Diagnostic imaging of rotator cuff injuries. A systematic review of diagnostic accuracy and reliability

 

 

Nº de DOI: 10.34896/RSI.2026.49.79.002

 

 

AUTHORS

1. José Vinicio Cóndor Chano. General Practitioner. Affiliated with Hospital Carlos Andrade Marín. Graduate of Universidad UTE. Based in Quito, Ecuador. https://orcid.org/0009-0000-0402-6847
2. Karla Gissel Ipiales Iles. General Practitioner. Affiliated with Clínicas Privadas del Ecuador. Graduate of Universidad UTE. Based in Ibarra, Ecuador. https://orcid.org/0009-0009-1779-5413
3. Claudia Dayana Barragán Silva. General Practitioner with a Master’s Degree in Public Health with a concentration in Primary Health Care. Affiliated with Clínicas Privadas del Ecuador. Graduate of Universidad Nacional de Chimborazo. Based in Riobamba, Ecuador. https://orcid.org/0009-0007-7910-4505
4. Victor Andrés Veloz Cárdenas. General Practitioner. Affiliated with Hospital General Latacunga. Graduate of Universidad Regional Autónoma de los Andes. Based in Latacunga, Ecuador. https://orcid.org/0009-0007-6809-7311
5. Margareth Lisbeth Sánchez Castro. General Practitioner. Affiliated with Instituto Ecuatoriano de Seguridad Social. Graduate of Universidad San Gregorio de Portoviejo. Based in Guayaquil, Ecuador. https://orcid.org/0000-0002-4944-9357

 

ABSTRACT

Objective: To synthesise published evidence on the diagnostic accuracy and reliability of plain radiography, US, MRI, MRA, computed tomography (CT) and artificial-intelligence-assisted interpretation for rotator cuff tears and their muscular consequences.

Methodology: Systematic review of diagnostic accuracy evidence, structured according to the PRISMA-DTA statement. Journal articles with a verifiable digital object identifier were identified through web-based searches of indexed biomedical literature on 18 July 2026 and through reference-list searching. A total of 51 articles were retained, of which 10 were systematic reviews or meta-analyses. Pooled estimates are reported as published by the original authors; no new pooling was undertaken.

Results: For full-thickness tears, US, MRI and MRA each achieved pooled sensitivity and specificity above 0.90 in one large meta-analysis, and US performance was similar whether the examiner was a radiologist, a sonographer or an orthopaedic surgeon. For partial-thickness tears, pooled US sensitivity and specificity were 0.84 and 0.89. One meta-analysis of 65 articles ranked MRA highest, whereas a Cochrane review found no significant difference between MRI and US and could not pool MRA data for partial tears. For bursal-sided partial-thickness tears, MRA and MRI had similar value. Plain radiographic measurement of the acromiohumeral interval was highly specific but insensitive. Semi-quantitative grading of fatty infiltration showed only fair to good reliability, and quantitative alternatives were more reproducible. Deep-learning models applied to MRI reported accuracies from 71.0% to 100%, with no significant difference from human experts in the studies that compared them.

Conclusions: US and MRI are broadly comparable for full-thickness tears, MRA may add value for selected partial-thickness lesions, and radiography remains a useful first step. The main weaknesses of the evidence are retrospective design, surgically enriched populations, heterogeneity and limited direct comparison of all modalities in the same patients. Prospective multimodality studies with standardised reference standards are needed.

KEY WORDS

Rotator cuff injuries, diagnostic imaging, ultrasonography, magnetic resonance imaging, magnetic resonance arthrography, computed tomography, sensitivity and specificity, artificial intelligence.

RESUMEN

Objetivo: Sintetizar la evidencia publicada sobre la precisión diagnóstica y la fiabilidad de la radiografía convencional, la ecografía, la RM, la angiorresonancia (ARM), la tomografía computarizada (TC) y la interpretación asistida por inteligencia artificial para las roturas del manguito rotador y sus consecuencias musculares.

Metodología: Revisión sistemática de la evidencia sobre precisión diagnóstica, estructurada según la declaración PRISMA-DTA. Se identificaron artículos de revistas con un identificador de objeto digital (DOI) verificable mediante búsquedas en la web de literatura biomédica indexada el 18 de julio de 2026 y mediante la revisión de listas de referencias. Se seleccionaron 51 artículos en total, de los cuales 10 eran revisiones sistemáticas o metaanálisis. Las estimaciones combinadas se presentan tal como fueron publicadas por los autores originales; no se realizó ninguna nueva combinación de datos.

Resultados: Para las roturas de espesor completo, la ecografía, la RM y la ARM alcanzaron, cada una, una sensibilidad y especificidad combinadas superiores a 0,90 en un metaanálisis de gran tamaño; además, el rendimiento de la ecografía fue similar independientemente de si el examinador era radiólogo, ecografista o cirujano ortopédico. Para las roturas de espesor parcial, la sensibilidad y especificidad combinadas de la ecografía fueron de 0,84 y 0,89, respectivamente. Un metaanálisis de 65 artículos situó a la ARM como la técnica con mejores resultados, mientras que una revisión Cochrane no halló diferencias significativas entre la RM y la ecografía, y no pudo combinar los datos de ARM para las roturas parciales. En el caso de las roturas de espesor parcial en la cara bursal, la ARM y la RM mostraron un valor similar. La medición radiográfica convencional del intervalo acromiohumeral resultó altamente específica pero poco sensible. La gradación semicuantitativa de la infiltración grasa mostró una fiabilidad entre aceptable y buena, mientras que las alternativas cuantitativas resultaron más reproducibles. Los modelos de aprendizaje profundo aplicados a la RM reportaron precisiones de entre el 71,0 % y el 100 %, sin diferencias significativas respecto a los expertos humanos en los estudios que realizaron comparaciones directas.

Conclusiones: La ecografía y la RM son ampliamente comparables para las roturas de espesor completo; la ARM puede aportar valor en lesiones seleccionadas de espesor parcial, y la radiografía sigue siendo un primer paso útil. Las principales debilidades de la evidencia son el diseño retrospectivo, las poblaciones con una prevalencia elevada de casos quirúrgicos, la heterogeneidad y la limitada comparación directa de todas las modalidades en los mismos pacientes. Se requieren estudios prospectivos multimodales que utilicen patrones de referencia estandarizados.

PALABRAS CLAVE

Lesiones del manguito rotador, diagnóstico por imagen, ecografía, resonancia magnética, artrorresonancia magnética, tomografía computarizada, sensibilidad y especificidad, inteligencia artificial.

