- Email: [email protected]
Diagnostic accuracy of MRI in adults with suspect brachial plexus lesions: A multicentre retrospective study with surgical findings and clinical follow-up as reference standard
European Journal of Radiology 81 (2012) 2666–2672
Contents lists available at SciVerse ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Diagnostic accuracy of MRI in adults with suspect brachial plexus lesions: A multicentre retrospective study with surgical findings and clinical follow-up as reference standard Alberto Tagliafico a,∗ , Giulia Succio b , Giovanni Serafini b , Carlo Martinoli c a
Institute of Anatomy, Department of Experimental Medicine, University of Genoa, Largo Rosanna Benzi 8, 16132 Genoa, Italy Department of Radiology, Santa Corona Hospital, Pietra Ligure, Italy via XXV Aprile, 38- Pietra Ligure, 17027 Savona, Italy c Radiology Department, DISC, Università di Genova, Largo Rosanna Benzi 8, 16138 Genova, Italy b
a r t i c l e
i n f o
Article history: Received 7 September 2011 Received in revised form 12 October 2011 Accepted 14 October 2011 Keywords: Brachial Plexus MRI Sensitivity Specificity Accuracy
a b s t r a c t Objective: To evaluate brachial plexus MRI accuracy with surgical findings and clinical follow-up as reference standard in a large multicentre study. Materials and methods: The research was approved by the Institutional Review Boards, and all patients provided their written informed consent. A multicentre retrospective trial that included three centres was performed between March 2006 and April 2011. A total of 157 patients (men/women: 81/76; age range, 18–84 years) were evaluated: surgical findings and clinical follow-up of at least 12 months were used as the reference standard. MR imaging was performed with different equipment at 1.5 T and 3.0 T. The patient group was divided in five subgroups: mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation, and other. Sensitivity, specificity with 95% confidence intervals (CIs), positive predictive value (PPV), pre-testprobability (the prevalence), negative predictive value (NPV), pre- and post-test odds (OR), likelihood ratio for positive results (LH+), likelihood ratio for negative results (LH−), accuracy and post-test probability (post-P) were reported on a per-patient basis. Results: The overall sensitivity and specificity with 95% CIs were: 0.810/0.914; (0.697–0.904). Overall PPV, pre-test probability, NPV, LH+, LH−, and accuracy: 0.823, 0.331, 0.905, 9.432, 0.210, 0.878. Conclusions: The overall diagnostic accuracy of brachial plexus MRI calculated on a per-patient base is relatively high. The specificity of brachial plexus MRI in patients suspected of having a space-occupying mass is very high. The sensitivity is also high, but there are false-positive interpretations as well. © 2011 Elsevier Ireland Ltd. All rights reserved.
1. Introduction The brachial plexus provides motor and sensory innervations to the upper extremities. Pathologic conditions involving the brachial plexus are relatively uncommon and frequently misdiagnosed with disastrous consequences [1]. Brachial plexopathy causes weakness, sensory loss, and loss of reflexes in body regions innervated by the C5-T1 segmental distribution [2]. The clinical diagnosis is usually confirmed by electrodiagnostic studies (EMG). Clinical evaluation of this anatomical region is a real challenge, and electrophysiological studies may not be sufficient to identify and locate the cause of brachial plexus impairment. Currently, MR imaging (MRI) is considered the evaluation of choice because of its multi-planar
∗ Corresponding author. Tel.: +39 0103537885/3479745122; fax: +39 0103537882. E-mail addresses: alberto.tagliafi[email protected], albertotagliafi[email protected] (A. Tagliafico). 0720-048X/$ – see front matter © 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ejrad.2011.10.007
capabilities and excellent soft-tissue contrast [3]. However, MRI has inherent limitations in specific clinical settings and can be variably supported by other modalities depending on the underlying clinical problem. The use of MRI also depends on the management policy of the referring clinicians and peripheral nerve surgeons [4]. Electromyography and MRI are often used as complementary tools to assess brachial plexopathy [1]. Previous studies have evaluated the diagnostic accuracy of brachial plexus MRI in traumatic root avulsions [4], in brachial plexopathy, and in primary or secondary tumors [4–7]. Different studies have demonstrated that brachial plexus MRI is useful in chronic inflammatory demyelinating polyradiculo-neuropathy [8], Parsonage-Turner syndrome, herpes zoster infection and thoracic outlet syndrome [9–14]. Reported values of MRI diagnostic accuracy range between 52% and 88% for traumatic root avulsions and reach 96% of sensitivity for tumor detection [5–14]. However, most of the published papers on brachial plexus MRI evaluated small series or selected pathological conditions [9–15]. Moreover, in the last decade, the diagnostic workup of brachial plexus pathology has been
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
2667
Table 1 Parameters used for MR imaging of the brachial plexus at 3-T and 1.5-T. T1-w tSE sequences
Repetition time (ms) Echo time (ms) Flip angle Voxel size (mm) Matrix Slice thickness (mm) Phase resolution Echo train length per slice Bandwidth Hz/px SAR (W/kg body weight) Acquisition time
T2-w tSE sequences
3-T MR imaging
1.5-T MR imaging
3-T MR imaging
1.5-T MR imaging
600 9.1 90 0.2 × 0.2 × 2 320 × 320 4 100 3 122 1.8 5 min.14 s
500 14
4220 89 90 0.2 × 0.2 × 2 320 × 320 4 100 17 162 1.6 5 min.10 s
3300 63
0.2 × 0.2 × 2 320 × 320 4 100 5 124 1.2 5 min.45 s
continually reviewed. Testing the diagnostic accuracy of current imaging strategies such as MRI is still important to avoid the underutilisation or inappropriate use of the available imaging techniques [4]. Therefore, the purpose of this study is to evaluate brachial plexus MRI accuracy in a large multicentre study with surgical findings and clinical follow-up as reference standard. 2. Materials and methods The research was approved by the Institutional Review Board. Patients provided their written informed consent to use of the data for statistical analysis and publication. This study of diagnostic accuracy was planned according to the Standards for Reporting of Diagnostic Accuracy (STARD) guidelines [16]. 2.1. Patients A multicentre retrospective trial was performed between March 2006 and April 2011, including three centres with proven imaging experience in brachial plexus MRI (at least 20 brachial plexus MR examinations per year and at least 1000 body MR examinations per year) and MR equipment with a 1.5–3.0-T magnet, 15 mT/m gradients, and at least 12 coil elements obtained by the combination of body, neck and spine-array coils. A total of 194 patients were enrolled; the centres provided 32, 88, and 74 cases, respectively. Patients 18 years old or older with suspicious brachial plexus lesions referred by neurologists, neurosurgeons and oncologists at each enrolling centre and surgical findings (obtained no later than 30 days after MR evaluation) or clinical follow-up of at least 12 months were used as the reference standards. Surgical findings included the reports for brachial plexus traumas and the results of biopsy for tumoral lesions. A level-by-level analysis based upon level-by-level exploration has not been done. Clinical follow-up included clinical and electrophysiological evaluation performed by the referring physicians as described elsewhere [17–19]. The exclusion criteria were absolute contraindications to MR, pregnancy or breastfeeding, severe renal failure, known hypersensitivity to gadolinium chelates and clinical status limiting data reliability. 2.2. MRI The imaging protocol included T1-weighted turbo spin-echo (tSE) sequences in the coronal, transverse and sagittal oblique planes and T2-weighed turbo spin-echo (tSE) sequences with fat saturation in the coronal, transverse and sagittal oblique planes (as perpendicular as possible to the brachial plexus nerves). T1weighted turbo spin-echo sequences with fat saturation following
0.2 × 0.2 × 2 320 × 320 4 100 14 124 1.3 6 min.36 s
