–4.37,5.42]	10.23	0*	15.52	8.34	42.99
Right	Infraspinatus	2	6.57 ± 7.97	[–1.03,16.51]	13.21	0*	7.49	6.56	20.73

RDC = reproducibility coefficient; SD = standard deviation.

Means for the two-random effect method, negative or zero component estimate occurs.

Repeatability of Single-Image Measures, Between-Operator and Between-Machine Reproducibility of Grayscale Mean Intensity Parameter

The repeatability metrics for the grayscale mean are summarized in Table 5. The 14L5 transducer acquired higher grayscale mean values than the UHF22 transducer, but the standard deviation values were lower than those from the UHF22. Poor repeatability was observed since the between-image SD values were equal to or larger than the between-participant SD. As a result, ICC(1,1) for grayscale mean had one case as excellent, six cases as moderate, and nine cases as poor reliability. Even though ICC(1,3) was for an average of three repeated measurements, only one case was considered excellent, six cases as good, two cases as moderate, and seven cases as poor.

Table 5.

Repeatability Measurements of Grayscale Mean Measurements on B-mode Rotator Cuff Muscles.

Conditions					Summary statistics of grayscale mean			ICC estimates of grayscale mean
Operator group	Side (left/right)	Transducer	Muscle type	Mean	Between-participants SD	Between-image SD	RC	ICC (1,1)	ICC (1,3)
1	Left	14L5	Supraspinatus	238.14 ± 6.86	5.54	5.35	14.83	0.52	0.76
	Left	14L5	Infraspinatus	234.92 ± 3.70	2.45	3.18	8.80	0.37	0.64
2	Left	14L5	Supraspinatus	238.81 ± 7.65	0*	8.07	22.35	0*	0*
	Left	14L5	Infraspinatus	234.18 ± 5.06	2.30	4.74	13.12	0.19	0.41
1	Right	14L5	Supraspinatus	239.44 ± 6.22	4.45	5.18	14.34	0.42	0.69
	Right	14L5	Infraspinatus	239.16 ± 6.54	5.41	5.02	13.91	0.54	0.78
2	Right	14L5	Supraspinatus	235.94 ± 5.71	0*	5.76	15.95	0*	0*
	Right	14L5	Infraspinatus	234.50 ± 4.67	0*	5.21	14.44	0*	0*
1	Left	UHF22	Supraspinatus	53.82 ± 13.26	13.43	8.23	22.81	0.73	0.89
	Left	UHF22	Infraspinatus	64.48 ± 14.73	7.14	13.66	37.83	0.21	0.45
2	Left	UHF22	Supraspinatus	57.62 ± 3.96	0.37	3.95	10.95	0.01	0.03
	Left	UHF22	Infraspinatus	56.29 ± 17.71	0*	19.49	53.99	0*	0*
1	Right	UHF22	Supraspinatus	56.69 ± 23.42	22.00	16.49	45.67	0.64	0.83
	Right	UHF22	Infraspinatus	75.52 ± 38.99	47.99	11.76	35.57	0.94	0.98
2	Right	UHF22	Supraspinatus	59.73 ± 29.21	23.83	22.63	62.69	0.53	0.77
	Right	UHF22	Infraspinatus	52.42 ± 18.21	16.90	12.67	35.09	0.64	0.84

RC = repeatability coefficient; ICC = intraclass correlation coefficient; SD = standard deviation.

means for the one-random effect method, a negative or zero component estimate occurs. This means the true variance equals zero.

Table 6 demonstrates the QIBA between-operator reproducibility metrics of the grayscale mean parameter. Between-operator SD values are larger than between-participant SD values for two cases, and QIBA-reproducibility SD values are larger than between-participant SD values for six cases. Consequently, the operators, interaction, and error terms introduced a larger amount of variability of grayscale mean parameter compared with the reproducibility of the iBSC parameter.

Table 6.

Operator Reproducibility Estimate of Single Grayscale Mean Measures Under Various Conditions on Human Shoulder Muscles (Supraspinatus and Infraspinatus), Calculated Using the Two-Random Effect Model.

