Clinical and prognostic significance of emergency MRI findings in neck infections

We retrospectively reviewed patients who had undergone emergency MRI for suspected neck infection between April 1, 2013, and December 31, 2018, in a single academic tertiary care referral center. We obtained permission from the hospital district board, and written patient consent was not obtained due to the retrospective nature of the study. Institutional review board (IRB) review (approval or waiver) was not sought, because it is not required by the national legislature for retrospective studies of existing data. Neck MRI was performed on these patients due to clinical suspicion of severe or deep neck infection. The inclusion criteria were (1) emergency MRI evidence of infection: high signal of fat-suppressed T2-weighted Dixon images suggesting edema, or high signal of fat-suppressed post-Gadolinium (post-Gd) T1-weighted Dixon images suggesting abnormal tissue enhancement; (2) final clinical diagnosis of infection; and (3) diagnostic image quality as deemed by the radiologist reading the study. Origin of the infection was determined as the final clinical diagnosis from the medical records. MRI was performed prior to hospital admission at the emergency radiology department using a Philips Ingenia 3-T system with a dS HeadNeckSpine coil configuration (Philips Healthcare). Details of the MRI are provided in the Supplementary information. We routinely administered a Gd-based contrast agent (gadoterate meglumine, Dotarem®, Guerbet). Demographic, clinical, and laboratory variables were extracted from medical records: age (years), sex (male/female), body mass index (BMI, kg/m²), duration of symptoms prior to imaging (days), C-reactive protein (CRP, mg/L), white blood cell count (WBC, × 10⁹/L), body temperature (°C), surgery (yes/no), treatment in an ICU (yes/no), and LOS (days). The need for ICU treatment was determined only on clinical grounds, e.g., to secure adequate ventilation or vital functions. Laboratory values were taken on admission or, in the case of multiple assessments, the most recent prior to imaging.

The main location of the infection was tabulated, as per final clinical diagnosis. In addition, we pooled two major groups of infections separately: pharyngotonsillar (including tonsillitis, peritonsillar abscess, and parapharyngeal space involvement) and odontogenic infections (mostly oral cavity, masticator, and buccal spaces, with a clinical confirmation of odontogenic origin by an oral surgeon).

We defined an abscess as an abnormal T2-hyperintense collection with a low apparent diffusion coefficient (ADC) surrounded by abnormal tissue enhancement on post-Gd T1-weighted images and no enhancement in the center. In analyses including abscess measurements, we excluded patients in whom purulence or an abscess cavity could not be surgically demonstrated (false positives, N = 14). We measured the maximal diameter of the abscesses from post-Gd T1-weighted images. Substantial interobserver agreement for the detection of an abscess has been shown previously [8].

Retropharyngeal edema (RPE) was defined as a hyperintense signal of at least 2 mm in anteroposterior thickness between the prevertebral muscles posteriorly and the superior pharyngeal constrictor muscle anteriorly, in at least two consecutive axial fat-suppressed T2-weighted Dixon images. Within the RPE, we further defined enhancing phlegmon as hyperintense areas on post-Gd T1-weighted Dixon images, and free fluid as any non-enhancing hypointense areas on post-Gd T1-weighted Dixon images within the phlegmon tissue. Free fluid was deemed purulent if it had a low apparent diffusion coefficient (ADC) and non-purulent if it had a high ADC on visual inspection. At the level of maximal RPE, we measured the thickness (mm) and craniocaudal position relative to the cervical vertebrae. We also recorded whether the RPE was suprahyoid, infrahyoid, or both.

Mediastinal edema (ME) was defined as a hyperintense signal in the soft tissues in axial or coronal fat-suppressed T2-weighted Dixon images at or below the level of the thoracic inlet, using the superior border of the manubrium sterni (anterior ME), or the first thoracic vertebra (posterior ME) as superior borders. Anterior ME was often contiguous with edema in the visceral space, while posterior ME was often a continuation of infrahyoid RPE.

RPE, ME, and abscess diameter were defined by a board-certified radiologist, fellowship trained in both neuroradiology and head and neck radiology (J.Hi., 10 years of experience in diagnostic radiology). To ensure the reliability of these assessments, another board-certified radiologist (J.N., 8 years of experience in diagnostic radiology) independently evaluated images for three different findings (RPE, ME, abscess diameter). For each finding, images from a random sample of 60 patients were evaluated, and each of these random samples was evaluated on separate occasions.