gadolinium (0.1 mmol/kg of body weight administered at the rate of 2 mL/s, using an automatic injector) were obtained in patients suspected to have neoplasm, radiation injury, inflammation, or abscess, or nerve entrapment and stretch injury. Three different contrast media were used (MultiHance, Bracco Imaging, Milan, Italy; Magnevist, Bayer HealthCare, Berlin, Germany; Doterem, Guerbet, Villepinte, France). The brachial plexus MRI protocols employed in this study have been previously used at 1.5 T and 3.0 T [18–21]. Technical parameters for all sequences are summarised in Table 1. The imaging time for a complete examination was approximately 40 min. 2.3. Analysis of cases Cases with incomplete surgical/histopathological data or follow-up and cases with important imaging artifacts (e.g., a patient’s movement) were excluded from the analysis. At the central unit (off-site), two experienced radiologists with 7 and 9 years of experience in MRI checked all the examinations with respect to the protocol and the quality of the images. As a consequence, 37 patients were excluded from the analysis, and 157 patients remained for analysis. For these 157 cases (men/women: 81/76; age range, 18–84 years), an off-site interpretation of the MR images was performed independently by the two experienced radiologists, who were aware of the entry criterion and blinded to all of the patients’ data and to the results of surgical findings and clinical follow-up (G.S and C.M.). The two reviewers interpreted the MR examinations, blinded to each other and to the clinical data of the patients. Image analysis was repeated after 1 month to assess intra-observer agreement. All images were transferred to an external workstation equipped with Osirix software (Osirix 3.6 v 32-bit) and evaluated in a random fashion; pathological findings on MRI were evaluated on the basis of established criteria (Table 2) [2,4,20,21]. Comparison of brachial plexus MRI findings and the reference standard was performed offsite by a third radiologist (A.T.) who matched MR and reference standard reports, creating a database for the data analysis. An abnormality was considered to be depicted successfully (true-positive) if concordance was present between reported MRI findings and the reference standard. Abnormalities revealed by the
Table 2 Normal vs. abnormal brachial plexus, see references [2] and [4]. Abnormality on Brachial Plexus MRI Loss of fat planes around all or part of a plexus component Diffuse or focal enlargement of a component (especially, the presence of an eccentric or nodular mass) Marked hyperintensity on T2-weighted images and/or enhancement on T1-weighted images with fat suppression Altered fascicular pattern
2668
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
Table 3 Sensitivity and specificity with 95% Confidence Intervals (CIs) of the five groups of lesions examined. Per-patient MR imaging findings for the group overall
Mass lesion Traumatic injury Entrapment syndromes Post-treatment evaluation Other Total
TP 21 12 4 8 6 42
FP 1 2 1 4 1 9
FN 3 4 1 2 0 10
TN 35 20 6 23 12 96
reference standard but not MRI were classified as false-negative findings for the modality (e.g., subtle stretching injuries with positive reference standard and negative MRI). Because of the differences between brachial plexus lesions, the statistical analysis was performed on each subgroup (mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation, other) and on the entire group of lesions as a whole. This process was done after the initial analysis. Analysis was performed on a patient-by-patient basis. With regard to the surgical reports for brachial plexus trauma, only a correspondence between the area affected at surgical exploration and the area where MRI was performed provided additional information about the disease. A level-by-level analysis based upon level-by-level exploration was not performed. This process results in much higher sensitivity/specificity. To describe brachial plexus MRI performance, the sensitivity and specificity have been reported with their respective 95% confidence intervals (CIs). Positive predictive value, pre-test-probability (the prevalence), negative predictive value, likelihood ratio for positive results, likelihood ratio for negative results and accuracy were also reported. Logistic multiple regression was used to analyse the possible effect of independent variables, such as centre, MR equipment, abnormality, and patient age on accurate detection of the presence of an abnormality. Intra- and Inter-reader agreement in detecting or assessing lesion nature was determined by using generalised weighted statistics and was classified as excellent ( values > 0.80), good ( = 0.61–0.80), moderate ( = 0.41–0.60), fair ( = 0.21–0.40), or poor ( ≤ 0.20). All statistical tests were two sided at the P < .05 level of significance and were performed by using dedicated software (SPSS version 14, Chicago, Ill).
3. Results Among 194 patients eligible for this study, results from 37 patients (19.1%) were excluded because of incomplete surgical/histopathological data or follow-up (n = 17), major image artifacts (n = 12), incomplete protocol (n = 5) or failure of the contrast material injection (n = 3) during MRI. Among the 157 analysable patients enrolled in the study, the mean age was 49.6 years; median age was 43 years. Lesions detected by MRI and confirmed included n = 22 primary or secondary tumors (n = 20 surgically confirmed), n = 35 root avulsions and brachial plexus cord injuries (n = 30 surgically confirmed), n = 4 entrapment syndromes (n = 2 surgically confirmed), n = 4 recurrences after MR post-treatment evaluation (n = 4 surgically confirmed), n = 4 fibrous scars after MR post-treatment evaluation (n = 3 surgically confirmed), n = 4 Parsonnage–Turner syndrome and n = 2 autoimmune neuritis.
Sensitivity 0.875 0.840 0.800 0.800 1 0.810
95% CI 0.789–0.96 0.701–0.915 0.465–0.981 0.740–1 0.761–1 0.697–0.904
Specificity 0.972 0.910 0.857 0.851 0.923 0.914
95% CI 0.790–0.973 0.641–0.933 0.487–0.974 0.750–0.977 0.757–1 0.697–0.904
3.1. Severity of disease In the “mass lesion” group, there were n = 8 primary neurogenic tumors (n = 2 malignant peripheral nerve sheath tumors pT1/N0/M0, n = 6 Schwannomas), n = 9 pT2/N2/M1 recurrent breast cancers and n = 5 pT3/N3/M1 recurrent breast cancers. No cases of Pancoast tumors have been recorded. In the “traumatic injury” group, there were 12 Sunderland V lesions, 4 Sunderland IV lesions and 2 Sunderland III lesions [22]. The “post-treatment” evaluation group included patients (n = 9) who underwent chemo- and/or radiotherapy for breast cancer recurrence and one patient with non-Hodgkin lymphoma. Parsonnage–Turner syndrome and autoimmune neuritis were defined by the referring physicians. For per-patient statistical analysis, lesions were divided into five groups: mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation, and other. Parsonnage–Turner and autoimmune neuritis were considered together for analysis (“other”). The sensitivity and specificity with 95% CI for different types of lesions are reported in Table 3. Positive predictive value, pre-test-probability (the prevalence), negative predictive value, likelihood ratio for positive results, likelihood ratio for negative results and accuracy are reported in Table 4.
3.2. Handling of indeterminate results Indeterminate results as to whether the result was correct or no as reported by the two radiologists were evaluated off-site by a third radiologist. This radiologist was the same who matched the MRI findings and the reference standard reports (A.T.). This evaluation was performed in consensus with a fourth independent radiologist (M.C., with 20 years of experience in musculoskeletal imaging) who was blinded to all clinical and radiologic information. A total of seven MR studies were initially considered indeterminate (n = 2 traumatic injuries; n = 3 post-treatment evaluation; n = 2 inflammatory neuropathies). After consensus evaluation, these MRI studies were included in the analysis. Logistic multiple regression did not identify any significant influence of centre, MR equipment (124 patients evaluated at 1.5 T and 33 patients evaluated at 3 T), abnormality evaluated, or patient age on the accurate detection of the presence of an abnormality (p > 0.05).
3.3. Intra- and inter-observer agreement Kappa values for the intra-observer agreement of the two readers are considered good ( value reader 1: 0.81; 95% CI 0.72–0.92; value reader 2: 0.79; 95% CI 0.73–0.89). Kappa values for the interobserver agreement between the two readers are considered good ( value 0.71; 95% CI 0.51–0.91).
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
2669
Table 4 Positive predictive value (PPV), pre-test-probability (prevalence), negative predictive value (NPV), likelihood ratio for positive results (LH + ), likelihood ratio for negative results (LH−), and accuracy related to Table 3 results. MR Imaging performance indicator related to the results of Table 3
Mass lesion Traumatic injury Entrapment syndromes Post-treatment evaluation Other Total
PPV 0.954 0.913 0.800 0.666 0.857 0.823
Pre-test probability 0.4 0.531 0.417 0.270 0.315 0.331
Fig. 1. A 39-year-old man with left-sided symptoms who underwent brachial plexus MRI following a motorcycle accident. On the right brachial plexus, marked hyperintensity on T2-weighted images has been recorded (arrows). Clinical follow-up and electrophysiological examination were negative. This case has been considered a false positive.