Conditions				Summary statistics of grayscale mean
Side (left/right)	Muscle type	Transducer	Mean	Range	Between-participants SD	Between-operators SD	QIBA-Reproducibility SD	Between-image SD	RDC
Left	Supraspinatus	14L5	238.48 ± 6.94	[228.02,250.10]	4.55	0*	5.82	6.85	16.11
Left	Infraspinatus	14L5	234.55 ± 4.24	[229.35,241.90]	3.31	0.46	3.34	4.03	9.26
Right	Supraspinatus	14L5	237.56 ± 5.98	[228.06,248.88]	3.97	1.92	5.24	5.51	14.52
Right	Infraspinatus	14L5	236.83 ± 5.94	[229.14,248.10]	1.52	1.61	6.05	5.12	16.77
Left	Supraspinatus	UHF22	55.72 ± 9.54	[33.29,71.90]	5.74	0*	8.40	6.6	23.28
Left	Infraspinatus	UHF22	61.89 ± 16.60	[38.87,87.49]	6.56	7.19	16.19	16.83	44.86
Right	Supraspinatus	UHF22	58.09 ± 25.14	[33.80,113.18]	25.40	0*	16.23	19.46	44.96
Right	Infraspinatus	UHF22	63.97 ± 31.42	[30.50,117.92]	29.88	6.41	24.33	12.22	67.40

RDC = reproducibility coefficient; SD = standard deviation.

Means for the two-random effect method, negative or zero component estimate occurs.

Table 7 shows the poor reproducibility of grayscale mean measurements with between-machine SD values larger than the between-participant SD values in every case. Moreover, between-operator reproducibility is much smaller than between-machine reproducibility since the mean RDC for the between-operator test is 29.65 ± 20.44, and RDC for the between-machine test is 349.02 ± 12.66. The poor reproducibility between machines can be explained by differences in grayscale mean values between the S2000 (236.87 ± 5.86) and Vevo MD transducers (59.88 ± 21.88).

Table 7.

Machine Reproducibility Estimate of Single Grayscale Mean Measures Under Various Conditions on Human Shoulder Muscles (Supraspinatus and Infraspinatus), Calculated Using the Two-Random Effect Model.

Conditions				Summary statistics of grayscale mean
Side (left/right)	Muscle type	Operator	Mean	Range	Between-participants SD	Between-machines SD	QIBA-reproducibility SD	Between-image SD	RDC
Left	Supraspinatus	1	145.98 ± 96.79	[33.29,250.10]	0*	129.93	130.87	6.94	362.50
Left	Infraspinatus	1	151.20 ± 88.04	[47.22,241.90]	0*	118.20	118.87	9.91	329.27
Right	Supraspinatus	1	141.04 ± 96.35	[33.48,248.88]	0*	126.70	128.80	12.77	356.76
Right	Infraspinatus	1	157.34 ± 89.52	[31.19,248.10]	0*	112.44	119.12	9.04	329.95
Left	Supraspinatus	2	148.22 ± 94.80	[53.45,248.08]	0*	128.09	128.26	6.35	355.27
Left	Infraspinatus	2	145.23 ± 93.72	[38.87,241.57]	0*	125.73	126.37	14.18	350.05
Right	Supraspinatus	2	154.61 ± 93.44	[38.42,243.96]	0*	123.90	126.39	15.69	350.10
Right	Infraspinatus	2	143.46 ± 95.93	[30.50,240.08]	0*	128.40	129.33	9.69	358.25

RDC = reproducibility coefficient; SD = standard deviation.

Means for the two-random effect method, negative or zero component estimate occurs.

Repeatability of Single-Image Measures, Between-Operator and Between-Machine Reproducibility of GLCM Contrast Parameter

The statistical metrics for GLCM contrast is summarized in Table 8. The descriptive statistics indicate that the supraspinatus muscle had a slightly reduced GLCM contrast mean value than the infraspinatus muscle. Supraspinatus and infraspinatus muscles have similar standard deviation values across all individual groups. Poor repeatability was observed with GLCM contrast since the between-image SD values were generally equal to or larger than the between-participant SD. ICC(1,1) demonstrated one case as excellent, one case as good, one case as moderate, and thirteen cases as poor reliability. ICC(1,3) demonstrated had two cases as excellent, one case as good, three cases as moderate, and ten cases as poor reliability.

