Background
We investigated whether a short, 5-min magnetic resonance imaging (MRI) protocol consisting of only axial T2-weighted and diffusion-weighted imaging (DWI) sequences can discriminate between tonsillar infections, peritonsillar abscesses and deeply extending abscesses in a retrospective, blinded, multireader setting.

Methods
We included patients sent by emergency physicians with suspected pharyngotonsillar infections who underwent emergency neck 3-T MRI from April 1 2013 to December 31 2018. Three radiologists (with 10−16 years of experience) reviewed the images for abscesses and their extension into deep neck spaces. Data were reviewed first using only axial T2-weighted Dixon images and DWI (short protocol) and second including other sequences and contrast-enhanced T1-weighted Dixon images (full protocol). Diagnostic accuracy, interobserver agreement, and reader confidence were measured. Surgical findings and clinical course served as standard of reference.

Results
The final sample consisted of 52 patients: 13 acute tonsillitis with no abscesses, 19 peritonsillar abscesses, and 20 deeply extending abscesses. Using the short protocol, diagnostic accuracy for abscesses across all readers was good-to-excellent: sensitivity 0.93 (95% confidence interval 0.87−0.97), specificity 0.85 (0.70−0.93), accuracy 0.91 (0.85−0.95). Using the full protocol, respective values were 0.98 (0.93−1.00), 0.85 (0.70−0.93), and 0.95 (0.90−0.97), not significantly different compared with the short protocol. Similar trends were seen with detecting deep extension. Interobserver agreement was similar between protocols. However, readers had higher confidence in diagnosing abscesses using the full protocol.

Conclusions
Short MRI protocol showed good-to-excellent accuracy for tonsillar abscesses. Contrast-enhanced images improved reader confidence but did not affect diagnostic accuracy or interobserver agreement.