3.4. Safety Adverse reactions to contrast media were recorded in five patients out of 111/157 who received contrast media. All adverse reactions were mild and resolved within 24 h. A short period of dizziness at the beginning of the examination with the 3-T MR system was reported by two patients. One patient reported a temporary increase of body temperature (37.3 ◦ C) with the 3.0-T MR system. Some examples are presented in figures (Figs. 1–6). 4. Discussion Brachial plexus assessment is still a great challenge: difficulties are encountered in patient management, timing, the types of investigations to be requested, and the indication for surgery [4,5]. Some brachial plexopathies cannot be definitively confirmed or localised without imaging studies. The role of imaging assessment is cru-
Fig. 2. 83-year-old woman who underwent brachial plexus MRI following radiotherapy and chemotherapy for breast cancer with axillary nodal metastasis. On the left brachial plexus, a speculated mass with suspicious arterial and neural invasion has been reported. The initial diagnostic hypothesis was breast cancer recurrence (arrows). Surgical exploration confirmed the presence of the tumor. This case has been considered as a true positive.
NPV 0.921 0.833 0.857 0.920 1 0.905
LH+ 31.499 9.240 5.599 5.400 13.000 9.432
LH− 0.128 0.176 0.233 0.234 0 0.210
Accuracy 0.933 0.872 0.833 0.837 0.947 0.878
cial, especially at the early stage of some diseases or in patients with mild plexopathies, and the role of MRI to study the brachial plexus is under evaluation [4,18–21]. Diagnostic performance evaluation of brachial plexus MRI in a clinical setting is important to avoid underutilisation or inappropriate use of MRI [4]. Therefore, this study was planned to evaluate brachial plexus MRI accuracy in a clinical setting. The study included three different centres and different MRI equipment. A total of 157 patients with complete reference standard data represent one of the largest cohorts evaluated in the radiological literature Fig. 7. Concerning inclusion criteria, a potential referral bias towards the three specialty centres is an unavoidable aspect of the data. However, defined and precise indications to brachial plexus MRI have not been clearly defined: the referring clinician evaluation and clinical thinking is critical in referring a patient to the MRI unit. The clinical evaluation of the patients considered in this study affects the prevalence of the disease and therefore the evaluation of diagnostic accuracy (positive and negative predictive values) of the index test (brachial plexus MRI). In this study, the pre-test probability, which is the prevalence, ranged between 0.270 for the post-treatment evaluation group to the 0.531 for the traumatic injuries group. The data related to the group with the lowest value of prevalence may have at least two meanings: the first is that referring clinicians ask for MRI even when the chance of detecting an abnormality is low. The second is that MRI performed after chemo- and radio-therapy did no detect any abnormality in cases of a complete response to treatment. This data is concordant to the increasing use of imaging modalities (MRI and CT) in oncology to monitor therapies. Another issue worthy of note is that patient recruitment has been performed in tertiary care centres. This is a spectrum bias influencing the predictive values of the test (brachial plexus MRI). Another factor influencing diagnostic performance is the spectrum of patients examined: in tertiary centres, patients are more likely to have advanced or florid disease, whereas in primary or secondary centres, the number of patients with early disease is higher. In this study, there was a balance between advanced and early diseases. This is evident, especially for the “mass lesion” group. Primary malignant neurogenic tumors were pT1, whereas recurrent breast cancers were pT2 and pT3, with a majority of pT2 lesions. In the “traumatic injury” group, there were 12 Sunderland V lesions, 4 Sunderland IV lesions and 2 Sunderland III lesions. These lesions represent advanced disease and are likely to result in a positive MRI even with the standard protocol employed in this study. In the traumatic setting, patients often have multiple lesions; therefore, it has been decided to assign a false negative if the reference standard recorded more lesions than those described by MRI. This is the reason why a relatively high number of false negatives have been reported. We arbitrary divided lesions into five groups: mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation and Parsonnage–Turner and autoimmune neuritis were considered together for analysis (“other”). This division was largely based on the clinical data available after the analysis of the images ended. On the base of the established criteria, some overlap exists between the
2670
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
Fig. 3. 55-year-old man who underwent brachial plexus MRI following surgery for a soft-tissue sarcoma at the level of the forearm and sensory disturbances at the level of the first three fingers. The rounded slightly hyperintense dot on fluid sensitive images (arrow) and isointense to the muscle on T1-weighted images (on the right) was misinterpreted as a nerve lesion. Surgical exploration identified a reactive rounded lymph node. This case was considered a false positive.
five group of patients, for example T2 hyper-intensity for traumatic stretching injuries and neuritis or loss of fascicular echotexture in tumors, entrapment syndromes and traumas. Moreover, intramuscular edema may be present in traumas and neuritis. However, we believe that it is helpful to give separate data for the different clinical conditions. New researches are needed to find out if new MRI sequences are helpful for differential diagnosis. Concerning the reference-standard, a clinical follow-up of 12 months has been considered sufficient for a true negative MRI examination or to consider a mass lesion stable. For entrapment syndromes, Parsonage–Turner syndrome and autoimmune neuritis, the reference standard was the sum of clinical, electrodiagnostic and laboratory data. This is a potential limitation because these syndromes are relatively rare and do not have a “perfect” reference standard. In this study, both 3.0-T and 1.5-T MRI systems were used. Logistic regression analysis excluded the possibility that a 3.0-T MRI yielded better results. These data are in agreement with preliminary data on 3.0-T brachial plexus MRI that showed that, in a series of 30 patients, pathological findings have been demonstrated equally well with both field strengths [21]. Further research
Fig. 4. 56-year-old man who underwent brachial plexus MRI for suspected immune mediated neuritis. Serum analysis was positive for anti-nucleus antibody. Anti-GM1 and anti-MAG antibody detection was negative. Chest-X rays did not reveal any cervical rib or other abnormalities. No sensory impairment was present. EMG testing revealed spontaneous activity in the supraspinatus, and infraspinatus. Cervical MRI was normal. Brachial plexus MRI performed few days after admission showed no root compression with signs of intramuscular edema and atrophy at the level of the supraspinatus and infraspinatus (asterisk). Nerve hyperintensity was also recorded (arrow). The final diagnosis was Parsonage–Turner syndrome, an idiopathic brachial plexus neuritis. This case was considered as a true positive.
is warranted to assess whether diagnostic accuracy of 3.0-T in brachial plexus MRI is increased in larger series of patients, and if it changes patient management. In the statistical analysis, diagnostic accuracy has been calculated on a patient-by-patient basis to stress the possible influence on clinical impact of brachial plexus MRI. In clinical practice, clinicians may not make full use of the information provided by imaging studies. For example, knowing the exact number of nerves involved in case of traumatic neuropathy may not change the management of a young patient with more than one nerve involved. Because diagnostic research should considered not only the accuracy of diagnostic tests but also their practical clinical value, the analysis has not been performed lesion-by-lesion. A limitation of this study is that surgical/histologic proof was not achieved in every patient. In the “mass lesion” group, brachial plexus MRI obtained very high values of sensitivity and specificity. There was one false positive due to a reactive rounded lymph node misinterpreted as a nerve lesion in a patients with a history of soft-tissue sarcoma. This pitfall is difficult to be avoided especially when the finding is near the neurovascular bundle. Standard fluid sensitive sequences were not helpful in this case and we believe that new sequences such as fibre tractography may help in differentiating neural from nonneural structures [19]. Among false negative findings it is worthy of mention the case of a patient operated on for a malignant peripheral nerve tumor in which MRI depicted only cystic changes after surgery, but failed to identify the residual tumor. In these cases the role of new advanced MRI sequences may be important. The highest rate of false positive MRI was registered in the “post-treatment” evaluation group where MRI failed to differentiate recurrent cancer
Fig. 5. 56-year-old woman who underwent brachial plexus MRI for a benign nerve tumor (Schwannoma) at the level of C7 previously seen on ultrasound. MRI depicted the tumor (arrow) and the relationship with the adjacent soft tissues.