Table 8.

GLCM Contrast Repeatability Measurements on Shoulder Muscle.

Conditions					Summary statistics of GLCM contrast			ICC estimates of GLCM contrast
Operator group	Side (left/right)	Transducer	Muscle type	Mean	Between-participants SD	Between-image SD	RC	ICC (1,1)	ICC (1,3)
1	Left	14L5	Supraspinatus	0.22 ± 0.03	0.02	0.03	0.08	0.39	0.66
	Left	14L5	Infraspinatus	0.23 ± 0.04	0*	0.04	0.11	0*	0*
2	Left	14L5	Supraspinatus	0.20 ± 0.04	0*	0.04	0.12	0*	0*
	Left	14L5	Infraspinatus	0.22 ± 0.03	0.003	0.03	0.08	0.01	0.04
1	Right	14L5	Supraspinatus	0.21 ± 0.04	0.021	0.04	0.11	0.22	0.47
	Right	14L5	Infraspinatus	0.20 ± 0.02	0.02	0.02	0.05	0.42	0.69
2	Right	14L5	Supraspinatus	0.21 ± 0.03	0*	0.03	0.09	0*	0*
	Right	14L5	Infraspinatus	0.22 ± 0.03	0*	0.03	0.09	0*	0*
1	Left	UHF22	Supraspinatus	0.18 ± 0.07	0.004	0.002	0.12	0.71	0.88
	Left	UHF22	Infraspinatus	0.22 ± 0.04	5.8e-4	1.4e-3	0.10	0.29	0.55
2	Left	UHF22	Supraspinatus	0.14 ± 0.04	0.003	2e-4	0.04	0.94	0.97
	Left	UHF22	Infraspinatus	0.17 ± 0.05	0*	4e-3	0.18	0*	0*
1	Right	UHF22	Supraspinatus	0.17 ± 0.05	0*	0.003	0.15	0*	0*
	Right	UHF22	Infraspinatus	0.24 ± 0.06	0.063	0.032	0.089	0.84	0.94
2	Right	UHF22	Supraspinatus	0.18 ± 0.05	0*	0.003	0.15	0*	0*
	Right	UHF22	Infraspinatus	0.17 ± 0.05	0*	2.8e-3	0.15	0*	0*

RC = repeatability coefficient; ICC = intraclass correlation coefficient; SD = standard deviation.

Means for the one-random effect method, a negative or zero component estimate occurs.

Table 9 demonstrates that between-operator SD values are larger than between-participant SD values for four cases, and the QIBA reproducibility SD values are larger than between-participant SD values for seven cases, indicating that most of the variability was introduced by the operators, interactions, and error terms. Operators demonstrated poorer reproducibility for both grayscale means and GLCM contrast measurements compared with iBSC values.

Table 9.

Operator Reproducibility Estimate of Single-GLCM Contrast Measures Under Various Conditions on Human Shoulder Muscles (Supraspinatus and Infraspinatus), Calculated Using the Two-Random Effect Model.

Conditions				Summary statistics of GLCM contrast
Side (left/right)	Muscle type	Transducer	Mean	Range	Between-participants SD	Between-operators SD	QIBA-reproducibility SD	Between-Image SD	RDC
Left	Supraspinatus	14L5	0.21 ± 0.04	[0.13, 0.27]	0.0005	0.0003	0.036	0.001	0.1
Left	Infraspinatus	14L5	0.23 ± 0.03	[0.18, 0.27]	0*	0*	0.032	0.001	0.09
Right	Supraspinatus	14L5	0.21 ± 0.04	[0.16,0.25]	0.0002	0*	0.032	0.001	0.09
Right	Infraspinatus	14L5	0.21 ± 0.028	[0.16,0.27]	0*	0*	0.036	0.001	0.1
Left	Supraspinatus	UHF22	0.16 ± 0.06	[0.09,0.26]	0.060	0.026	0.039	0.087	0.11
Left	Infraspinatus	UHF22	0.19 ± 0.05	[0.11,0.30]	0.022	0.034	0.054	0.050	0.15
Right	Supraspinatus	UHF22	0.17 ± 0.05	[0.10,0.26]	0.011	0*	0.048	0.056	0.13
Right	Infraspinatus	UHF22	0.21 ± 0.06	[0.11,0.30]	0*	0.030	0.068	0.043	0.19

RDC = reproducibility coefficient; SD = standard deviation.

