Upper airway morphometry in acute neck infections: Clinical associations and deep learning-based segmentation

Upper airway (UA) compromise is a significant and potentially life-threatening complication of acute neck infection (ANI), but objective, quantitative imaging features are lacking. This study aimed to evaluate the association between UA morphometry in emergency neck MRI and disease severity in ANI, and to develop and evaluate a deep learning (DL)-based model for automatic UA segmentation in ANI patients.

A total of 322 patients were included. UA lumen was manually segmented, and prespecified morphometric features were extracted. UA morphometry was correlated with clinical outcomes. Interobserver agreement for all extracted features was evaluated with the Intraclass Correlation Coefficient (ICC). An nnU-Net model was trained on manually segmented data, and its performance was evaluated using the Dice Similarity Coefficient (DSC) and feature-specific ICCs.

ICU admission occurred in 33/322 patients (10.2 %) and prolonged stay (>5 days) in 47/319 (14.7 %). Minimum cross-sectional area, compactness, maximum midline shift, and length of airway narrowing were associated with ICU admission (false-discovery-rate q ≤ 0.008) but not with prolonged stay, and did not significantly improve an established clinical model (AUC 0.80 to 0.84, DeLong p = 0.07 to 0.12). Interobserver agreement was moderate-to-excellent. The nnU-Net reached a median DSC of 0.920 with ICCs of 0.853 to 0.973 and reproduced the ICU associations.

Quantitative UA morphometry in ANI is reproducible. A narrow caliber, a long narrowed segment, and lateral displacement of the airway are associated with ICU admission, without demonstrated incremental value over established predictors in the prespecified comparison. nnU-Net segmentation reproduces the features and their univariable associations.