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
2671
Fig. 6. 56-year-old woman who underwent brachial plexus MRI after surgery for malignant peripheral nerve tumor. MRI depicted a post-surgical cystic formation with no evidence of tumor. The patient had recurrence after 6 months. This case was considered as a false negative.
from radiation fibrosis. These data are worst than those previously reported for breast cancer patients, but differences in equipment, patient selection, therapies, and image timing may have influenced the results. We believe that also in this clinical setting advanced MRI sequences, such as DWI and DTI, may increase the diagnostic accuracy of MRI. Among patients with Parsonnage–Turner syndrome and autoimmune neuritis, brachial plexus MRI identified every patient affected. One false positive has been reported relative to a marked T2 hyperintensity of the brachial plexus nerve, which was misinterpreted as pathologic. We explain this false positive with two considerations. The first is that increased signal intensity of the brachial plexus nerve on fluid-sensitive images may be present without a clear relation to pathology. The second is that MRI detected a subclinical abnormality before the reference standard became positive.
Considering the entire examination, high NPV means that brachial plexus MRI only rarely misclassifies a sick patient as being healthy. The relatively low positive predictive value in the “posttreatment group” confirms that some of the positive results are false positives and that it is necessary to follow-up any positive result with a more reliable test (biopsy?) or to select patients more appropriately. The inappropriate use of brachial plexus MRI must also be avoided, considering that the test is rather expensive. Regarding the sample size, 95% confidence interval values suggest that an accurate estimate of the overall diagnostic accuracy has been reached as a whole. A limitation of this study is that sample size calculation was not performed at the planning stage. However, the relatively narrow confidence intervals of sensitivity and specificity in study groups as a whole suggest that the sample size of this study is sufficient.
Fig. 7. 63-year-old woman who underwent brachial plexus MRI after chemo- and radio- therapy for lymphoma. At the level of the costo-clavear space the nerves are enlarged with loss of the normal fascicular pattern. A scarring area, hypointense on T1-weighted images and slightly hyperintense on fluid sensitive sequences determines and encasement of the nerves. Clinical follow-up confirmed that this was a true negative finding in which MRI correctly differentiated fibrosis and scarring tissue from recurrence.
2672
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
4.1. Clinical applicability of the study findings and new horizons Considering that the main limitation of this study is its retrospective nature, the overall diagnostic accuracy of brachial plexus MRI calculated on a per-patient basis is relatively high. The specificity of brachial plexus MRI in patients suspected of having a space-occupying mass is very high. The sensitivity is also high, and a few false-positive cases must be taken into account, especially if a careful patient selection has not been performed by the referring clinician. The results of this study open new research horizons: for example, the assessment of a possible increase in diagnostic accuracy related to the introduction of new diffusion-weighted sequences. Furthermore, it should be important to identify the pathological conditions in which brachial plexus MRI is more likely to give clinically useful information. Finally, the clinical impact of brachial plexus MRI in defining which group of patients will benefit from modifications of the approaches to therapeutic management and prognosis must be evaluated. These studies are critical to rationalise the use of imaging studies and to avoid unnecessary and expensive MRI studies. Conflict of interest This piece of the submission is being sent via mail.
Acknowledgments The authors would like to thank all of the residents, radiographers and medical students who participated in data collection. We would like to thank Luisa Altafini, MD; Isabella Garello, MD; Federigo Palmieri, MD; Alessandra Marchetti, MD; Matteo Ghidara, MD; Daniele Pace, MD.
References [1] Nardin RA, Patel MR, Gudas TF, Rutkove SB, Raynor EM. Electromyography MRI in the evaluation of radiculopathy. Muscle Nerve 1999;22:151–5. [2] Bowen BC, Seidenwurm DJ. Plexopathy. Expert panel on neurologic imaging. AJNR Am J Neuroradiol 2008;29:400–2. [3] Vargas MI, Viallon M, Nguyen D, Beaulieu JY, Delavelle J, Becker M. New approaches in imaging of the brachial plexus. Eur J Radiol 2010;74:403–10.
[4] Martinoli C, Gandolfo N, Perez MM, Klauser A, Palmieri F, Padua L, Tagliafico A. Brachial plexus and nerves about the shoulder. Semin Musculoskelet Radiol 2010;14:523–46. [5] Linda DD, Harish S, Stewart BG, Finlay K, Parasu N, Rebello RP. Multimodality imaging of peripheral neuropathies of the upper limb and brachial plexus. Radiographics 2010;30(5):1373–400. [6] Hayes CE, Tsuruda JS, Mathis CM, Maravilla KR, Kliot M, Filler AG. Brachial plexus: MR imaging with a dedicated phased array of surface coils. Radiology 1997;203:286–9. [7] Qayyum A, MacVicar AD, Padhani AR, Revell P, Husband JE. Symptomatic brachial plexopathy following treatment for breast cancer: utility of MR imaging with surface-coil techniques. Radiology 2000;214:837–42. [8] Bradley LJ, Wilhelm T, King RH, Ginsberg L, Orrell RW. Brachial plexus hypertrophy in chronic inflammatory demyelinating polyradiculoneuropathy. Neuromuscul Disord 2006;16:126–31. [9] Gaskin CM, Helms CA. Parsonage–Turner syndrome: MR imaging findings and clinical information of 27 patients. Radiology 2006;240:501–7. [10] Scalf RE, Wenger DE, Frick MA, Mandrekar JN, Adkins MC. MRI findings of 26 patients with Parsonage–Turner syndrome. AJR Am J Roentgenol 2007;189:W39–44. [11] Choi JY, Kang CH, Kim BJ, Park KW, Yu SW. Brachial plexopathy following herpes zoster infection: two cases with MRI findings. J Neurol Sci 2009;285:224–6. [12] Bilbey JH, Lamond RG, Mattrey RF. MR imaging of disorders of the brachial plexus. J Magn Reson Imaging 1994;4:13–8. [13] Adachi Y, Sato N, Okamoto T, Sasaki M, Komaki H, Yamashita F, et al. Brachial and lumbar plexuses in chronic inflammatory demyelinating polyradiculoneuropathy: MRI assessment including apparent diffusion coefficient. Neuroradiology 2010 [Epub ahead of print]. [14] Aralasmak A, Karaali K, Cevikol C, Uysal H, Senol U. MR imaging findings in brachial plexopathy with thoracic outlet syndrome. AJNR Am J Neuroradiol 2010;31(3):410–7. [15] Viallon M, Vargas MI, Jlassi H, Lövblad KO, Delavelle J. High-resolution and functional magnetic resonance imaging of the brachial plexus using an isotropic 3D T2 STIR (short term inversion recovery) SPACE sequence and diffusion tensor imaging. Eur Radiol 2008;18:1018–23. [16] www.stard-statement.org (accessed 09.03.11). [17] Millesi H. Brachial plexus injuries. In: Jupiter JB, editor. Flynn’s hand surgery. 4th ed. Baltimore, MD: Williams & Wilkins; 1991. p. 457–63. [18] Bowen B, Pattany P, Saraf-Lavi E, Maravilla KR. The brachial plexus: normal anatomy, pathology, and MR imaging. Neuroimaging Clin N Am 2004;14:59–85. [19] Tagliafico A, Calabrese M, Puntoni M, Pace D, Baio G, Neumaier CE, et al. Brachial plexus MR imaging: accuracy and reproducibility of DTI-derived measurements and fibre tractography at 3.0-T. Eur Radiol 2011;21(8): 1764–71. [20] Tagliafico A, Succio G, Neumaier CE, Baio G, Serafini G, Ghidara M, et al. Brachial plexus assessment with three-dimensional isotropic resolution fast spin echo MRI: comparison with conventional MRI at 3.0T. Br J Radiol 2011 [Epub ahead of print] PubMed PMID: 21343321. [21] Tagliafico A, Succio G, Neumaier CE, Serafini G, Ghidara M, Calabrese M, et al. MR imaging of the brachial plexus: comparison between 1.5-T and 3-T MR imaging: preliminary experience. Skeletal Radiol 2010;40(6): 717–24. [22] Mark S, Greenberg. Injury classification system. Handbook of Neurosurgery, 41, 3rd ed. 1994. p. 1–412.