Means for the two-random effect method, negative or zero component estimate occurs.

Table 10 demonstrates that GLCM contrast of S2000 mean value (0.21 ± 0.03) is higher than Vevo MD mean value (0.18 ± 0.06). GLCM contrast also demonstrated poor between-machine reproducibility since the between-machine and QIBA reproducibility SD values are larger than the between-participant SD values for nearly every case, indicating the variability introduced by the different machines, the interaction between machines and participants, and measurement errors are greater than the between-participant variability. Overall, iBSC measurements exhibited more repeatable and reproducible results than both the grayscale means and GLCM contrast parameters.

Table 10.

Machine Reproducibility Estimate of Single GLCM Contrast Measures Under Various Conditions on Human Shoulder Muscles (Supraspinatus and Infraspinatus), Calculated Using the Two-Random Effect Model.

Conditions				Summary statistics of GLCM contrast
Side (left/right)	Muscle type	Operator	Mean	Range	Between-participants SD	Between-machines SD	QIBA-reproducibility SD	Between-image SD	RDC
Left	Supraspinatus	1	0.20 ± 0.05	[0.11,0.27]	0*	0*	0.071	0.032	0.20
Left	Infraspinatus	1	0.22 ± 0.04	[0.15,0.27]	0*	0*	0.035	0.032	0.10
Right	Supraspinatus	1	0.17 ± 0.05	[0.10,0.26]	0*	0.020	0.049	0.045	0.14
Right	Infraspinatus	1	0.24 ± 0.06	[0.16,0.30]	0.032	0.0095	0.046	0.032	0.13
Left	Supraspinatus	2	0.17 ± 0.05	[0.088,0.26]	0*	0.018	0.058	0.032	0.16
Left	Infraspinatus	2	0.20 ± 0.05	[0.11,0.27]	0.016	0.045	0.063	0.045	0.18
Right	Supraspinatus	2	0.18 ± 0.05	[0.11,0.24]	0*	0.013	0.046	0.045	0.13
Right	Infraspinatus	2	0.17 ± 0.05	[0.11,0.23]	0.004	0.032	0.055	0.045	0.15

RDC = reproducibility coefficient; SD = standard deviation.

Means for the two-random effect method, negative or zero component estimate occurs.

Discussion

This study investigated the repeatability and reproducibility of BSC, grayscale, and GLCM texture-based analyses in two representative participants’ healthy and degenerated rotator cuff muscles, where operators were free to adjust system settings for optimal image quality. Published data suggests that BSC-based measurements obtained using raw RF data and calibrated phantoms are system-independent,³⁴ and repeatability and reproducibility investigations have been performed in soft tissues such as a liver²⁷ and median nerve²⁸ with promising results. This is an early if not the first study to evaluate BSC using raw RF data in normal and pathologic rotator cuff musculature.

QUS, grayscale and GLCM textural analysis have been performed on skeletal muscles through grayscale^14–17 and GLCM textural¹⁸ analyses on uncalibrated B-mode images, though the repeatability and reproducibility testing of these measurements are lacking, particularly under varying conditions in accordance with guidelines endorsed by QIBA. It has been suggested that grayscale and GLCM textural analyses are much more dependent on the settings adjusted by the operators^19,35 than QUS features. However, image optimization is not only common practice, but important in musculoskeletal ultrasound, where higher frequency transducers are routinely utilized and structures can vary greatly in depth and attenuation.³⁶ Without image optimization on a per patient basis, the visualization of tissue characteristics, boundaries, and ultimately diagnoses can be severely impaired.²⁰

General Overview of BSC-Based Outcomes

The participants were chosen to represent extreme ends of the range of rotator cuff pathology typically encountered in clinical practice. MRI was used to determine muscle status, which is recognized as the gold standard,^1,37 but is much slower, less convenient and less cost-effective compared with US. On the routine B-mode images, both radiologists in this study identified the fatty degeneration of the rotator cuff muscles in the injured participant. This is consistent with prior research showing moderate accuracy (72%–85%) of US for substantial fatty atrophy of rotator cuff muscles using qualitative B-mode image evaluation.⁷ In the future, clinically evaluation of QUS compared to MRI for a sufficient sample size would be an important next step.

