Automated deep-learning segmentation of neck abscesses on MRI shows excellent volumetric agreement with manual segmentation.

An automated deep-learning model segmented neck abscesses on multiparametric MRI with a mean Dice similarity coefficient of 0.828 in a study of 226 patients. A simpler model using only post-contrast T1-weighted images was non-inferior, offering a streamlined alternative.

• The study used 226 patients with surgically confirmed neck abscesses from a single-center emergency MRI database and employed five-fold cross-validation for evaluation.
• Volumetric agreement with manual segmentation was excellent, with an intraclass correlation coefficient of 0.970 for the best model and 0.933 for the simpler, single-sequence model.
• The authors note that using only post-contrast T1-weighted imaging avoids the need for image registration, eliminating a potential confound.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.