MRI edema patterns in acute neck infections: A multicenter and multidisciplinary interobserver study

Abstract

Purpose To assess the diagnostic accuracy and interobserver agreement of radiologists and surgeons in detecting prognostic MRI edema patterns, retropharyngeal (RPE) and mediastinal (ME) edema, in acute neck infections, supporting their clinical integration into emergency imaging workflows. These patterns are known to predict disease severity, yet their reliable identification across specialties remains untested until now. Methods In this multicenter observational study, 28 readers (comprising radiology residents, board-certified radiologists, and head and neck surgeons from five academic hospitals) evaluated axial T2-weighted Dixon MRI images from 60 patients with clinically confirmed acute neck infections. Readers, blinded to clinical data such as patient symptoms or laboratory results, assessed images for RPE and ME presence and rated their confidence. Sensitivity, specificity, accuracy, and interobserver agreement (Fleiss’ kappa) were calculated against a consensus standard set by two fellowship-trained neuroradiologists. Intraobserver consistency was tested by repeating 10 randomly selected cases, unbeknownst to readers. Results Overall sensitivity, specificity, and accuracy were 0.92, 0.80, and 0.87 for RPE and 0.88, 0.86, and 0.85 for ME, with substantial interobserver agreement (κ = 0.61 for RPE, 0.52 for ME). Radiologists outperformed surgeons in RPE sensitivity ( p = 0.007), RPE accuracy ( p = 0.04), ME sensitivity ( p = 0.01), and ME accuracy ( p = 0.02), with intraobserver consistency reaching 90 %. Confidence was high, averaging 4.3 for RPE and 4.1 for ME, and strongly correlated with accuracy (p < 0.001), reinforcing diagnostic reliability. Conclusion Physicians reliably detect MRI edema patterns across expertise levels, validating their use for risk stratification in acute neck infections and supporting broader adoption in clinical practice where MRI is available.