Objective Ultrasound (US) is a more accessible alternative to MRI for rotator cuff tear (RCT) assessment. Rapid US can identify RCTs, but accurate US interpretation remains challenging, even for experts. We performed a retrospective cross-sectional study to evaluate intraobserver and interobserver agreement and accuracy in RCT detection with US between expert and non-expert readers. Materials and methods Long-axis cine sweeps of supraspinatus tendon from 138 patients with suspected RCTs were analyzed and compared to corresponding MRI results (reference standard). Three non-experts (graduate student, medical student, radiology resident) and three experts (fellowship-trained musculoskeletal (MSK) radiologist, MSK sonographer, and radiology fellow) independently classified tendons as normal, full-thickness tear, or partial-thickness tear by predefined criteria. Results For all-tear detection (ATD), expert accuracy (%) ranged from 68.8 (95% CI, 61.1–76.5) to 84.8% (78.8–90.9), and non-expert accuracy from 63.0 (54.9–71.1) to 79.7% (73.0–86.4). For full-thickness tear detection (FTTD), expert accuracy ranged from 75.4 (68.2–82.6) to 87.0% (81.4–92.6), and non-expert accuracy from 73.9 (66.6–81.2) to 79.0% (72.2–85.8). Interobserver agreement for ATD was moderate among experts (Light’s kappa (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">$\kappa $</EquationSource> <EquationSource Format="MATHML"><math> <mi>κ</mi> </math></EquationSource> </InlineEquation>) = 0.42 (95% CI, 0.30–0.58)) and fair among non-experts (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">$\kappa $</EquationSource> <EquationSource Format="MATHML"><math> <mi>κ</mi> </math></EquationSource> </InlineEquation> = 0.38 (0.32–0.50)). Intraobserver agreement was fair (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">$\kappa $</EquationSource> <EquationSource Format="MATHML"><math> <mi>κ</mi> </math></EquationSource> </InlineEquation> = 0.35 (0.29–0.43)). Our experts achieved higher sensitivity, resulting in improved overall accuracy. Conclusion Ultrasound readers of all experience levels accurately identified over 80% of RCTs when compared with MRI. Expert readers achieved higher accuracy, sensitivity, and reliability than non-experts. Ultrasound reliably detects RCTs and remains the preferred imaging modality for its accessibility and patient comfort, underscoring the need for standardization, training, and technological improvements.

Evaluating diagnostic consistency: intra and interobserver agreement in rotator cuff tear classification using ultrasound videos

期刊论文
发表日期：

2025年8月28日

查看全文

Shrimanti Ghosh,
Jessica Knight,
Stephanie Wichuk,
Natasha Akhlaq,
Daniel Durham,
Cassandra Gallant,
Steel McDonald,
Michael Xie,
Vedur Verma,
Vincent Man,
Abhilash R. Hareendranathan,
Jacob L. Jaremko

摘要

Objective

Ultrasound (US) is a more accessible alternative to MRI for rotator cuff tear (RCT) assessment. Rapid US can identify RCTs, but accurate US interpretation remains challenging, even for experts. We performed a retrospective cross-sectional study to evaluate intraobserver and interobserver agreement and accuracy in RCT detection with US between expert and non-expert readers.

Materials and methods

Long-axis cine sweeps of supraspinatus tendon from 138 patients with suspected RCTs were analyzed and compared to corresponding MRI results (reference standard). Three non-experts (graduate student, medical student, radiology resident) and three experts (fellowship-trained musculoskeletal (MSK) radiologist, MSK sonographer, and radiology fellow) independently classified tendons as normal, full-thickness tear, or partial-thickness tear by predefined criteria.

Results

For all-tear detection (ATD), expert accuracy (%) ranged from 68.8 (95% CI, 61.1–76.5) to 84.8% (78.8–90.9), and non-expert accuracy from 63.0 (54.9–71.1) to 79.7% (73.0–86.4). For full-thickness tear detection (FTTD), expert accuracy ranged from 75.4 (68.2–82.6) to 87.0% (81.4–92.6), and non-expert accuracy from 73.9 (66.6–81.2) to 79.0% (72.2–85.8). Interobserver agreement for ATD was moderate among experts (Light’s kappa ( $\kappa $ $κ$ ) = 0.42 (95% CI, 0.30–0.58)) and fair among non-experts ( $\kappa $ $κ$ = 0.38 (0.32–0.50)). Intraobserver agreement was fair ( $\kappa $ $κ$ = 0.35 (0.29–0.43)). Our experts achieved higher sensitivity, resulting in improved overall accuracy.

Conclusion

Ultrasound readers of all experience levels accurately identified over 80% of RCTs when compared with MRI. Expert readers achieved higher accuracy, sensitivity, and reliability than non-experts. Ultrasound reliably detects RCTs and remains the preferred imaging modality for its accessibility and patient comfort, underscoring the need for standardization, training, and technological improvements.