In our study, we found that the magnitude of iBSC of the injured participant was higher than the control, which might be explained by more densely packed scatterers in the injured muscles of injured participant.^38,39 The increased scatterers in the rotator cuff muscle may be due to fat infiltration or fibrosis after the primary injury to the tendon,⁴⁰ somewhat analogous to elevated BSC values in the fatty liver.^11,41

Repeatability and Reproducibility Metrics

Our results indicate that iBSC measurements in rotator cuff muscles are repeatable. Overall, the repeatability and reproducibility metrics (RC, RDC) are comparable to previous literature.²⁸ The good repeatability metrics demonstrate the capability of iBSC to distinguish between the healthy and injured rotator cuff muscles (particularly Goutallier grade 3 degenerated RC) with varied US settings. ICC(1,3) with a better repeatability performance suggests that future clinical application of BSC based measurements should obtain multiple repeated measurements. Moreover, the between-operators SD is comparatively low to between-participants SD. Therefore, operators did not contribute significantly to the overall variability within this study. Table 8 also demonstrates excellent between-machine reproducibility with low variability across different US platforms.

In contrast, poor repeatability for grayscale mean and GLCM contrast parameters was shown since the between-participants SD values are comparable or smaller than the between-images SD values for most groups. Most of the ICC results were classified as poor, consistent with poor repeatability of grayscale and GLCM analyses for the detection of rotator cuff muscle injury with varying operator presets.^19,42 Similarly, the QIBA reproducibility SD values are greater than the between-participant SD values, consistent with poor reproducibility between operators. These results confirm the importance of maintaining US settings if grayscale mean and GLCM contrast are to be utilized. The between-machine variability is greater than the between-participant variability, so a clear distinction between healthy and injured rotator cuff muscles could not be made. The signal treatment methods, hardware and transducer (e.g., beam shape, element size, elements interspace, and lens properties), and frequency bandwidths vary between the two US platforms. Among these sources of variability, the differences in signal treatment methods is especially important for altering the grayscale and textural outcomes.⁴²

In our study, iBSC reliability measurements of the rotator cuff muscles achieved an average ICC > 0.8, which compares favorably with other modalities and parameters. Specifically, an average ICC of 0.8 was reported for stiffness values of the supraspinatus muscle using magnetic resonance elastography,⁴³ and ICC values >0.75 was reported for shear modulus obtained with a shear wave elastography.⁴⁴ However, the grayscale means measurements (mean ICC = 0.35) and GLCM contrast measurements (mean ICC = 0.2) are not robust with varying US settings.

Limitations

First, our study only included one healthy volunteer and one patient with chronic rotator cuff disease. However, the intent of our study was to establish the repeatability and reproducibility of quantitative imaging across different operators and platforms in normal and pathological muscle; studies of this type often contain 1 to 2 volunteers.^45,46 Still, future clinical studies should be conducted on more participants to formally compare control and disease groups and determine how patient-to-patient variability (e.g., gender, age, BMI, and various disease states) affect BSC-based outcomes in various muscles. Second, our study was not exhaustive with regards to the QUS imaging and analytical methods that could have been used, plus two US scanners were employed. For instance, others have previously modeled envelope statistics using uncalibrated RF data⁴⁷ or analyzed image intensity of B-mode images normalized with a reference phantom.⁴⁸ Therefore, future studies should be performed to determine the repeatability and reproducibility of these additional methods. Third, the accuracy of the QUS metrics in this study was not determined since it requires a reference confirmation (e.g., ex vivo histological samples). However, a separate modality, MRI, was used in this study to confirm the state of the participant muscles, and the MRI exams demonstrated that both subjects were symmetric from side to side. Fourth, we fixed the GLCM offsets ( $Δ x = 0$ and $Δ y = λ / 8$ ), which define pixel relationship of varying direction and distance. However, further studies about the relationship between offset and US GLCM outcomes on muscles should be conducted. Fifth, multiple frames of reference data could be utilized to calculate the reference spectrum in order to reduce the noise of reference spectral estimate and improve the R&R metrics. However, without the multiple frames averaged for reference data, more repeatable and reproducible results were obtained for iBSC parameters compared with grayscale/textural analysis.