INTRODUCTION

Rotator cuff injuries are among the most frequent causes of shoulder pain and functional loss, and the imaging pathway chosen for a patient influences diagnosis, surgical planning and postoperative surveillance. Several meta-analyses have compared ultrasonography (US), magnetic resonance imaging (MRI) and magnetic resonance arthrography (MRA), but their conclusions are not uniform, and direct head-to-head evidence remains scarce.

OBJECTIVE

The primary objective of this review was to synthesise published evidence on the diagnostic accuracy of radiography, US, MRI, MRA and CT or CT arthrography for the detection and characterisation of full-thickness and partial-thickness rotator cuff tears, using arthroscopic or open surgical findings as the preferred reference standard. The secondary objectives were: (1) to summarise the reliability of imaging-based assessment of muscle atrophy and fatty infiltration; (2) to describe the role of imaging after rotator cuff repair; (3) to summarise the emerging evidence on artificial-intelligence-assisted interpretation; and (4) to propose, from the evidence gathered, a pragmatic framework for modality selection.

METHODOLOGY

Study design and reporting framework:

This work is a systematic review of diagnostic test accuracy evidence. Its reporting was structured according to the Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) statement⁵. Because many of the included records were themselves systematic reviews or meta-analyses that had already pooled data, the present review does not perform a new meta-analysis. Instead it provides a structured qualitative synthesis in which pooled estimates are reproduced as published and interpreted in context. The protocol was not prospectively registered.

Eligibility criteria:

Records were eligible if they met all of the following criteria: (1) a full article in a peer-reviewed journal with a digital object identifier (DOI); (2) a population of patients with suspected or confirmed rotator cuff pathology; (3) evaluation of at least one imaging modality (radiography, US, MRI, MRA, CT or CT arthrography, or computer-assisted interpretation of these); and (4) reporting of diagnostic accuracy, reliability or the clinical performance of imaging after repair. Systematic reviews and meta-analyses, diagnostic accuracy studies, reliability studies and methodological guidance relevant to diagnostic reviews were all admissible. Books, book chapters, conference abstracts, preprints and records without a verifiable DOI were excluded, as were studies of animals or cadavers.

Information sources and search strategy:

Searches were run on 18 July 2026 through a web-based search tool that indexes journal publisher sites, PubMed Central and institutional repositories. The search combined terms for the condition («rotator cuff», «rotator cuff tear», «supraspinatus», «subscapularis»), the index tests («ultrasonography», «sonography», «magnetic resonance imaging», «MR arthrography», «computed tomography», «radiograph», «acromiohumeral») and the outcome or design («diagnostic accuracy», «sensitivity and specificity», «reliability», «meta-analysis», «systematic review», «Goutallier», «fatty infiltration», «deep learning»). Priority was given to published systematic reviews and meta-analyses, and the reference lists of the retrieved syntheses were then searched by hand to locate key primary studies and classic methodological papers. No language filter was intended, although all retained records were available in English.

Selection and verification:

Titles and abstracts were screened against the eligibility criteria and full-text pages or reference-list entries were consulted for retained records. For every retained record, the title, author list, journal, year, volume, page range and DOI were checked against a publisher page, a repository page or an authoritative reference list in another retrieved article. Records whose bibliographic details or DOI could not be confirmed were not used. Screening and extraction were performed by a single reviewer, and no independent duplicate screening was carried out.

Data extraction:

For each record the following items were extracted: study design, sample size, population characteristics, index test, reference standard, number and experience of readers, main diagnostic metrics (sensitivity, specificity, accuracy, area under the curve), reliability metrics (kappa and intraclass correlation coefficients) and the principal conclusion. Where a secondary study reported a statistic drawn from another article, the secondary source was cited and the statistic was attributed to it.

Appraisal of methodological quality:

The quality of primary diagnostic studies is conventionally appraised with QUADAS or its revision QUADAS-2⁶. In this review, quality was not re-scored for each primary study. Instead, the appraisals reported by the included systematic reviews were taken into account when they were available, and design-level threats to validity (retrospective design, absence of reader blinding, single readers, verification bias and spectrum bias) were noted when the source reported them.

Synthesis:

Findings were organised by imaging modality and by clinical question and are presented in narrative form and in tables. Pooled estimates are reported exactly as published by the original authors, with the review in which they appeared cited alongside. No new statistical pooling, heterogeneity testing or formal grading of the certainty of evidence was undertaken.

Records retained:

A total of 51 journal articles were retained and form the reference list. They comprised 10 systematic reviews or meta-analyses, 2 narrative or methodological reviews, 2 reporting or quality-assessment guidance papers and 37 primary diagnostic accuracy, reliability or technical studies. The publication dates range from 1991 to 2025. Because the search was run as a web-based search with reference-list expansion rather than as a series of database exports, a conventional PRISMA flow diagram with record counts at each stage is not reported, and this is acknowledged as a limitation in the Discussion.

RESULTS

Characteristics of the evidence base:

The retained evidence spans more than three decades, from an early prospective comparison of MRI, CT arthrography and US with operative findings to recent systematic reviews of acromiohumeral distance and of artificial intelligence. Most primary studies compared the index test with arthroscopic or open surgical findings. In the meta-analysis of bursal-sided partial-thickness tears, for example, 11 of the 12 included studies were retrospective and 11 used shoulder arthroscopy as the reference standard^1,2,7,8,9. In the Cochrane review, the median prevalence of any rotator cuff tear was 80%, which illustrates how strongly the underlying populations were enriched by surgical referral⁴. Table 1 summarises the pooled findings reported by the principal syntheses.

Source: Prepared by the authors.

Plain radiography:

Plain radiographs have a limited direct role in detecting a tendon defect, and clinical examination together with plain film has been described as insufficient for diagnosing cuff tears¹. Their value lies in indirect signs and in excluding other causes of pain. Acromial morphology is one example: acromial spurs have been classified into six types according to their shape, and the heel-type spur has been proposed as a possible risk factor for rotator cuff tears¹⁴.

The most studied radiographic sign is narrowing of the acromiohumeral interval, which reflects superior migration of the humeral head. In a prospective study of 86 shoulders divided into non-full-thickness tears, full-thickness tears up to 3 cm and full-thickness tears larger than 3 cm, an upright interval of 7.0 mm or less had a sensitivity of 27.9% and a specificity of 100% for full-thickness tears, with an accuracy of 64%¹⁵. A supine interval of 6.5 mm or less had a sensitivity of 32.6%, a specificity of 100% and an accuracy of 66.3%. The interval measured on supine radiographs and on MRI was significantly lower than on upright radiographs, by 1.34 to 1.37 mm and 1.62 to 1.87 mm respectively¹⁵. The authors concluded that an upright interval of 7 mm or less is appropriate for ruling in full-thickness tears but that the value was not relevant as a cut-off for supine radiographs or MRI.