Contents lists available at SciVerse ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Diagnostic accuracy of MRI in adults with suspect brachial plexus lesions: A multicentre retrospective study with surgical findings and clinical follow-up as reference standard Alberto Tagliafico a,∗ , Giulia Succio b , Giovanni Serafini b , Carlo Martinoli c a
Institute of Anatomy, Department of Experimental Medicine, University of Genoa, Largo Rosanna Benzi 8, 16132 Genoa, Italy Department of Radiology, Santa Corona Hospital, Pietra Ligure, Italy via XXV Aprile, 38- Pietra Ligure, 17027 Savona, Italy c Radiology Department, DISC, Università di Genova, Largo Rosanna Benzi 8, 16138 Genova, Italy b
a r t i c l e
i n f o
Article history: Received 7 September 2011 Received in revised form 12 October 2011 Accepted 14 October 2011 Keywords: Brachial Plexus MRI Sensitivity Specificity Accuracy
a b s t r a c t Objective: To evaluate brachial plexus MRI accuracy with surgical findings and clinical follow-up as reference standard in a large multicentre study. Materials and methods: The research was approved by the Institutional Review Boards, and all patients provided their written informed consent. A multicentre retrospective trial that included three centres was performed between March 2006 and April 2011. A total of 157 patients (men/women: 81/76; age range, 18–84 years) were evaluated: surgical findings and clinical follow-up of at least 12 months were used as the reference standard. MR imaging was performed with different equipment at 1.5 T and 3.0 T. The patient group was divided in five subgroups: mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation, and other. Sensitivity, specificity with 95% confidence intervals (CIs), positive predictive value (PPV), pre-testprobability (the prevalence), negative predictive value (NPV), pre- and post-test odds (OR), likelihood ratio for positive results (LH+), likelihood ratio for negative results (LH−), accuracy and post-test probability (post-P) were reported on a per-patient basis. Results: The overall sensitivity and specificity with 95% CIs were: 0.810/0.914; (0.697–0.904). Overall PPV, pre-test probability, NPV, LH+, LH−, and accuracy: 0.823, 0.331, 0.905, 9.432, 0.210, 0.878. Conclusions: The overall diagnostic accuracy of brachial plexus MRI calculated on a per-patient base is relatively high. The specificity of brachial plexus MRI in patients suspected of having a space-occupying mass is very high. The sensitivity is also high, but there are false-positive interpretations as well. © 2011 Elsevier Ireland Ltd. All rights reserved.
1. Introduction The brachial plexus provides motor and sensory innervations to the upper extremities. Pathologic conditions involving the brachial plexus are relatively uncommon and frequently misdiagnosed with disastrous consequences [1]. Brachial plexopathy causes weakness, sensory loss, and loss of reflexes in body regions innervated by the C5-T1 segmental distribution [2]. The clinical diagnosis is usually confirmed by electrodiagnostic studies (EMG). Clinical evaluation of this anatomical region is a real challenge, and electrophysiological studies may not be sufficient to identify and locate the cause of brachial plexus impairment. Currently, MR imaging (MRI) is considered the evaluation of choice because of its multi-planar
∗ Corresponding author. Tel.: +39 0103537885/3479745122; fax: +39 0103537882. E-mail addresses: alberto.tagliafi[email protected], albertotagliafi[email protected] (A. Tagliafico). 0720-048X/$ – see front matter © 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ejrad.2011.10.007
capabilities and excellent soft-tissue contrast [3]. However, MRI has inherent limitations in specific clinical settings and can be variably supported by other modalities depending on the underlying clinical problem. The use of MRI also depends on the management policy of the referring clinicians and peripheral nerve surgeons [4]. Electromyography and MRI are often used as complementary tools to assess brachial plexopathy [1]. Previous studies have evaluated the diagnostic accuracy of brachial plexus MRI in traumatic root avulsions [4], in brachial plexopathy, and in primary or secondary tumors [4–7]. Different studies have demonstrated that brachial plexus MRI is useful in chronic inflammatory demyelinating polyradiculo-neuropathy [8], Parsonage-Turner syndrome, herpes zoster infection and thoracic outlet syndrome [9–14]. Reported values of MRI diagnostic accuracy range between 52% and 88% for traumatic root avulsions and reach 96% of sensitivity for tumor detection [5–14]. However, most of the published papers on brachial plexus MRI evaluated small series or selected pathological conditions [9–15]. Moreover, in the last decade, the diagnostic workup of brachial plexus pathology has been
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
2667
Table 1 Parameters used for MR imaging of the brachial plexus at 3-T and 1.5-T. T1-w tSE sequences
Repetition time (ms) Echo time (ms) Flip angle Voxel size (mm) Matrix Slice thickness (mm) Phase resolution Echo train length per slice Bandwidth Hz/px SAR (W/kg body weight) Acquisition time
T2-w tSE sequences
3-T MR imaging
1.5-T MR imaging
3-T MR imaging
1.5-T MR imaging
600 9.1 90 0.2 × 0.2 × 2 320 × 320 4 100 3 122 1.8 5 min.14 s
500 14
4220 89 90 0.2 × 0.2 × 2 320 × 320 4 100 17 162 1.6 5 min.10 s
3300 63
0.2 × 0.2 × 2 320 × 320 4 100 5 124 1.2 5 min.45 s
continually reviewed. Testing the diagnostic accuracy of current imaging strategies such as MRI is still important to avoid the underutilisation or inappropriate use of the available imaging techniques [4]. Therefore, the purpose of this study is to evaluate brachial plexus MRI accuracy in a large multicentre study with surgical findings and clinical follow-up as reference standard. 2. Materials and methods The research was approved by the Institutional Review Board. Patients provided their written informed consent to use of the data for statistical analysis and publication. This study of diagnostic accuracy was planned according to the Standards for Reporting of Diagnostic Accuracy (STARD) guidelines [16]. 2.1. Patients A multicentre retrospective trial was performed between March 2006 and April 2011, including three centres with proven imaging experience in brachial plexus MRI (at least 20 brachial plexus MR examinations per year and at least 1000 body MR examinations per year) and MR equipment with a 1.5–3.0-T magnet, 15 mT/m gradients, and at least 12 coil elements obtained by the combination of body, neck and spine-array coils. A total of 194 patients were enrolled; the centres provided 32, 88, and 74 cases, respectively. Patients 18 years old or older with suspicious brachial plexus lesions referred by neurologists, neurosurgeons and oncologists at each enrolling centre and surgical findings (obtained no later than 30 days after MR evaluation) or clinical follow-up of at least 12 months were used as the reference standards. Surgical findings included the reports for brachial plexus traumas and the results of biopsy for tumoral lesions. A level-by-level analysis based upon level-by-level exploration has not been done. Clinical follow-up included clinical and electrophysiological evaluation performed by the referring physicians as described elsewhere [17–19]. The exclusion criteria were absolute contraindications to MR, pregnancy or breastfeeding, severe renal failure, known hypersensitivity to gadolinium chelates and clinical status limiting data reliability. 2.2. MRI The imaging protocol included T1-weighted turbo spin-echo (tSE) sequences in the coronal, transverse and sagittal oblique planes and T2-weighed turbo spin-echo (tSE) sequences with fat saturation in the coronal, transverse and sagittal oblique planes (as perpendicular as possible to the brachial plexus nerves). T1weighted turbo spin-echo sequences with fat saturation following
0.2 × 0.2 × 2 320 × 320 4 100 14 124 1.3 6 min.36 s