Conclusion

QUS BSC-based measurements using the reference phantom method demonstrate good repeatability and reproducibility for rotator cuff muscle evaluation across two different operators and two US platforms. Furthermore, because of apparent system and settings independence, BSC-based measurements demonstrate greater repeatability and reproducibility compared with grayscale and GLCM textural-based analysis. Therefore, BSC-based measurements may be preferable to grayscale and textural analyses for the evaluation of rotator cuff muscle degeneration in real clinical settings. In the meantime, BSC-based measurements require US scanner research mode to acquire RF data and additional offline processing.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We gratefully acknowledge funding from the VA R&D service (I01CX001388, I01BX005952, and I01CX002118), NIH (R01AR075825 and K01AR080257), the Department of Defense (W81XWH-20-1-0927).

ORCID iDs

Yuanshan Wu

Aiguo Han

References

Ashir

Lombardi

Jerban

Chang

EY.

Magnetic resonance imaging of the shoulder. Pol J Radiol. 2020;85:e420-e39.

Sambandam

Khanna

Gul

Mounasamy

Rotator cuff tears: an evidence based approach. World J Orthop. 2015;6(11):902-18.

Kuzel

Grindel

Papandrea

Ziegler

Fatty infiltration and rotator cuff atrophy. J Am Acad Orthop Surg. 2013;21(10):613-23.

Park

Rhee

SM.

Treatment strategy for irreparable rotator cuff tears. Clin Orthop Surg. 2018;10(2):119-34.

Chang

Chung

CB.

Current concepts on imaging diagnosis of rotator cuff disease. Semin Musculoskelet Radiol. 2014;18(4):412-24.

Singh

JP.

Shoulder ultrasound: what you need to know. Indian J Radiol Imaging. 2012;22(4):284-92.

Strobel

Hodler

Meyer

Pfirrmann

Pirkl

Zanetti

Fatty atrophy of supraspinatus and infraspinatus muscles: accuracy of US. Radiology. 2005;237(2):584-9.

Ghoshal

Oelze

O’Brien

WD.

Quantitative ultrasound history and successes. Dordrecht: Springer; 2013, pp.21-42.

Insana

Zagzebski

Madsen

Improvements in the spectral difference method for measuring ultrasonic attenuation. Ultrason Imaging. 1983;5(4):331-45.

10.

Lizzi

Ostromogilsky

Feleppa

Rorke

Yaremko

MM.

Relationship of ultrasonic spectral parameters to features of tissue microstructure. IEEE Trans Ultrason Ferroelectr Freq Control. 1987;34(3):319-29.

11.

Han

Erdman

Simpson

Andre

O’Brien

. (eds) Early detection of fatty liver disease in mice via quantitative ultrasound. In: 2014 IEEE International Ultrasonics Symposium, 2014. New York, NY: IEEE.

12.

Alnazer

Falou

Nasr

Azar

Hysi

Wirtzfeld

, et al. Quantitative ultrasound imaging for the differentiation between fresh and decellularized mouse kidneys. Annu Int Conf IEEE Eng Med Biol Soc. 2019;2019:6624-7.

13.

Rohrbach

Wodlinger

Wen

Mamou

Feleppa

High-frequency quantitative ultrasound for Imaging Prostate Cancer using a novel micro-ultrasound Scanner. Ultrasound Med Biol. 2018;44(7):1341-54.

14.

Harris-Love

Seamon

Teixeira

Ismail

Ultrasound estimates of muscle quality in older adults: reliability and comparison of Photoshop and ImageJ for the grayscale analysis of muscle echogenicity. PeerJ. 2016;4: e1721. doi: 10.7717/peerj.1721.

15.

Nielsen

Jensen

Darvann

Jørgensen

Bakke

Quantitative ultrasound tissue characterization in shoulder and thigh muscles–a new approach. BMC Musculoskelet Disord. 2006;7:2.

16.

