Esophageal cancer is the eighth most common cancer worldwide [1]. In the West, adenocarcinoma is now the predominate form of esophageal cancer, having recently surpassed the incidence of esophageal squamous cell carcinoma [2]. Surgical resection remains the only curative treatment for adenocarcinoma of the esophagus but carries significant morbidity and mortality. It is essential for imaging to accurately identify patients with localized, resectable disease.

Locoregional staging is currently undertaken by CT and endoscopic sonography. The latter is more accurate than CT in determining T stage, with rates between 60% and 80% depending on tumor stage and operator experience [3, 4]. However, endoscopic sonography examinations may be limited by tightly stenosed tumors and differentiation of ulcer-associated inflammation from infiltration [5, 6]. The transducers used are high frequency (7.5-12 MHz) to maximize resolution, but have a restricted sonographic range, limiting visualization of structures that are deep to the tumor. This limits determination of tumor-free surgical resection planes.

To date, MRI has not been used for locoregional staging. An early study [7] comparing conventional CT and 0.35-T MRI found MRI to have lower accuracy for staging esophageal tumors than CT, although a more recent study [8] concluded that MRI and CT had similar accuracy in predicting resectability of tumors in patients with esophageal cancer. However, because CT also provides information regarding distant metastatic disease and is more widely available, it has remained the technique of choice. Technical challenges have to be overcome when using MRI for esophageal imaging—the deep location of the esophagus and the degree of movement related to cardiac motion, peristalsis, and respiration, combined with the relatively slow acquisition time of MRI, conspire to degrade image quality. Recent research has focused mainly on endoluminal MRI because reducing the distance between the coil and the esophagus increases the signal-to-noise ratio, improving image quality. However, limited in vivo and ex vivo studies are available on endoluminal MRI esophageal imaging. Analysis of the optimal sequence indicates that T2-weighted images provide the best delineation of the layers of the esophageal wall [9-12]. To our knowledge, the use of an external surface coil to acquire high-resolution images of the esophagus has not previously been described.

The aim of this pilot study was to establish whether MRI of esophageal cancer is feasible using an externally placed surface coil. We also evaluated the ability of the technique to depict normal esophageal wall and periesophageal planes and to clearly delineate tumor by correlation with histopathology of the resected specimen.

Ten patients with confirmed esophageal carcinoma (seven with adenocarcinoma, two with squamous cell carcinoma, and one with spindle-cell melanoma), who were deemed medically fit for surgery and shown to have resectable disease using endoscopic sonography and CT, underwent high-resolution MRI of their primary tumors on the day before surgery. The study group contained 7 men and 3 women with median age of 60 years (range, 47-82 years). Nine patients had preoperative combination chemotherapy, and one patient with spindle cell melanoma underwent primary surgery. Nine patients underwent esophagogastrectomy, and one underwent total gastrectomy and esophagogastrectomy with a colonic interposition. The resections were performed by two surgeons with 16 and 17 years experience in esophageal surgery.

Ethics approval was obtained from the local research ethics committee. Informed, written consent was obtained from each patient.

The study shows that surface-coil MRI of the esophagus is feasible and, when using a T2-weighted sequence, that the esophageal wall layers are accurately depicted and the tumor can be identified separately from surrounding tissue. The results regarding the optimum sequence concur with the in vitro studies performed by Yamada et al. [12, 14]. We chose not to use respiratory or cardiac gating in an attempt to minimize scanning time, which undoubtedly impaired the quality of images obtained. Further work is needed to assess whether this is a satisfactory compromise. We acknowledge that the use of oral and IV contrast media and multiplanar imaging also require investigation to maximize the potential of this technique.

Statistical analysis of the study is limited by the small sample size and the fact that individual patients contributed an average of five of the total 50 images analyzed in the study. Interpretation errors were therefore carried over between the images, exaggerating their impact overall. The appearance of the mucinous tumor, for example, was misinterpreted by all three interpreters, and measurements of tumor depth were underestimated on four MR images as, in each case, high signal intensity within the submucosa was considered to be edema rather than tumor. The measurements taken by the interpreters were at the interface with the unaffected esophageal wall rather than at the maximal extent of the tumor. Overestimation of disease occurred in tumors at the gastroesophageal junction because of the obliquity of the sections. This will undoubtedly remain a difficult area to image, as it is for endoscopic sonography. Further work using tangential MR images at the gastroesophageal junction may provide more accurate assessment of these tumors.

Tumor filling in the esophageal lumen resulted in inaccuracies in tumor measurement because it was difficult to accurately measure the single wall thickness. In reality, it is the mural and, in particular, the extramural component of the tumor that is important for staging and assessment of resectability, and these distances are likely to be more accurately measured.

The discrepancies between interpreters 1 and 2 were essentially for distances greater than 5 mm. At short distances, more important for tumor staging, the correlation between the interpreters was good. Interpreter 3 had 26 years of cross-sectional imaging experience, but had limited previous experience in gastrointestinal MRI, which might explain the wider coefficient of repeatability for interpreter 3 and illustrates that a learning curve is associated with image interpretation.

We acknowledge a limitation within the study design involving interpreter 2, who potentially could have been unblinded by the process of matching the MR images with histology. All attempts were made to prevent this, as explained in the methodology. The results indicate that unblinding is unlikely to have occurred because results from this interpreter were similar to the other two.

Specific correlation with pathologic T staging is required to reinforce the potential of this imaging technique. However, this study has shown that external surface-coil MRI of the esophagus is technically feasible and enables detailed imaging of the esophageal wall, delineation of tumor, and depiction of the surrounding periesophageal tissues.

The authors thank Cheryl Richardson and Erica Scurr for their expertise in the development of the MRI technique. We also thank Andrew Norman for his advice regarding the statistical analysis of the study data, Ian Chau for his contribution to the patient treatment protocol that incorporated the MRI study and for his recruitment of patients, and finally Sally Legge for her assistance in the recruitment of patients to the study.

Address correspondence to A. M. Riddell ([email protected]).