gadolinium (0.1 mmol/kg of body weight administered at the rate of 2 mL/s, using an automatic injector) were obtained in patients suspected to have neoplasm, radiation injury, inflammation, or abscess, or nerve entrapment and stretch injury. Three different contrast media were used (MultiHance, Bracco Imaging, Milan, Italy; Magnevist, Bayer HealthCare, Berlin, Germany; Doterem, Guerbet, Villepinte, France). The brachial plexus MRI protocols employed in this study have been previously used at 1.5 T and 3.0 T [18–21]. Technical parameters for all sequences are summarised in Table 1. The imaging time for a complete examination was approximately 40 min. 2.3. Analysis of cases Cases with incomplete surgical/histopathological data or follow-up and cases with important imaging artifacts (e.g., a patient’s movement) were excluded from the analysis. At the central unit (off-site), two experienced radiologists with 7 and 9 years of experience in MRI checked all the examinations with respect to the protocol and the quality of the images. As a consequence, 37 patients were excluded from the analysis, and 157 patients remained for analysis. For these 157 cases (men/women: 81/76; age range, 18–84 years), an off-site interpretation of the MR images was performed independently by the two experienced radiologists, who were aware of the entry criterion and blinded to all of the patients’ data and to the results of surgical findings and clinical follow-up (G.S and C.M.). The two reviewers interpreted the MR examinations, blinded to each other and to the clinical data of the patients. Image analysis was repeated after 1 month to assess intra-observer agreement. All images were transferred to an external workstation equipped with Osirix software (Osirix 3.6 v 32-bit) and evaluated in a random fashion; pathological findings on MRI were evaluated on the basis of established criteria (Table 2) [2,4,20,21]. Comparison of brachial plexus MRI findings and the reference standard was performed offsite by a third radiologist (A.T.) who matched MR and reference standard reports, creating a database for the data analysis. An abnormality was considered to be depicted successfully (true-positive) if concordance was present between reported MRI findings and the reference standard. Abnormalities revealed by the
Table 2 Normal vs. abnormal brachial plexus, see references [2] and [4]. Abnormality on Brachial Plexus MRI Loss of fat planes around all or part of a plexus component Diffuse or focal enlargement of a component (especially, the presence of an eccentric or nodular mass) Marked hyperintensity on T2-weighted images and/or enhancement on T1-weighted images with fat suppression Altered fascicular pattern
2668
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
Table 3 Sensitivity and specificity with 95% Confidence Intervals (CIs) of the five groups of lesions examined. Per-patient MR imaging findings for the group overall
Mass lesion Traumatic injury Entrapment syndromes Post-treatment evaluation Other Total
TP 21 12 4 8 6 42
FP 1 2 1 4 1 9
FN 3 4 1 2 0 10
TN 35 20 6 23 12 96
reference standard but not MRI were classified as false-negative findings for the modality (e.g., subtle stretching injuries with positive reference standard and negative MRI). Because of the differences between brachial plexus lesions, the statistical analysis was performed on each subgroup (mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation, other) and on the entire group of lesions as a whole. This process was done after the initial analysis. Analysis was performed on a patient-by-patient basis. With regard to the surgical reports for brachial plexus trauma, only a correspondence between the area affected at surgical exploration and the area where MRI was performed provided additional information about the disease. A level-by-level analysis based upon level-by-level exploration was not performed. This process results in much higher sensitivity/specificity. To describe brachial plexus MRI performance, the sensitivity and specificity have been reported with their respective 95% confidence intervals (CIs). Positive predictive value, pre-test-probability (the prevalence), negative predictive value, likelihood ratio for positive results, likelihood ratio for negative results and accuracy were also reported. Logistic multiple regression was used to analyse the possible effect of independent variables, such as centre, MR equipment, abnormality, and patient age on accurate detection of the presence of an abnormality. Intra- and Inter-reader agreement in detecting or assessing lesion nature was determined by using generalised weighted statistics and was classified as excellent ( values > 0.80), good ( = 0.61–0.80), moderate ( = 0.41–0.60), fair ( = 0.21–0.40), or poor ( ≤ 0.20). All statistical tests were two sided at the P < .05 level of significance and were performed by using dedicated software (SPSS version 14, Chicago, Ill).
3. Results Among 194 patients eligible for this study, results from 37 patients (19.1%) were excluded because of incomplete surgical/histopathological data or follow-up (n = 17), major image artifacts (n = 12), incomplete protocol (n = 5) or failure of the contrast material injection (n = 3) during MRI. Among the 157 analysable patients enrolled in the study, the mean age was 49.6 years; median age was 43 years. Lesions detected by MRI and confirmed included n = 22 primary or secondary tumors (n = 20 surgically confirmed), n = 35 root avulsions and brachial plexus cord injuries (n = 30 surgically confirmed), n = 4 entrapment syndromes (n = 2 surgically confirmed), n = 4 recurrences after MR post-treatment evaluation (n = 4 surgically confirmed), n = 4 fibrous scars after MR post-treatment evaluation (n = 3 surgically confirmed), n = 4 Parsonnage–Turner syndrome and n = 2 autoimmune neuritis.
Sensitivity 0.875 0.840 0.800 0.800 1 0.810
95% CI 0.789–0.96 0.701–0.915 0.465–0.981 0.740–1 0.761–1 0.697–0.904
Specificity 0.972 0.910 0.857 0.851 0.923 0.914
95% CI 0.790–0.973 0.641–0.933 0.487–0.974 0.750–0.977 0.757–1 0.697–0.904
3.1. Severity of disease In the “mass lesion” group, there were n = 8 primary neurogenic tumors (n = 2 malignant peripheral nerve sheath tumors pT1/N0/M0, n = 6 Schwannomas), n = 9 pT2/N2/M1 recurrent breast cancers and n = 5 pT3/N3/M1 recurrent breast cancers. No cases of Pancoast tumors have been recorded. In the “traumatic injury” group, there were 12 Sunderland V lesions, 4 Sunderland IV lesions and 2 Sunderland III lesions [22]. The “post-treatment” evaluation group included patients (n = 9) who underwent chemo- and/or radiotherapy for breast cancer recurrence and one patient with non-Hodgkin lymphoma. Parsonnage–Turner syndrome and autoimmune neuritis were defined by the referring physicians. For per-patient statistical analysis, lesions were divided into five groups: mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation, and other. Parsonnage–Turner and autoimmune neuritis were considered together for analysis (“other”). The sensitivity and specificity with 95% CI for different types of lesions are reported in Table 3. Positive predictive value, pre-test-probability (the prevalence), negative predictive value, likelihood ratio for positive results, likelihood ratio for negative results and accuracy are reported in Table 4.
3.2. Handling of indeterminate results Indeterminate results as to whether the result was correct or no as reported by the two radiologists were evaluated off-site by a third radiologist. This radiologist was the same who matched the MRI findings and the reference standard reports (A.T.). This evaluation was performed in consensus with a fourth independent radiologist (M.C., with 20 years of experience in musculoskeletal imaging) who was blinded to all clinical and radiologic information. A total of seven MR studies were initially considered indeterminate (n = 2 traumatic injuries; n = 3 post-treatment evaluation; n = 2 inflammatory neuropathies). After consensus evaluation, these MRI studies were included in the analysis. Logistic multiple regression did not identify any significant influence of centre, MR equipment (124 patients evaluated at 1.5 T and 33 patients evaluated at 3 T), abnormality evaluated, or patient age on the accurate detection of the presence of an abnormality (p > 0.05).
3.3. Intra- and inter-observer agreement Kappa values for the intra-observer agreement of the two readers are considered good ( value reader 1: 0.81; 95% CI 0.72–0.92; value reader 2: 0.79; 95% CI 0.73–0.89). Kappa values for the interobserver agreement between the two readers are considered good ( value 0.71; 95% CI 0.51–0.91).
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
2669
Table 4 Positive predictive value (PPV), pre-test-probability (prevalence), negative predictive value (NPV), likelihood ratio for positive results (LH + ), likelihood ratio for negative results (LH−), and accuracy related to Table 3 results. MR Imaging performance indicator related to the results of Table 3
Mass lesion Traumatic injury Entrapment syndromes Post-treatment evaluation Other Total
PPV 0.954 0.913 0.800 0.666 0.857 0.823
Pre-test probability 0.4 0.531 0.417 0.270 0.315 0.331
Fig. 1. A 39-year-old man with left-sided symptoms who underwent brachial plexus MRI following a motorcycle accident. On the right brachial plexus, marked hyperintensity on T2-weighted images has been recorded (arrows). Clinical follow-up and electrophysiological examination were negative. This case has been considered a false positive.