Sarto

Spörri

Fitze

Quinlan

Narici

Franchi

MV.

Implementing ultrasound imaging for the assessment of muscle and tendon properties in elite sports: Practical Aspects, methodological considerations and Future Directions. Sports Med. 2021;51(6):1151-70.

17.

Stock

Thompson

BJ.

Echo intensity as an indicator of skeletal muscle quality: applications, methodology, and future directions. Eur J Appl Physiol. 2021;121(2):369-80.

18.

Wilkinson

Ashman

Baker

Watson

Smith

AC.

Quantitative muscle ultrasonography using 2D textural analysis: a novel approach to assess skeletal muscle structure and quality in chronic kidney disease. Ultrason Imaging. 2021;43(3):139-48.

19.

Vara

Rustici

Sechi

Mosconi

Lucidi

Golfieri

Texture analysis on ultrasound: the effect of time gain compensation on histogram metrics and gray-level matrices. J Med Phys. 2020;45(4):249-55.

20.

Zander

Hüske

Hoffmann

Cui

X-W

Dong

Lim

, et al. Ultrasound image optimization (“Knobology”): B-Mode. Ultrasound Int Open. 2020;6(1):E14-24.

21.

Sullivan

Obuchowski

Kessler

Raunig

Gatsonis

Huang

, et al.; RSNA-QIBA Metrology Working Group. Metrology standards for Quantitative Imaging Biomarkers. Radiology. 2015;277(3):813-25.

22.

Lim

Huang

Statistical considerations for repeatability and reproducibility of quantitative imaging biomarkers. BJR Open. 2022; August 22 4(1): 20210083. doi: 10.1259/bjro.20210083.

23.

Somerson

Hsu

Gorbaty

Gee

AO.

Classifications in brief: Goutallier classification of fatty infiltration of the rotator cuff musculature. Clin Orthop Relat Res. 2016;474(5):1328-32.

24.

Tsuchiya

Bois

Matthewson

Oiwa

IKY

. The relationship between preoperative Goutallier stage and retear rates following posterosuperior rotator cuff repair: a systematic review. J Shoulder Elbow Surg. 2023;32(2):435-43.

25.

Khair

Lehman

Tsouris

Gulotta

LV.

A systematic review of preoperative fatty infiltration and rotator cuff outcomes. HSS J. 2016;12(2):170-6.

26.

Nordberg

Hall

TJ.

Effective scatterer diameter estimates for broad scatterer size distributions. Ultrason Imaging. 2015;37(1):3-21.

27.

Han

Andre

Deiranieh

Housman

Erdman

Loomba

, et al. Repeatability and reproducibility of the ultrasonic attenuation coefficient and backscatter coefficient measured in the right lobe of the liver in adults with known or suspected nonalcoholic fatty liver disease. J Ultrasound Med. 2018;37(8):1913-27.

28.

Barrere

Han

Chang

Andre

Shah

SB.

Repeatability, reproducibility and sources of variability in the assessment of backscatter coefficient and texture parameters from high-frequency ultrasound acquisitions in human median nerve. Ultrasound Med Biol. 2023;49:122-35.

29.

Moran

Bush

Bamber

JC.

Ultrasonic propagation properties of excised human skin. Ultrasound Med Biol. 1995;21(9):1177-90.

30.

Glueck

Mottley

Sobel

Miller

Pérez

JE.

Changes in ultrasonic attenuation and backscatter of muscle with state of contraction. Ultrasound Med Biol. 1985;11(4):605-10.

31.

Greenleaf

JF.

Tissue Characterization With Ultrasound. Boca Raton, FL: CRC Press; 1986.

32.

Haralick

Shanmugam

Dinstein

Textural features for image classification. IEEE Trans Syst Man Cybern. 1973;SMC-3(6):610-21.

33.

Koo

MY.

A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med. 2016;15(2):155-63.

34.

Boote

Zagzebski

Madsen

Hall

TJ.

Instrument-independent acoustic backscatter coefficient imaging. Ultrason Imaging. 1988;10(2):121-38.

35.

Alqahtani

Coleman

Pugh

Nokes

LDM

. (eds) Tissue characterization: Influence of ultrasound setting on texture features in vivo. In: 2010 International Conference of Medical Image Analysis and Clinical Application, 10–13 June 2010.