A comparison of five methods of measuring craniocaudal humeral position on conventional radiographs of 280 subjects found that the acromiohumeral interval and the upward migration index had the highest discriminative accuracy for detecting a cuff tear, and the authors recommended either as an indirect measure¹⁶. A systematic review of acromiohumeral distance as a diagnostic and prognostic biomarker found that reduced values were frequently associated with full-thickness tears, whereas larger values typically characterised asymptomatic individuals, but it also documented substantial heterogeneity in imaging protocols, measurement definitions and diagnostic thresholds⁸.

Ultrasonography:

US has been the most extensively meta-analysed modality. A review of 62 studies including 6007 patients and 6066 shoulders, all of which compared US with arthroscopic or open surgical findings, reported a pooled sensitivity of 0.96 and a specificity of 0.93 for full-thickness tears, and 0.84 and 0.89 respectively for partial-thickness tears¹⁰. The authors concluded that US is an appropriate technique for assessing cuff tears and that its accuracy is superior for full-thickness than for partial-thickness lesions. A more recent review restricted to studies using arthroscopy as the reference standard reported a median accuracy of 0.93 for full-thickness supraspinatus tears and 0.81 for partial-thickness tears¹³. The same review included a meta-analysis of five studies that compared US and MRI directly and found no statistically significant difference in sensitivity, specificity or accuracy for supraspinatus tears of any thickness (P = 0.31 to 0.55), full-thickness tears (P = 0.63 to 0.97) or partial-thickness tears (P = 0.13 to 0.81) [13]. It also noted that an earlier review had reported a pooled US sensitivity of 0.95 and specificity of 0.72 for tears of any size, and that this earlier work had been limited by small size and by mixing arthroscopy and MRI as reference standards¹³.

Operator dependence is a recurrent concern for US. In a meta-analysis that included a secondary analysis by examiner type, the accuracy of US was similar whether the examination was performed by a trained radiologist, a sonographer or an orthopaedist¹¹. Taking accuracy, cost and safety together, the authors of that review considered US the best option¹¹. Another review qualified the conclusion by noting that US shows statistically equivalent capability to MRI for experienced operators, which suggests that training and experience remain relevant modifiers¹³.

Three-dimensional ultrasound has also been examined. A meta-analysis reported that three-dimensional shoulder US is highly accurate for full-thickness tears but may lack accuracy for partial-thickness tears¹⁷. A study in a community hospital compared two-dimensional and three-dimensional shoulder US with MRI for supraspinatus tears, with attention to room efficiency¹⁸. Earlier primary studies compared US with MRI and with arthroscopy in 71 consecutive cases, with MRI and surgery in symptomatic patients, and with MRI in a prospective series of 77 consecutive patients with a surgical reference standard^19,20,21. Taken together, these studies underpin the meta-analytic conclusion that US and MRI are comparable for the detection of tears, particularly when the reference standard is surgical.

Magnetic resonance imaging:

A meta-analysis of 18 studies of MRI alone reported pooled sensitivity and specificity of 0.93 and 0.88 for tears of any type and found that diagnostic accuracy for full-thickness tears was significantly better than for partial-thickness tears¹². In the comparative meta-analysis of 65 articles, MRI and US did not differ significantly in sensitivity or specificity for either tear category, and the summary area under the curve of MRI was 0.878³. The Cochrane review likewise found no significant difference between MRI and US for any tear or for partial-thickness tears⁴.

Technical factors appear to matter. In the meta-analysis of bursal-sided partial-thickness tears, meta-regression showed that publication year and magnetic field strength accounted for heterogeneity in the specificity of MRI¹. Primary studies at 3.0 T have asked whether contrast injection is needed. One study of 3-T shoulder MRI explicitly addressed the question of whether MRA is necessary, and another compared non-contrast MRI with indirect MRA for rotator cuff tears in a cohort of 333 shoulders^22,23. Diffusion-weighted imaging has been explored as a quantitative adjunct for partial-thickness tears in a series of 146 patients²⁴. MRI also shows secondary findings: in 238 patients, cystic changes at the supraspinatus and infraspinatus insertion sites were associated with age and with rotator cuff disorders²⁵.

Magnetic resonance arthrography and CT arthrography:

The meta-analysis of 65 articles reported that MRA was more sensitive and specific than MRI or US for both full-thickness and partial-thickness tears (P < 0.05), with the highest summary area under the curve (0.935)³. By contrast, for bursal-sided partial-thickness tears the pooled sensitivity of MRA and MRI was identical at 0.77, with specificities of 0.98 and 0.96, areas under the curve of 0.88 and 0.82, and diagnostic odds ratios of 73.01 and 37.12 respectively, and the authors concluded that the two techniques have similar diagnostic value¹. Within that meta-analysis, false-positive MRA findings were attributed to inflamed tendon and false-negative findings to failure of contrast to pass into the bursa, and the authors noted that MRA is an invasive procedure with a longer examination time and a risk of infection and other complications from the injection¹. A review of direct MR arthrography describes the technique and its current use²⁶.

Technique has evolved. Three-dimensional isotropic T1-weighted fast spin-echo MRA was compared with two-dimensional MRA in 49 patients, and three-dimensional isotropic indirect MRA was compared with two-dimensional conventional indirect MRA for the diagnosis of rotator cuff tears in 205 patients^27,28. A prospective study compared indirect isotropic and conventional MRA for labral lesions and cuff tears²⁹. MRA has been used to study particular lesion patterns: delaminated tears were examined for prevalence, characteristics and diagnostic accuracy using indirect MRA in 231 patients, and various partial-thickness tears were compared on MRA with arthroscopic correlation in 202 patients^33,31. A large series of 275 patients assessed cuff tears with MRA and arthroscopic correlation³². For the subscapularis tendon, detection and grading at MRA was described in an early radiological study, and the injection approach has been compared at 3.0 T, with anterior trans-subscapularis and posterior routes tested for subscapularis tears^33,34.

The Cochrane review could not pool MRA data for the detection of any cuff tear or of partial-thickness tears, and it found that no study had evaluated MRA, MRI and US in the same population⁴. Consequently, the apparent superiority of MRA in older indirect comparisons rests on cross-study comparisons, which are exposed to differences in populations and technique.

CT and CT arthrography occupy a smaller place in recent diagnostic literature. An early prospective study compared MRI, CT arthrography and US with operative findings in the painful shoulder⁷. In more recent practice, CT arthrography has been used mainly for postoperative evaluation. In one study of 75 patients with full-thickness tears, 16-detector CT arthrography was used after repair, and the authors justified the choice by pointing out that it was cheaper than MRI, was not examiner-dependent in the way that US is, and avoided metal artefact from anchors that limits MRI near the repair³⁵.

