In the present study we compared the accuracy of US and MRI in the detection of rotator cuff tears in patients who underwent both imaging techniques. In our institution, US is the method of choice in evaluating patients with shoulder complaints. As in other countries, the use of US has increased significantly [20]. In our study the use of US obviates the need for further imaging in 95% of the cases.
The high lifetime prevalence of shoulder pain of 66% [21] and the moderate reliability and reproducibility of clinical history and clinical examination may be an explanation for the large number of US examinations. In this study we focussed on the presence or absence of rotator cuff tears. Other diagnoses that are often made with US, e.g., subacromial bursitis and impingement syndrome, were not evaluated. The combination of clinical history, clinical examination and ultrasound fulfil the need for diagnostic certainty and permit the initiation of therapy in most cases. The disadvantage of the liberal use of US may be a large number of negative findings as was an additional finding of our study, because the indication to perform US of the shoulder was shoulder pain and/or disability instead of suspicion for having a rotator cuff tear. Another disadvantage could be a large number of false-positive findings, as there is a chance of up to 50% of finding abnormalities in an asymptomatic shoulder [22]. Therefore, good clinical examination remains of utmost importance in the evaluation of patients with shoulder complaints.
Another purpose of this retrospective study was to evaluate the diagnostic accuracy of US in cases in which additional MRI was requested and to compare these two techniques.
US and MRI findings were compared with surgical findings and appeared comparably accurate in diagnosing full-thickness tears (94% and 94%, respectively) and less, but also comparably accurate for the detection of partial-thickness tears (81% and 84%, respectively). Our findings substantiate those reported by Dinnes et al. [11] and Teefey et al. [9], who showed that US and MRI have comparable accuracy for identifying partial-thickness and full-thickness rotator cuff tears. Although Teefey et al. [9] performed a prospective study, while we retrospectively analysed findings in a more diverse patient population from daily practice, both studies show that MRI of the shoulder provides, with regard to the rotator cuff, little additional information following an US examination. However, in our selected study group in 7 (10%) of the 68 patients MRI detected intraarticular pathology, which changed the therapy strategy. Furthermore, for some surgeons MRI may have additional value to assess fatty infiltration of the rotator cuff; however, we agree with others that US can depict fatty infiltration and atrophy of the rotator cuff as reliably as MRI [23, 24]. Although it is known that US and MRI have comparable accuracy for identifying and measuring the size of full-thickness and partial-thickness rotator cuff tears [9], MRI may be used to define the precise location and extent of a rotator cuff tear; however, this was not the case in our series.
There are several limitations of our study. A potential drawback is the operator dependency of US [25–27] and MRI [16, 28]. In an unpublished study we evaluated the learning curve and the interobserver variability of US in a series of 200 patients. If US was performed in a standardised manner, the interobserver agreement was excellent. The kappa coefficient was calculated to be 0.80 (SE = 0.05).
Furthermore, the study design was prone to bias. For example, the study population of the 207 patients who underwent US and MRI but did not undergo surgery probably differs from the 68 patients who were operated upon (selection bias). On the other hand, the agreement between US and MRI in the group of 207 patients was 86%, approximately similar to that in the group of 68 patients (85%), indicating that the result of our study was not biased by this selection.
Also, a verification or workup bias was present because imaging findings were known by the surgeon and influenced the decision whether or not to treat surgically and thus influenced patient selection. Preoperative knowledge of the imaging results caused diagnostic review bias, as a result of a more thorough exploration of the cuff in order to find a RCT identified using US or MRI, which of course influences the gold standard.
The value of MRI as a follow-up examination is probably underestimated due to the low threshold to request US and consequently overuse of US.
Finally, there is an imperfect standard bias, which occurs when the reference standard is not 100% accurate. In our opinion the so-called gold standard is such a potential cause of bias. Waldt et al. [29] showed that the diagnosis of small partial-thickness tears are restricted because of difficulties in the differentiation among fibre tearing, tendinitis, synovitic changes and superficial fraying at tendon margins. Interobserver variability is also introduced by varying definitions and/or synonyms used by both sonologists and surgeons. Kuhn et al. [30] showed that six currently described rotator cuff classification systems have demonstrated little interobserver agreement among experienced shoulder surgeons. In our experience in these studies the ‘gold standard’ was more a ‘silver handicap’, especially with regard to the detection of partial-thickness rotator cuff tears (Fig. 4). When we assume that the seven cases in which both US and MRI showed a non-surgically proven PTT were true-positives, then the sensitivity, specificity, accuracy, PPV and NPV of US would increase to 94%, 90%, 91%, 75%, 98%, and those of MRI to 81%, 98%, 94%, 93%, 94%, respectively, which is almost as good as the accuracy for diagnosing full-thickness rotator cuff tears. The imperfect standard bias may cause an underestimation of the reported accuracy for diagnosing partial-thickness rotator cuff tears with US and MRI.
At the RSNA meeting of 2008 a special focus session was dedicated to the question “Musculoskeletal US: Has the Time Come?” We have demonstrated that diagnostic US of the shoulder in patients with periarticular complaints in our institution performed by a radiologist fulfils the clinical need for diagnosis and further management. We are of the opinion that if musculoskeletal radiologists ignore increasing requests for US of the shoulder, these examinations will soon be performed by rheumatologists [27, 31], orthopaedic surgeons [32–34], physiotherapists or family physicians who have been reported to be able to performing US of the shoulder equally well.
In summary, in patients with periarticular shoulder pain, US is a reliable diagnostic tool that obviates the need for further imaging in most cases. Our study established that US and MRI yield comparably high sensitivity, diagnostic accuracy and positive predictive value in detecting full-thickness rotator cuff tears. In detecting partial-thickness rotator cuff tears both tests are less accurate; however, US appears to be more sensitive than MRI.
Finally, following US of the shoulder performed by a dedicated radiologist, MRI offers little additional value, with regard to the detection of rotator cuff tears. Of course local setting and other factors such as equipment availability, personal expertise and preference, patient preference [35] and cost effectiveness [36] may play a role in choosing which imaging technique will be used.