36.

Jacobson

. Chapter 1 - Introduction. In: Jacobson

, ed. Fundamentals of Musculoskeletal Ultrasound. 2nd ed. Philadelphia, PA: W.B. Saunders; 2013, pp.1-e15.

37.

Yazigi Junior

Anauate Nicolao

Archetti Netto

Matsunaga

Lim Lee

Torres Ogata

, et al. Magnetic resonance imaging reproducibility for rotator cuff partial tears in patients up to 60 years. BMC Musculoskelet Disord. 2019;20(1):383.

38.

Kawasaki

An integrated backscatter ultrasound technique for the detection of coronary and carotid atherosclerotic lesions. Sensors. 2015;15(1):979-94.

39.

Lizzi

Astor

Feleppa

Shao

Kalisz

Statistical framework for ultrasonic spectral parameter imaging. Ultrasound Med Biol. 1997;23(9):1371-82.

40.

Osti

Buda

Del Buono

Fatty infiltration of the shoulder: diagnosis and reversibility. Muscles Ligaments Tendons J. 2013;3(4):351-4.

41.

Wear

Han

Rubin

Gao

Lavarello

Cloutier

, et al. US backscatter for liver fat quantification: an AIUM-RSNA QIBA pulse-echo quantitative ultrasound initiative. Radiology. 2022;305(3):526-37.

42.

Steffel

Brown

Korcarz

Varghese

Stein

Wilbrand

, et al. Influence of ultrasound system and gain on grayscale median values. J Ultrasound Med. 2019;38(2):307-19.

43.

Ito

Numano

Ueki

Habe

Maeno

Takamoto

, et al. Magnetic resonance elastography of the supraspinatus muscle: a preliminary study on test-retest repeatability and wave quality with different frequencies and image filtering. Magn Reson Imaging. 2020;71:27-36.

44.

Sakaki

Taniguchi

Katayose

Kura

Okamura

Effects of shoulder abduction on the stiffness of supraspinatus muscle regions in rotator cuff tear. Clin Anat. 2022;35(1):94-102.

45.

Fox

Sakaie

Lee

Debbins

Liu

Arnold

, et al. A validation study of multicenter diffusion tensor imaging:reliability of fractional anisotropy and diffusivity values. AJNR Am J Neuroradiol. 2012;33(4):695-700.

46.

Prohl

Scherrer

Tomas-Fernandez

Filip-Dhima

Kapur

Velasco-Annis

, et al. Reproducibility of structural and diffusion tensor imaging in the TACERN multi-center Study. Front Integr Neurosci. 2019;13:24.

47.

Sikdar

Diao

Turo

Stanley

Sharma

Chambliss

, et al. Quantification of muscle tissue properties by modeling the statistics of ultrasound image intensities using a mixture of gamma distributions in children with and without cerebral palsy. J Ultrasound Med. 2018;37(9):2157-69.

48.

Zaidman

Holland

Anderson

Pestronk

Calibrated quantitative ultrasound imaging of skeletal muscle using backscatter analysis. Muscle Nerve. 2008;38(1):893-8.

High-frequency Quantitative Ultrasound Imaging of Human Rotator Cuff Muscles: Assessment of Repeatability and Reproducibility

Abstract

Keywords

Introduction

Materials and Methods

Study Design and Ultrasound Acquisition

Analytical Method

QUS Analysis Methods

Image GLCM Texture Analysis

Grayscale Analysis

ROI Selection

Statistical Analysis

Results

B-Mode, BSC, and GLCM Textural Measurement Results

Repeatability of Single-Image Measures, Between-Operator and Between-Machine Reproducibility of iBSC Parameter

Repeatability of Single-Image Measures, Between-Operator and Between-Machine Reproducibility of Grayscale Mean Intensity Parameter

Repeatability of Single-Image Measures, Between-Operator and Between-Machine Reproducibility of GLCM Contrast Parameter

Discussion

General Overview of BSC-Based Outcomes

Repeatability and Reproducibility Metrics

Limitations

Conclusion

Footnotes

Declaration of Conflicting Interests

Funding

ORCID iDs

References