Muscle quality: atrophy and fatty infiltration:

The Goutallier classification grades fatty infiltration from 0 (no fat) through 1 (some fatty streaks), 2 (less fat than muscle), 3 (as much fat as muscle) to 4 (more fat than muscle)³⁶. It was introduced for axial CT and subsequently adapted to MRI³⁷. Greater fatty infiltration has been associated with worse anatomical and functional outcomes of repair^38,39.

The reliability of the grading is a recurring theme. In the original comparison of CT and MRI, kappa values of 0.68 to 0.83 on CT and 0.82 to 0.93 on MRI were reported for the grading, according to a later study that cited that work [35]. That later study of 75 full-thickness tears, using five readers (two musculoskeletal radiologists and three orthopaedic surgeons), found interobserver intraclass correlation coefficients of 0.6 to 0.72 on MRA and 0.43 to 0.6 on CT arthrography; the coefficients were higher among radiologists (0.58 to 0.78) than among surgeons (0.32 to 0.68), and intraobserver values ranged from 0.26 to 0.81³⁵. The simplified three-stage system did not improve reliability. The authors cautioned that published data on fatty degeneration should be interpreted carefully because of the relatively low reliability of the grading³⁵.

Validity is also questionable. When five shoulder surgeons graded the MRI scans of 42 patients with cuff tears, the correlation between the Goutallier grade and the fat-to-water ratio measured by MR spectroscopy was weak (R = 0.35, p < 0.05)⁴⁰. Quantitative alternatives have therefore been proposed. Proton MR spectroscopy was used to assess supraspinatus fat content in volunteers and patients⁴¹. A CT method that measured mean muscle density in Hounsfield units was reported to have an intraclass correlation coefficient of 0.98, compared with 0.63 for visual rating⁴². US has been evaluated for fatty atrophy of the supraspinatus and infraspinatus, and extended-field-of-view sonography was tested for its effect on interrater reliability in the detection of muscle atrophy^43,44.

The choice of slice also matters. The most commonly used sagittal section is the scapular Y view, which was described for atrophy assessment⁴⁵. A 2025 study of 91 patients with tears compared the Y view with a more medial section at the suprascapular notch (the r view) ³⁶. In the r view, interobserver and intraobserver coefficients were at least 0.80 in all three retraction groups, whereas reliability in the Y view declined as retraction increased, and in patients with the most severe retraction the Y view gave significantly higher fat-infiltration grades than the r view (P < 0.01) ³⁶. The authors acknowledged an imbalance between sexes, small subgroups of about 30 patients, an MRI-only design and the absence of surgical or biopsy confirmation³⁶.

Fatty infiltration also behaves differently across the cuff after surgery. A study of MRI after arthroscopic supraspinatus repair found that fat infiltration of the supraspinatus and infraspinatus increased significantly after surgery regardless of tendon integrity, whereas no significant progression was observed in the subscapularis⁴⁶.

Imaging after rotator cuff repair:

Postoperative imaging asks a different question, namely whether the repaired tendon is intact. MRI was used to correlate tendon integrity, fatty degeneration and atrophy with intraoperative and clinical findings after repair of massive tears, and repair integrity has been related to functional outcome after double-row repair^47,48. In a multicentre concordance study of 113 repairs performed with a suture-bridge technique, the concordance between US and MRI readings was 85% (kappa 0.40) when MRI was read by a central radiologist and 92% (kappa 0.70) when MRI was read by the investigating surgeon⁴⁹. The authors concluded that US may be used to evaluate repaired tendon integrity and constitutes a comparable alternative to MRI, while advising investigators to compare postoperative US with MRI for a period before relying solely on US⁴⁹.

Artificial-intelligence-assisted interpretation:

A deep-learning model for screening rotator cuff tears in three planes of routine shoulder MRI was developed in 794 scans, using the YOLO v8 architecture trained on tears labelled by three musculoskeletal radiologists [50]. The area under the curve was highest when all planes were used together (0.94), and among single planes the axial plane performed best (0.71), followed by the sagittal (0.70) and coronal (0.68) planes⁵⁰. A systematic review of MRI-based artificial-intelligence models, covering 2020 to November 2024, found accuracies from 71.0% to 100% and no significant differences in accuracy, sensitivity, specificity or precision between artificial intelligence and human experts in studies that included such comparisons⁹. A broader review described the application of artificial intelligence in the management of cuff tears as promising yet experimental⁵¹.

Methodological quality of the primary evidence:

Across the included syntheses, several recurrent threats to validity were evident. In the bursal-sided partial-thickness meta-analysis, QUADAS-2 scores of the 12 included studies ranged from 7 to 11, only seven studies stated that reader blinding was used, one study used a prospective design, and heterogeneity was substantial, with an I² of 84.4% for the specificity of MRA¹. Smith and colleagues appraised their 62 studies with the QUADAS form¹⁰. The Cochrane review highlighted that no single study compared all three tests in the same patients and that the populations were dominated by patients with a high prevalence of tears⁴. A large share of studies used arthroscopy as the reference standard, which is generally applied only to patients who proceed to surgery and therefore creates a risk of verification bias.

DISCUSSION

Principal findings:

Three conclusions emerge from the evidence reviewed. First, for full-thickness tears, the three principal techniques perform similarly well, with sensitivity and specificity above 0.90 in the most comprehensive pooled analysis¹¹. Second, the performance of all techniques is lower for partial-thickness tears, and it is in this category that the choice between US, MRI and MRA is genuinely contested. Third, ancillary information, such as muscle quality and repair integrity, is important for management but is measured with variable reliability.

Interpreting the comparison of US, MRI and MRA:

The apparent disagreement between meta-analyses is largely explained by what each was able to compare. The meta-analysis that ranked MRA first pooled 65 articles and compared modalities across studies³. The Cochrane review, which used stricter inclusion criteria to restrict participants to those with suspected cuff tears for whom surgery was being considered, found no difference between MRI and US and could not pool MRA studies for the most important categories⁴. Neither review had access to a body of studies in which all three tests were performed in the same patients, and the Cochrane authors explicitly called for such studies⁴. The pooled estimates should therefore be read as evidence that US and MRI are broadly equivalent, rather than as proof that MRA is superior or that it adds nothing.

The bursal-sided partial-thickness meta-analysis illustrates why this subtlety matters. MRA and MRI had identical pooled sensitivity of 0.77, so no gain from contrast was demonstrated¹. The authors reasoned that contrast may add little for bursal-sided lesions because the contrast injected into the joint may fail to reach the bursal surface and may yield false-negative results, while an inflamed tendon may yield false-positive results¹. This anatomical logic is consistent with the established concept that arthrography is intrinsically better suited to the articular surface than to the bursal surface. For articular-sided and intratendinous lesions, the evidence summarised here does not allow a firm conclusion, which highlights an important gap that future meta-analyses should address separately for each partial-thickness subtype.