3.4. Safety Adverse reactions to contrast media were recorded in five patients out of 111/157 who received contrast media. All adverse reactions were mild and resolved within 24 h. A short period of dizziness at the beginning of the examination with the 3-T MR system was reported by two patients. One patient reported a temporary increase of body temperature (37.3 ◦ C) with the 3.0-T MR system. Some examples are presented in figures (Figs. 1–6). 4. Discussion Brachial plexus assessment is still a great challenge: difficulties are encountered in patient management, timing, the types of investigations to be requested, and the indication for surgery [4,5]. Some brachial plexopathies cannot be definitively confirmed or localised without imaging studies. The role of imaging assessment is cru-
Fig. 2. 83-year-old woman who underwent brachial plexus MRI following radiotherapy and chemotherapy for breast cancer with axillary nodal metastasis. On the left brachial plexus, a speculated mass with suspicious arterial and neural invasion has been reported. The initial diagnostic hypothesis was breast cancer recurrence (arrows). Surgical exploration confirmed the presence of the tumor. This case has been considered as a true positive.
NPV 0.921 0.833 0.857 0.920 1 0.905
LH+ 31.499 9.240 5.599 5.400 13.000 9.432
LH− 0.128 0.176 0.233 0.234 0 0.210
Accuracy 0.933 0.872 0.833 0.837 0.947 0.878
cial, especially at the early stage of some diseases or in patients with mild plexopathies, and the role of MRI to study the brachial plexus is under evaluation [4,18–21]. Diagnostic performance evaluation of brachial plexus MRI in a clinical setting is important to avoid underutilisation or inappropriate use of MRI [4]. Therefore, this study was planned to evaluate brachial plexus MRI accuracy in a clinical setting. The study included three different centres and different MRI equipment. A total of 157 patients with complete reference standard data represent one of the largest cohorts evaluated in the radiological literature Fig. 7. Concerning inclusion criteria, a potential referral bias towards the three specialty centres is an unavoidable aspect of the data. However, defined and precise indications to brachial plexus MRI have not been clearly defined: the referring clinician evaluation and clinical thinking is critical in referring a patient to the MRI unit. The clinical evaluation of the patients considered in this study affects the prevalence of the disease and therefore the evaluation of diagnostic accuracy (positive and negative predictive values) of the index test (brachial plexus MRI). In this study, the pre-test probability, which is the prevalence, ranged between 0.270 for the post-treatment evaluation group to the 0.531 for the traumatic injuries group. The data related to the group with the lowest value of prevalence may have at least two meanings: the first is that referring clinicians ask for MRI even when the chance of detecting an abnormality is low. The second is that MRI performed after chemo- and radio-therapy did no detect any abnormality in cases of a complete response to treatment. This data is concordant to the increasing use of imaging modalities (MRI and CT) in oncology to monitor therapies. Another issue worthy of note is that patient recruitment has been performed in tertiary care centres. This is a spectrum bias influencing the predictive values of the test (brachial plexus MRI). Another factor influencing diagnostic performance is the spectrum of patients examined: in tertiary centres, patients are more likely to have advanced or florid disease, whereas in primary or secondary centres, the number of patients with early disease is higher. In this study, there was a balance between advanced and early diseases. This is evident, especially for the “mass lesion” group. Primary malignant neurogenic tumors were pT1, whereas recurrent breast cancers were pT2 and pT3, with a majority of pT2 lesions. In the “traumatic injury” group, there were 12 Sunderland V lesions, 4 Sunderland IV lesions and 2 Sunderland III lesions. These lesions represent advanced disease and are likely to result in a positive MRI even with the standard protocol employed in this study. In the traumatic setting, patients often have multiple lesions; therefore, it has been decided to assign a false negative if the reference standard recorded more lesions than those described by MRI. This is the reason why a relatively high number of false negatives have been reported. We arbitrary divided lesions into five groups: mass lesion, traumatic injury, entrapment syndromes, post-treatment evaluation and Parsonnage–Turner and autoimmune neuritis were considered together for analysis (“other”). This division was largely based on the clinical data available after the analysis of the images ended. On the base of the established criteria, some overlap exists between the
2670
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
Fig. 3. 55-year-old man who underwent brachial plexus MRI following surgery for a soft-tissue sarcoma at the level of the forearm and sensory disturbances at the level of the first three fingers. The rounded slightly hyperintense dot on fluid sensitive images (arrow) and isointense to the muscle on T1-weighted images (on the right) was misinterpreted as a nerve lesion. Surgical exploration identified a reactive rounded lymph node. This case was considered a false positive.
five group of patients, for example T2 hyper-intensity for traumatic stretching injuries and neuritis or loss of fascicular echotexture in tumors, entrapment syndromes and traumas. Moreover, intramuscular edema may be present in traumas and neuritis. However, we believe that it is helpful to give separate data for the different clinical conditions. New researches are needed to find out if new MRI sequences are helpful for differential diagnosis. Concerning the reference-standard, a clinical follow-up of 12 months has been considered sufficient for a true negative MRI examination or to consider a mass lesion stable. For entrapment syndromes, Parsonage–Turner syndrome and autoimmune neuritis, the reference standard was the sum of clinical, electrodiagnostic and laboratory data. This is a potential limitation because these syndromes are relatively rare and do not have a “perfect” reference standard. In this study, both 3.0-T and 1.5-T MRI systems were used. Logistic regression analysis excluded the possibility that a 3.0-T MRI yielded better results. These data are in agreement with preliminary data on 3.0-T brachial plexus MRI that showed that, in a series of 30 patients, pathological findings have been demonstrated equally well with both field strengths [21]. Further research
Fig. 4. 56-year-old man who underwent brachial plexus MRI for suspected immune mediated neuritis. Serum analysis was positive for anti-nucleus antibody. Anti-GM1 and anti-MAG antibody detection was negative. Chest-X rays did not reveal any cervical rib or other abnormalities. No sensory impairment was present. EMG testing revealed spontaneous activity in the supraspinatus, and infraspinatus. Cervical MRI was normal. Brachial plexus MRI performed few days after admission showed no root compression with signs of intramuscular edema and atrophy at the level of the supraspinatus and infraspinatus (asterisk). Nerve hyperintensity was also recorded (arrow). The final diagnosis was Parsonage–Turner syndrome, an idiopathic brachial plexus neuritis. This case was considered as a true positive.
is warranted to assess whether diagnostic accuracy of 3.0-T in brachial plexus MRI is increased in larger series of patients, and if it changes patient management. In the statistical analysis, diagnostic accuracy has been calculated on a patient-by-patient basis to stress the possible influence on clinical impact of brachial plexus MRI. In clinical practice, clinicians may not make full use of the information provided by imaging studies. For example, knowing the exact number of nerves involved in case of traumatic neuropathy may not change the management of a young patient with more than one nerve involved. Because diagnostic research should considered not only the accuracy of diagnostic tests but also their practical clinical value, the analysis has not been performed lesion-by-lesion. A limitation of this study is that surgical/histologic proof was not achieved in every patient. In the “mass lesion” group, brachial plexus MRI obtained very high values of sensitivity and specificity. There was one false positive due to a reactive rounded lymph node misinterpreted as a nerve lesion in a patients with a history of soft-tissue sarcoma. This pitfall is difficult to be avoided especially when the finding is near the neurovascular bundle. Standard fluid sensitive sequences were not helpful in this case and we believe that new sequences such as fibre tractography may help in differentiating neural from nonneural structures [19]. Among false negative findings it is worthy of mention the case of a patient operated on for a malignant peripheral nerve tumor in which MRI depicted only cystic changes after surgery, but failed to identify the residual tumor. In these cases the role of new advanced MRI sequences may be important. The highest rate of false positive MRI was registered in the “post-treatment” evaluation group where MRI failed to differentiate recurrent cancer
Fig. 5. 56-year-old woman who underwent brachial plexus MRI for a benign nerve tumor (Schwannoma) at the level of C7 previously seen on ultrasound. MRI depicted the tumor (arrow) and the relationship with the adjacent soft tissues.
