Predicting lymph node metastasis in pancreatobiliary cancer with magnetic resonance imaging: A prospective analysis

Abstract

ObjectivesTo prospectively investigate the diagnostic potential of lymph node (LN) magnetic resonance (MR) imaging features. MethodsA radiologist determined the maximum diameters in the short and long axes, shape, signal intensities on T1- and T2-weighted imaging, pattern of enhancement, and apparent diffusion coefficient (ADC) on diffusion-weighted MR images of LNs and annotated measurable (≥5 mm in short-axis diameter) LNs. Surgically harvested LNs were correlated with the pathologic findings. Univariable and multivariable generalized estimating equation analyses were performed to evaluate predictive power. ResultsOf 80 LNs, 29 (36.3%) were positive and 51 (63.7%) negative for metastasis. The mean short-axis diameter of metastatic LNs (10.59 ± 4.30 mm) was larger than that of benign LNs (7.96 ± 2.10 mm). The ADC was significantly (P < 0.001) lower in metastatic than non-metastatic LNs. The area under the curve (AUC) of a univariable model using only the mean ADC was 0.845 (95% confidence interval [CI], 0.743–0.927), and the mean-ADC cutoff value for predicting LN metastasis was 0.901 × 10−3 mm2/s. The AUC of a multivariable model including round shape, heterogeneous enhancement, and the mean ADC was 0.917 (95% CI, 0.845–0.972), with a sensitivity, specificity, overall accuracy, and positive and negative predictive values of 89.7%, 82.4%, 85.0%, 74.3%, and 93.3%, respectively. ConclusionThe short-axis diameter and ADC were different between benign and metastatic LNs in pancreatobiliary cancer. However, round shape, heterogeneous enhancement, and a low ADC value (<0.901 × 10−3 mm2/s) may be the most reliable diagnostic features of multiple metastatic LNs.