The role of operator and reader expertise:

Reported US performance appears robust across examiner types, but this finding sits beside the observation that US is equivalent to MRI specifically in the hands of experienced operators^11,13. The two statements are compatible if one accepts that the examiners in the underlying studies were typically trained to a standard that the primary authors considered adequate. In practice, the strength of US depends on local expertise, on the quality of the equipment and on the clinician’s ability to perform a dynamic, focused examination. The reliability data for muscle grading point in the same direction for MRI: in a study of five readers, radiologists produced higher and less variable agreement than orthopaedic surgeons, and more experienced readers tended to give more consistent results³⁵. The implication is that structured reporting and training may improve diagnostic performance as much as technological upgrades.

Why partial-thickness tears remain difficult:

Partial-thickness tears are difficult to detect because the abnormal tissue is a small proportion of the tendon cross-section and because tendinopathy, intratendinous degeneration and partial tears share imaging features. The lower accuracy of US for partial-thickness than for full-thickness tears is paralleled by the observation that the diagnostic accuracy of MRI for full-thickness tears is significantly higher than for partial-thickness tears^10,12,13. Partial-thickness lesions matter clinically because they may progress, and because the proportion of tendon thickness involved is relevant to treatment¹. The identification of subtype has therefore become more meaningful, as the authors of the bursal-sided meta-analysis emphasised¹. Delaminated tears, which are not captured by a simple full-thickness versus partial-thickness dichotomy, have been studied separately with indirect MRA³⁰. A stratified approach, in which accuracy is reported by tear subtype, thickness threshold and tendon, is likely to be more informative than the current practice of reporting a single figure for partial-thickness tears.

Radiography as a first step:

The evidence on radiographs supports a role as a first-line adjunct rather than as a diagnostic test for the tear itself. The acromiohumeral interval has high specificity, so a narrow interval strongly suggests a full-thickness tear, but low sensitivity means that a normal interval does not exclude one¹⁵. Measurement conditions influence the result: the interval is lower on supine radiographs and MRI than on upright radiographs, so thresholds derived in one position cannot be transferred to another¹⁵. The review of acromiohumeral distance confirmed that heterogeneity in protocols and thresholds limits the direct application of any single cut-off⁸. The acromiohumeral interval and the upward migration index appeared to be the most discriminative of five radiographic measures¹⁶. Combined with the observation that acromial morphology may relate to cuff disease, radiographs help in the assessment of the bony environment and in the exclusion of arthropathy or calcific deposits, but they cannot replace soft-tissue imaging¹⁴.

Muscle quality and the limits of the Goutallier grade:

Fatty infiltration is a prognostic marker, but the tool used to measure it is imperfect. The reliability is only fair to good, correlation with a quantitative reference is weak, and the choice of slice influences the result^35,36,40. The 2025 r-view study suggests that in massive tears with severe retraction the conventional Y view may overestimate fat infiltration because the muscle belly has retracted medially and the lateral slice samples tendinous tissue³⁶. If confirmed in independent cohorts, this would change how cuff tears are triaged for repair, because misclassifying a partially reparable tear as irreparable would distort surgical planning. Nonetheless, the r-view study was retrospective, relied on MRI alone and did not validate grades against biopsy, so its conclusions should be considered hypothesis-generating.

Imaging after repair:

The concordance data between US and MRI after repair are reassuring but incomplete. Agreement of 92% (kappa 0.70) between US and MRI read by the investigating surgeons, but of only 85% (kappa 0.40) when MRI was read by a central radiologist, illustrates that reader and context affect apparent concordance⁴⁹. Magnetic susceptibility artefact from anchors may limit MRI near the repair, whereas CT arthrography is less affected, as the authors of one study argued when selecting it for postoperative assessment³⁵. The clinical relevance of postoperative muscle changes is also noteworthy, because fat infiltration of the supraspinatus and infraspinatus can progress after surgery regardless of tendon integrity, and because fatty infiltration and atrophy have been reported not to improve after repair and to correlate with poor functional outcome^38,46.

Artificial intelligence:

The reported accuracy range of 71.0% to 100% in the systematic review of MRI-based models is too wide to be interpreted as a single performance figure [9]. Individual models show promising discrimination, as the area under the curve of 0.94 for the multiplane model demonstrates, but the plane-specific results (0.68 to 0.71) show that performance depends strongly on the information supplied to the model⁵⁰. Artificial intelligence is best viewed as an assistive technology that may reduce reading time and variability, a view in line with the description of current applications as promising yet experimental⁵¹. Before clinical adoption, models require external validation in multicentre datasets, calibration against surgical findings and demonstration that they improve decisions rather than merely match readers.

A pragmatic framework for modality selection:

Because no single study compares all modalities in the same patients, the following framework is a synthesis of the evidence rather than a validated algorithm. First, radiographs should be obtained to assess the bony environment and to look for indirect signs of cuff failure. Second, where expertise is available, US is a reasonable first cross-sectional test, particularly for suspected full-thickness tears, because it has high pooled accuracy, similar performance across examiner types and low cost and risk^10,11. Third, MRI is appropriate when comprehensive assessment is needed, such as for surgical planning, evaluation of muscle quality, assessment of the subscapularis or biceps, or when US is inconclusive or unavailable. Fourth, MRA may be considered when MRI is equivocal and a partial-thickness articular-surface or small full-thickness tear is suspected, bearing in mind that it is invasive and that its advantage is not established for bursal-sided tears^1,3. Fifth, CT arthrography may be an option when MRI is contraindicated or heavily degraded by metal³⁵.

Strengths and limitations of this review:

The strengths of this review are the inclusion of recent syntheses together with the classic primary studies, the verification of every bibliographic record and DOI against a source page, and the attention given to reliability and to postoperative and emerging applications, which are often omitted from reviews that focus on sensitivity and specificity alone.

Several limitations should be acknowledged. The search was web-based with reference-list expansion and was not run as a series of database exports, so a PRISMA flow diagram with record counts could not be provided and some eligible studies may have been missed. Screening and extraction were performed by a single reviewer without duplicate verification. The protocol was not registered. Pooled estimates were taken from published syntheses and were not re-derived, and several of those syntheses overlap in their primary studies, so the estimates are not independent. The quality of the primary studies was not re-scored, and the certainty of the evidence was not formally graded. Finally, a number of statements about the content of primary studies rely on the descriptions provided in the secondary sources that cited them, and these were attributed accordingly.

