MRI-based prognostic factors and risk stratification for treatment response in post-traumatic olfactory dysfunction

Objective To identify baseline magnetic resonance imaging (MRI) features associated with treatment response in post-traumatic olfactory dysfunction (PTOD) and to develop an exploratory MRI-based risk score. Methods This single-center retrospective cohort study included 41 patients with PTOD and 41 age- and sex-matched healthy controls. Olfactory bulb (OB) volume, morphology, and integrity, olfactory sulcus depth, and structural lesions in olfactory-related cortical regions were assessed on MRI. Olfactory function was evaluated using the Sniffin’ Sticks test, and treatment response was defined as an increase of ≥5.5 points in the Threshold-Discrimination-Identification score. Least absolute shrinkage and selection operator (LASSO) regression was used to select MRI features for model development, from which a simplified integer-based scoring system was derived. Results Compared with healthy controls, patients with PTOD had a shallower left olfactory sulcus and lower proportions of convex OB morphology bilaterally. After a median follow-up of 8 months, 16 of 41 patients (39.0%) met the treatment-response criterion. Three MRI features were retained in the LASSO model: left OB integrity, left orbital gyrus injury, and right temporal lobe injury. The resulting model had an apparent area under the receiver operating characteristic curve (AUC) of 0.885 (95% confidence interval [CI], 0.787–0.983). Full-pipeline internal validation yielded a stratified ten-fold cross-validated AUC of 0.733 (95% CI, 0.564–0.901) and a bootstrap optimism-corrected AUC of 0.725. The simplified three-factor scoring system had an apparent AUC of 0.869 (95% CI, 0.761–0.977) and a bootstrap bias-corrected AUC of 0.867, with treatment response rates of 75.0%, 50.0%, and 5.3% in the low-, intermediate-, and high-risk groups, respectively. The empirical MRI scoring scheme had an apparent AUC of 0.805 (95% CI, 0.662–0.926); the difference between the two scoring systems was not statistically significant (DeLong p = 0.079). Conclusion Baseline structural MRI features were associated with treatment response in PTOD. The exploratory three-factor MRI-based score showed potential for risk stratification within this cohort, but prospective external validation is required before clinical application.