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
2671
Fig. 6. 56-year-old woman who underwent brachial plexus MRI after surgery for malignant peripheral nerve tumor. MRI depicted a post-surgical cystic formation with no evidence of tumor. The patient had recurrence after 6 months. This case was considered as a false negative.
from radiation fibrosis. These data are worst than those previously reported for breast cancer patients, but differences in equipment, patient selection, therapies, and image timing may have influenced the results. We believe that also in this clinical setting advanced MRI sequences, such as DWI and DTI, may increase the diagnostic accuracy of MRI. Among patients with Parsonnage–Turner syndrome and autoimmune neuritis, brachial plexus MRI identified every patient affected. One false positive has been reported relative to a marked T2 hyperintensity of the brachial plexus nerve, which was misinterpreted as pathologic. We explain this false positive with two considerations. The first is that increased signal intensity of the brachial plexus nerve on fluid-sensitive images may be present without a clear relation to pathology. The second is that MRI detected a subclinical abnormality before the reference standard became positive.
Considering the entire examination, high NPV means that brachial plexus MRI only rarely misclassifies a sick patient as being healthy. The relatively low positive predictive value in the “posttreatment group” confirms that some of the positive results are false positives and that it is necessary to follow-up any positive result with a more reliable test (biopsy?) or to select patients more appropriately. The inappropriate use of brachial plexus MRI must also be avoided, considering that the test is rather expensive. Regarding the sample size, 95% confidence interval values suggest that an accurate estimate of the overall diagnostic accuracy has been reached as a whole. A limitation of this study is that sample size calculation was not performed at the planning stage. However, the relatively narrow confidence intervals of sensitivity and specificity in study groups as a whole suggest that the sample size of this study is sufficient.
Fig. 7. 63-year-old woman who underwent brachial plexus MRI after chemo- and radio- therapy for lymphoma. At the level of the costo-clavear space the nerves are enlarged with loss of the normal fascicular pattern. A scarring area, hypointense on T1-weighted images and slightly hyperintense on fluid sensitive sequences determines and encasement of the nerves. Clinical follow-up confirmed that this was a true negative finding in which MRI correctly differentiated fibrosis and scarring tissue from recurrence.
2672
A. Tagliafico et al. / European Journal of Radiology 81 (2012) 2666–2672
4.1. Clinical applicability of the study findings and new horizons Considering that the main limitation of this study is its retrospective nature, the overall diagnostic accuracy of brachial plexus MRI calculated on a per-patient basis is relatively high. The specificity of brachial plexus MRI in patients suspected of having a space-occupying mass is very high. The sensitivity is also high, and a few false-positive cases must be taken into account, especially if a careful patient selection has not been performed by the referring clinician. The results of this study open new research horizons: for example, the assessment of a possible increase in diagnostic accuracy related to the introduction of new diffusion-weighted sequences. Furthermore, it should be important to identify the pathological conditions in which brachial plexus MRI is more likely to give clinically useful information. Finally, the clinical impact of brachial plexus MRI in defining which group of patients will benefit from modifications of the approaches to therapeutic management and prognosis must be evaluated. These studies are critical to rationalise the use of imaging studies and to avoid unnecessary and expensive MRI studies. Conflict of interest This piece of the submission is being sent via mail.
Acknowledgments The authors would like to thank all of the residents, radiographers and medical students who participated in data collection. We would like to thank Luisa Altafini, MD; Isabella Garello, MD; Federigo Palmieri, MD; Alessandra Marchetti, MD; Matteo Ghidara, MD; Daniele Pace, MD.
References [1] Nardin RA, Patel MR, Gudas TF, Rutkove SB, Raynor EM. Electromyography MRI in the evaluation of radiculopathy. Muscle Nerve 1999;22:151–5. [2] Bowen BC, Seidenwurm DJ. Plexopathy. Expert panel on neurologic imaging. AJNR Am J Neuroradiol 2008;29:400–2. [3] Vargas MI, Viallon M, Nguyen D, Beaulieu JY, Delavelle J, Becker M. New approaches in imaging of the brachial plexus. Eur J Radiol 2010;74:403–10.
[4] Martinoli C, Gandolfo N, Perez MM, Klauser A, Palmieri F, Padua L, Tagliafico A. Brachial plexus and nerves about the shoulder. Semin Musculoskelet Radiol 2010;14:523–46. [5] Linda DD, Harish S, Stewart BG, Finlay K, Parasu N, Rebello RP. Multimodality imaging of peripheral neuropathies of the upper limb and brachial plexus. Radiographics 2010;30(5):1373–400. [6] Hayes CE, Tsuruda JS, Mathis CM, Maravilla KR, Kliot M, Filler AG. Brachial plexus: MR imaging with a dedicated phased array of surface coils. Radiology 1997;203:286–9. [7] Qayyum A, MacVicar AD, Padhani AR, Revell P, Husband JE. Symptomatic brachial plexopathy following treatment for breast cancer: utility of MR imaging with surface-coil techniques. Radiology 2000;214:837–42. [8] Bradley LJ, Wilhelm T, King RH, Ginsberg L, Orrell RW. Brachial plexus hypertrophy in chronic inflammatory demyelinating polyradiculoneuropathy. Neuromuscul Disord 2006;16:126–31. [9] Gaskin CM, Helms CA. Parsonage–Turner syndrome: MR imaging findings and clinical information of 27 patients. Radiology 2006;240:501–7. [10] Scalf RE, Wenger DE, Frick MA, Mandrekar JN, Adkins MC. MRI findings of 26 patients with Parsonage–Turner syndrome. AJR Am J Roentgenol 2007;189:W39–44. [11] Choi JY, Kang CH, Kim BJ, Park KW, Yu SW. Brachial plexopathy following herpes zoster infection: two cases with MRI findings. J Neurol Sci 2009;285:224–6. [12] Bilbey JH, Lamond RG, Mattrey RF. MR imaging of disorders of the brachial plexus. J Magn Reson Imaging 1994;4:13–8. [13] Adachi Y, Sato N, Okamoto T, Sasaki M, Komaki H, Yamashita F, et al. Brachial and lumbar plexuses in chronic inflammatory demyelinating polyradiculoneuropathy: MRI assessment including apparent diffusion coefficient. Neuroradiology 2010 [Epub ahead of print]. [14] Aralasmak A, Karaali K, Cevikol C, Uysal H, Senol U. MR imaging findings in brachial plexopathy with thoracic outlet syndrome. AJNR Am J Neuroradiol 2010;31(3):410–7. [15] Viallon M, Vargas MI, Jlassi H, Lövblad KO, Delavelle J. High-resolution and functional magnetic resonance imaging of the brachial plexus using an isotropic 3D T2 STIR (short term inversion recovery) SPACE sequence and diffusion tensor imaging. Eur Radiol 2008;18:1018–23. [16] www.stard-statement.org (accessed 09.03.11). [17] Millesi H. Brachial plexus injuries. In: Jupiter JB, editor. Flynn’s hand surgery. 4th ed. Baltimore, MD: Williams & Wilkins; 1991. p. 457–63. [18] Bowen B, Pattany P, Saraf-Lavi E, Maravilla KR. The brachial plexus: normal anatomy, pathology, and MR imaging. Neuroimaging Clin N Am 2004;14:59–85. [19] Tagliafico A, Calabrese M, Puntoni M, Pace D, Baio G, Neumaier CE, et al. Brachial plexus MR imaging: accuracy and reproducibility of DTI-derived measurements and fibre tractography at 3.0-T. Eur Radiol 2011;21(8): 1764–71. [20] Tagliafico A, Succio G, Neumaier CE, Baio G, Serafini G, Ghidara M, et al. Brachial plexus assessment with three-dimensional isotropic resolution fast spin echo MRI: comparison with conventional MRI at 3.0T. Br J Radiol 2011 [Epub ahead of print] PubMed PMID: 21343321. [21] Tagliafico A, Succio G, Neumaier CE, Serafini G, Ghidara M, Calabrese M, et al. MR imaging of the brachial plexus: comparison between 1.5-T and 3-T MR imaging: preliminary experience. Skeletal Radiol 2010;40(6): 717–24. [22] Mark S, Greenberg. Injury classification system. Handbook of Neurosurgery, 41, 3rd ed. 1994. p. 1–412.