Implications for research

Future research should prioritise prospective, multimodality studies in which US, MRI and MRA are performed in the same consecutive patients, interpreted by blinded readers, and verified against a standard reference that includes some patients not selected for surgery. Studies should report accuracy separately for articular-sided, bursal-sided and intratendinous partial-thickness tears and for each tendon. Consensus on measurement of acromiohumeral distance, on the slice used for fat grading and on the reporting of postoperative integrity would reduce heterogeneity. Artificial-intelligence tools should be tested prospectively across scanners and institutions.

CONCLUSIONS

The imaging of rotator cuff injuries is best supported by evidence for full-thickness tears, for which US, MRI and MRA each achieve high pooled sensitivity and specificity. Ultrasonography performs well when performed by trained examiners and is attractive on grounds of cost and safety, and MRI provides the most comprehensive assessment of the tendon, the muscle and associated pathology. MR arthrography may be more accurate in some pooled comparisons, but a Cochrane review could not confirm this because of the absence of direct three-way comparisons, and for bursal-sided partial-thickness tears the two MRI approaches had similar value. Plain radiographs have a supporting role through indirect signs, particularly a narrow acromiohumeral interval, which is specific but insensitive.

Partial-thickness tears remain the principal diagnostic challenge, and tear-subtype-specific accuracy data are needed. Semi-quantitative grading of fatty infiltration is only moderately reliable and is sensitive to the slice chosen, so quantitative methods and standardised sections deserve wider evaluation. After repair, US and MRI show reasonable but reader-dependent concordance. Deep-learning models report encouraging accuracy but require external validation before routine use.

The overall certainty of the evidence is limited by retrospective designs, surgically enriched populations, heterogeneity, overlap between syntheses and the scarcity of studies comparing all modalities in the same patients. Until such studies are available, modality selection should be individualised according to the clinical question, local expertise, patient factors and availability, and it should be based on the framework outlined above.