The results are expressed as percentages, means, and standard deviation (SD). For univariate assessments, we used Pearson T tests for continuous data and the chi square (Χ²) test and odds ratios (OR) for ordinal data. We evaluated inter-rater reliability using Cohens’ kappa for ordinal data and an intraclass correlation coefficient (ICC) for continuous data. In univariate analyses, we included age, sex, BMI, duration of symptoms, CRP, and WBC as clinical and laboratory parameters of interest. Body temperature was not included because of a large amount of missing data. For multivariate prediction models, we used binary logistic regression for dichotomous outcomes and linear regression for continuous outcomes. In these models, we only included variables that had been significant at the univariate level. We constructed three different models for both ICU treatment and LOS, in order to compare the added value of MRI findings. Model 1 had only significant clinical or laboratory predictors (no MRI findings). Model 2 added MRI edema patterns, and model 3 added abscesses. To make the models comparable, all had similar sample sizes. Model 1 only included patients in whom MRI edema patterns were available for analysis. Furthermore, in model 3, patients with no MRI evidence of abscess or false-positive abscesses in surgery were assigned “0” for maximal abscess diameter. In the regression models, all the predictors were entered into the model (fixed predictors). R² values were compared between models. Because of the retrospective nature of the study, the sample size varies in different statistical analyses depending on which data is available. p values less than 0.05 were considered statistically significant. We analyzed the data using IBM SPSS Statistics for Mac (version 26, copyright IBM Corporation 2019).

Our study included 371 patients (226 male, 145 female) with clinically confirmed acute neck infection (Table 1). The patients had a mean age of 41 (range 0–88 years). Pediatric patients (<18 years old) represented 12% (N = 45) of the total sample. The mean duration of symptoms before imaging was about 5 days. As many as 254 (69%) patients underwent surgery, and 50 (14%) patients were treated at the ICU. Mean LOS was 4 days (range 0–55). The two most common types of infection were pharyngotonsillar (36%) and odontogenic (29%), followed by other less common locations (35%) (Fig. 1).

RPE was found in 201 (54%) patients (Table 2). One hundred forty-nine patients (74%) had retropharyngeal enhancement and 52 (26%) patients had free fluid collections (Fig. 2.). The mean thickness of RPE was 4.7 ± 2.3 (SD) mm. The thickest portion of RPE was most commonly at the C3 level (50%). Relative to the hyoid bone, 115 (57%) patients had RPE on both sides, superior and inferior to the hyoid bone, whereas 84 (42%) patients had suprahyoid only and 2 (1.0%) patients infrahyoid only. RPE was seen in various types of neck infections (Fig. 3).

The images covered the upper mediastinum sufficiently to evaluate ME in 311 (84%) patients, among 81 of whom ME was present (26%): anterior in 59%, posterior in 48%, and both in 7.4% (Table 2, Fig. 4). ME was equally prevalent among pharyngotonsillar (15%) and odontogenic infections (23%), although it occurred mostly posteriorly in the former (88%) and anteriorly (95%) in the latter.

Two hundred seventy-four patients had MRI evidence of an abscess, of whom 229 underwent surgery. Excluding 14 false-positive abscesses, data from 260 abscesses were analyzed. The mean maximal abscess diameter was 35 mm. No differences were found between the prevalence (85% vs. 76%) or maximal diameters (33 vs. 35 mm) of the abscesses in pharyngotonsillar and odontogenic infections (respectively).

RPE was significantly more prevalent in the patients with an abscess (62%) than in those without (34%) (p < 0.001), whereas ME was equally prevalent (27% vs. 22%, p = 0.332).

In the independent inter-rater evaluation of 60 randomly selected patients, imaging outcomes were reliable, as indicated by kappa values consistent with substantial agreement (0.781 and 0.675 for RPE and ME, respectively), and ICC values consistent with good reliability (0.81 and 0.75 for RPE thickness and maximal abscess diameter, respectively).

Overall, the patients with RPE and ME had higher incidence of markers consistent with more severe disease, including higher incidence of ICU treatment, larger abscesses, higher CRP, and longer LOS, than those without (Fig. 5).