REFERENCES

1. Huang T, Liu J, Ma Y, Zhou D, Chen L, Liu F. Diagnostic accuracy of MRA and MRI for the bursal-sided partial-thickness rotator cuff tears: a meta-analysis. J Orthop Surg Res. 2019;14:436. doi: 10.1186/s13018-019-1460-y.
2. Yamaguchi K, Tetro AM, Blam O, Evanoff BA, Teefey SA, Middleton WD. Natural history of asymptomatic rotator cuff tears: a longitudinal analysis of asymptomatic tears detected sonographically. J Shoulder Elbow Surg. 2001;10(3):199-203. doi: 10.1067/mse.2001.113086.
3. de Jesus JO, Parker L, Frangos AJ, Nazarian LN. Accuracy of MRI, MR arthrography, and ultrasound in the diagnosis of rotator cuff tears: a meta-analysis. AJR Am J Roentgenol. 2009;192(6):1701-7. doi: 10.2214/AJR.08.1241.
4. Lenza M, Buchbinder R, Takwoingi Y, Johnston RV, Hanchard NCA, Faloppa F. Magnetic resonance imaging, magnetic resonance arthrography and ultrasonography for assessing rotator cuff tears in people with shoulder pain for whom surgery is being considered. Cochrane Database Syst Rev. 2013;(9):CD009020. doi: 10.1002/14651858.CD009020.pub2.
5. McInnes MDF, Moher D, Thombs BD, McGrath TA, Bossuyt PM, Clifford T, et al. Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: the PRISMA-DTA statement. JAMA. 2018;319(4):388-96. doi: 10.1001/jama.2017.19163.
6. Whiting PF, Rutjes AWS, Westwood ME, Mallett S, Deeks JJ, Reitsma JB, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. 2011;155(8):529-36. doi: 10.7326/0003-4819-155-8-201110180-00009.
7. Nelson MC, Leather GP, Nirschl RP, Pettrone FA, Freedman MT. Evaluation of the painful shoulder. A prospective comparison of magnetic resonance imaging, computerized tomographic arthrography, ultrasonography, and operative findings. J Bone Joint Surg Am. 1991;73(5):707-16. doi: 10.2106/00004623-199173050-00009.
8. Arráez-Aybar LA, García-de-Pereda-Notario CM, Palomeque-Del-Cerro L, Montoya-Miñano JJ. Acromiohumeral distance as a diagnostic and prognostic biomarker for shoulder disorders: a systematic review—acromiohumeral distance and shoulder disorders. J Funct Morphol Kinesiol. 2025;10(4):478. doi: 10.3390/jfmk10040478.
9. Longo UG, Bandini B, Mancini L, Merone M, Schena E, de Sire A, et al. Artificial intelligence in rotator cuff tear detection: a systematic review of MRI-based models. Diagnostics (Basel). 2025;15(11):1315. doi: 10.3390/diagnostics15111315.
10. Smith TO, Back T, Toms AP, Hing CB. Diagnostic accuracy of ultrasound for rotator cuff tears in adults: a systematic review and meta-analysis. Clin Radiol. 2011;66(11):1036-48. doi: 10.1016/j.crad.2011.05.007.
11. Roy JS, Braën C, Leblond J, Desmeules F, Dionne CE, MacDermid JC, et al. Diagnostic accuracy of ultrasonography, MRI and MR arthrography in the characterisation of rotator cuff disorders: a systematic review and meta-analysis. Br J Sports Med. 2015;49(20):1316-28. doi: 10.1136/bjsports-2014-094148.
12. Li WF, Shakir TM, Zhao Y, Chen T, Niu C, Wang Z. An evidence-based approach to assess the accuracy of MRI in diagnosing rotator cuff tears: a systematic review and meta-analysis. Iran J Radiol. 2019;16(2):e79079. doi: 10.5812/iranjradiol.79079.
13. Farooqi AS, Lee A, Novikov D, Kelly AM, Li X, Kelly JD, et al. Diagnostic accuracy of ultrasonography for rotator cuff tears: a systematic review and meta-analysis. Orthop J Sports Med. 2021;9(10):23259671211035106. doi: 10.1177/23259671211035106.
14. Oh JH, Kim JY, Lee HK, Choi JA. Classification and clinical significance of acromial spur in rotator cuff tear: heel-type spur and rotator cuff tear. Clin Orthop Relat Res. 2010;468(6):1542-50. doi: 10.1007/s11999-009-1058-5.
15. Sanguanjit P, Apivatgaroon A, Boonsun P, Srimongkolpitak S, Chernchujit B. The differences of the acromiohumeral interval between supine and upright radiographs of the shoulder. Sci Rep. 2022;12:9404. doi: 10.1038/s41598-022-13632-0.
16. Kolk A, Overbeek CL, de Groot JH, Nelissen RGHH, Nagels J. Reliability and discriminative accuracy of 5 measures for craniocaudal humeral position: an assessment on conventional radiographs. JSES Int. 2020;4(1):189-96. doi: 10.1016/j.jseint.2019.11.005.
17. Teng A, Liu F, Zhou D, He T, Chevalier Y, Klar RM. Effectiveness of 3-dimensional shoulder ultrasound in the diagnosis of rotator cuff tears: a meta-analysis. Medicine (Baltimore). 2018;97(37):e12405. doi: 10.1097/MD.0000000000012405.
18. Co S, Bhalla S, Rowan K, Aippersbach S, Bicknell S. Comparison of 2- and 3-dimensional shoulder ultrasound to magnetic resonance imaging in a community hospital for the detection of supraspinatus rotator cuff tears with improved worktime room efficiency. Can Assoc Radiol J. 2012;63(3):170-6. doi: 10.1016/j.carj.2011.02.003.
19. Teefey SA, Rubin DA, Middleton WD, Hildebolt CF, Leibold RA, Yamaguchi K. Detection and quantification of rotator cuff tears. Comparison of ultrasonographic, magnetic resonance imaging, and arthroscopic findings in seventy-one consecutive cases. J Bone Joint Surg Am. 2004;86(4):708-16. doi: 10.2106/00004623-200404000-00007.
20. Fotiadou AN, Vlychou M, Papadopoulos P, Karataglis DS, Palladas P, Fezoulidis IV. Ultrasonography of symptomatic rotator cuff tears compared with MR imaging and surgery. Eur J Radiol. 2008;68(1):174-9. doi: 10.1016/j.ejrad.2007.11.002.
21. Sipola P, Niemitukia L, Kröger H, Höfling I, Väätäinen U. Detection and quantification of rotator cuff tears with ultrasonography and magnetic resonance imaging—a prospective study in 77 consecutive patients with a surgical reference. Ultrasound Med Biol. 2010;36(12):1981-9. doi: 10.1016/j.ultrasmedbio.2010.09.001.
22. Magee T. 3-T MRI of the shoulder: is MR arthrography necessary? AJR Am J Roentgenol. 2009;192(1):86-92. doi: 10.2214/AJR.08.1097.
23. Lee JH, Yoon YC, Jung JY, Yoo JC. Rotator cuff tears: noncontrast MRI compared to MR arthrography. Skeletal Radiol. 2015;44(12):1745-54. doi: 10.1007/s00256-015-2228-z.
24. Lo HC, Hung ST, Kuo DP, Chen YL, Lee HM. Quantitative diffusion-weighted magnetic resonance imaging for the diagnosis of partial-thickness rotator cuff tears. J Shoulder Elbow Surg. 2016;25(9):1433-41. doi: 10.1016/j.jse.2016.01.020.
25. Fritz LB, Ouellette HA, O’Hanley TA, Kassarjian A, Palmer WE. Cystic changes at supraspinatus and infraspinatus tendon insertion sites: association with age and rotator cuff disorders in 238 patients. Radiology. 2007;244(1):239-48. doi: 10.1148/radiol.2441050029.
26. Grainger AJ, Elliott JM, Campbell RSD, Tirman PFJ, Steinbach LS, Genant HK. Direct MR arthrography: a review of current use. Clin Radiol. 2000;55(3):163-76. doi: 10.1053/crad.1999.0374.
27. Choo HJ, Lee SJ, Kim OH, Seo SS, Kim JH. Comparison of three-dimensional isotropic T1-weighted fast spin-echo MR arthrography with two-dimensional MR arthrography of the shoulder. Radiology. 2012;262(3):921-31. doi: 10.1148/radiol.11111261.
28. Lee JH, Yoon YC, Jee S, Kwon JW, Cha JG, Yoo JC. Comparison of three-dimensional isotropic and two-dimensional conventional indirect MR arthrography for the diagnosis of rotator cuff tears. Korean J Radiol. 2014;15(6):771-80. doi: 10.3348/kjr.2014.15.6.771.
29. Oh DK, Yoon YC, Kwon JW, Choi SH, Jung JY, Bae S, et al. Comparison of indirect isotropic MR arthrography and conventional MR arthrography of labral lesions and rotator cuff tears: a prospective study. AJR Am J Roentgenol. 2009;192(2):473-9. doi: 10.2214/AJR.08.1223.
30. Choo HJ, Lee SJ, Kim JH, Kim DW, Park YM, Kim OH, et al. Delaminated tears of the rotator cuff: prevalence, characteristics, and diagnostic accuracy using indirect MR arthrography. AJR Am J Roentgenol. 2015;204(2):360-6. doi: 10.2214/AJR.14.12555.
31. Chun KA, Kim MS, Kim YJ. Comparisons of the various partial-thickness rotator cuff tears on MR arthrography and arthroscopic correlation. Korean J Radiol. 2010;11(5):528-35. doi: 10.3348/kjr.2010.11.5.528.
32. Waldt S, Bruegel M, Mueller D, Holzapfel K, Imhoff AB, Rummeny EJ, et al. Rotator cuff tears: assessment with MR arthrography in 275 patients with arthroscopic correlation. Eur Radiol. 2007;17(2):491-8. doi: 10.1007/s00330-006-0370-7.
33. Pfirrmann CWA, Zanetti M, Weishaupt D, Gerber C, Hodler J. Subscapularis tendon tears: detection and grading at MR arthrography. Radiology. 1999;213(3):709-14. doi: 10.1148/radiology.213.3.r99dc03709.
34. Jung JY, Jee WH, Chun CW, Kim YS. Diagnostic performance of MR arthrography with anterior trans-subscapularis versus posterior injection approach for subscapularis tendon tears at 3.0 T. Eur Radiol. 2017;27(3):1303-11. doi: 10.1007/s00330-016-4467-3.
35. Oh JH, Kim SH, Choi JA, Kim Y, Oh CH. Reliability of the grading system for fatty degeneration of rotator cuff muscles. Clin Orthop Relat Res. 2010;468(6):1558-64. doi: 10.1007/s11999-009-0818-6.
36. Zhang G, Hu Q, Zhang J, Ying L, Hang D, Zhou X, et al. A new section of shoulder magnetic resonance imaging (MRI) used to assess the fat infiltration of the rotator cuff muscles. J Orthop Surg Res. 2025;20:433. doi: 10.1186/s13018-025-05832-x.
37. Fuchs B, Weishaupt D, Zanetti M, Hodler J, Gerber C. Fatty degeneration of the muscles of the rotator cuff: assessment by computed tomography versus magnetic resonance imaging. J Shoulder Elbow Surg. 1999;8(6):599-605. doi: 10.1016/S1058-2746(99)90097-6.
38. Gladstone JN, Bishop JY, Lo IKY, Flatow EL. Fatty infiltration and atrophy of the rotator cuff do not improve after rotator cuff repair and correlate with poor functional outcome. Am J Sports Med. 2007;35(5):719-28. doi: 10.1177/0363546506297539.
39. Goutallier D, Postel JM, Gleyze P, Leguilloux P, Van Driessche S. Influence of cuff muscle fatty degeneration on anatomic and functional outcomes after simple suture of full-thickness tears. J Shoulder Elbow Surg. 2003;12(6):550-4. doi: 10.1016/S1058-2746(03)00211-8.
40. Gilbert F, Böhm D, Eden L, Schmalzl J, Meffert RH, Köstler H, et al. Comparing the MRI-based Goutallier classification to an experimental quantitative MR spectroscopic fat measurement of the supraspinatus muscle. BMC Musculoskelet Disord. 2016;17:355. doi: 10.1186/s12891-016-1216-3.
41. Pfirrmann CWA, Schmid MR, Zanetti M, Jost B, Gerber C, Hodler J. Assessment of fat content in supraspinatus muscle with proton MR spectroscopy in asymptomatic volunteers and patients with supraspinatus tendon lesions. Radiology. 2004;232(3):709-15. doi: 10.1148/radiol.2323030442.
42. van de Sande MAJ, Stoel BC, Obermann WR, Tjong a Lieng JGS, Rozing PM. Quantitative assessment of fatty degeneration in rotator cuff muscles determined with computed tomography. Invest Radiol. 2005;40(5):313-9. doi: 10.1097/01.rli.0000160014.16577.86.
43. Strobel K, Hodler J, Meyer DC, Pfirrmann CWA, Pirkl C, Zanetti M. Fatty atrophy of supraspinatus and infraspinatus muscles: accuracy of US. Radiology. 2005;237(2):584-9. doi: 10.1148/radiol.2372041612.
44. Kavanagh EC, Koulouris G, Parker L, Morrison WB, Bergin D, Zoga AC, et al. Does extended-field-of-view sonography improve interrater reliability for the detection of rotator cuff muscle atrophy? AJR Am J Roentgenol. 2008;190(1):27-31. doi: 10.2214/AJR.07.2716.
45. Thomazeau H, Rolland Y, Lucas C, Duval JM, Langlais F. Atrophy of the supraspinatus belly. Assessment by MRI in 55 patients with rotator cuff pathology. Acta Orthop Scand. 1996;67(3):264-8. doi: 10.3109/17453679608994685.
46. Liem D, Lichtenberg S, Magosch P, Habermeyer P. Magnetic resonance imaging of arthroscopic supraspinatus tendon repair. J Bone Joint Surg Am. 2007;89(8):1770-6. doi: 10.2106/JBJS.F.00749.
47. Mellado JM, Calmet J, Olona M, Esteve C, Camins A, Pérez del Palomar L, et al. Surgically repaired massive rotator cuff tears: MRI of tendon integrity, muscle fatty degeneration, and muscle atrophy correlated with intraoperative and clinical findings. AJR Am J Roentgenol. 2005;184(5):1456-63. doi: 10.2214/ajr.184.5.01841456.
48. Sugaya H, Maeda K, Matsuki K, Moriishi J. Repair integrity and functional outcome after arthroscopic double-row rotator cuff repair: a prospective outcome study. J Bone Joint Surg Am. 2007;89(5):953-60. doi: 10.2106/JBJS.F.00512.
49. Codsi MJ, Rodeo SA, Scalise JJ, Moorehead TM, Ma CB. Assessment of rotator cuff repair integrity using ultrasound and magnetic resonance imaging in a multicenter study. J Shoulder Elbow Surg. 2014;23(10):1468-72. doi: 10.1016/j.jse.2014.01.045.
50. Lee KC, Cho Y, Ahn KS, Park HJ, Kang YS, Lee S, et al. Deep-learning-based automated rotator cuff tear screening in three planes of shoulder MRI. Diagnostics (Basel). 2023;13(20):3254. doi: 10.3390/diagnostics13203254.
51. Rodriguez HC, Rust B, Hansen PY, Maffulli N, Gupta M, Potty AG, et al. Artificial intelligence and machine learning in rotator cuff tears. Sports Med Arthrosc Rev. 2023;31(3):67-72. doi: 10.1097/JSA.0000000000000371.

APPENDICE

TABLE 1. Pooled findings reported by the principal systematic reviews and meta-analyses.

Source	Evidence base	Index test(s)	Principal result as reported
Smith et al. [10]	62 studies; 6007 patients; 6066 shoulders	US	Partial-thickness tears: sensitivity 0.84, specificity 0.89. Full-thickness tears: sensitivity 0.96, specificity 0.93.
de Jesus et al. [3]	65 articles	MRI, MRA, US	MRA more sensitive and specific than MRI or US for full-thickness and partial-thickness tears. Summary AUC: MRA 0.935, US 0.889, MRI 0.878. No significant MRI–US difference.
Lenza et al. [4]	20 studies	MRI, MRA, US	MRI versus US: no significant difference for any tear (P = 0.13) or partial-thickness tears (P = 1.0). MRA not pooled for any or partial tears. No study evaluated all three tests in the same population.
Roy et al. [11]	Systematic search of three databases	US, MRI, MRA	Full-thickness tears: sensitivity and specificity above 0.90 for all three tests. US accuracy similar across radiologists, sonographers and orthopaedists.
Li et al. [12]	18 studies	MRI	Any tear: pooled sensitivity 0.93, specificity 0.88. Accuracy higher for full-thickness than for partial-thickness tears.
Huang et al. [1]	12 studies; 1740 patients; 1741 shoulders	MRA, MRI	Bursal-sided partial-thickness tears: MRA sensitivity 0.77, specificity 0.98, AUC 0.88; MRI sensitivity 0.77, specificity 0.96, AUC 0.82. Similar diagnostic value.
Farooqi et al. [13]	Review with meta-analysis of 5 US–MRI comparison studies	US, MRI	Median US accuracy 0.93 for full-thickness and 0.81 for partial-thickness supraspinatus tears. No significant US–MRI difference.
Longo et al. [9]	Review of MRI-based AI models, 2020 to November 2024	Machine learning and deep learning on MRI	Accuracy 71.0% to 100%. No significant difference from human experts where compared (p > 0.05).

Abbreviations: AUC, area under the receiver operating characteristic curve; MRA, magnetic resonance arthrography; MRI, magnetic resonance imaging; US, ultrasonography; AI, artificial intelligence.