MRI predictive value for adenoma detection in Cushing's disease and reader dependence: a multi-observer study.

Abstract

Objective: MRI detection of pituitary microadenomas in Cushing's disease is inconsistent, and its impact on surgical outcomes remains debated. Diagnostic performance may vary by reader expertise and interpretation strategy. The authors sought to assess interrater agreement and compare diagnostic accuracy across expert MRI interpretation conditions.

Methods: In this retrospective study at a tertiary referral center, preoperative pituitary MRI from 90 patients with biochemically confirmed Cushing's disease who underwent surgery (2012-2022) were independently interpreted by a trained pituitary neuroradiologist and two experienced pituitary neurosurgeons. Interrater agreement (Cohen's κ) and diagnostic performance were evaluated across 5 reading conditions: each reader individually and majority and unanimous voting among the 3 readers. The reference standard was adenoma identification at surgery. The authors calculated sensitivity/specificity-derived metrics, including balanced accuracy, area under the receiver operating characteristic curve (AUC-ROC), positive likelihood ratio (PLR), and negative likelihood ratio (NLR). Three-month remission rates were compared between MRI-positive and MRI-negative interpretations within each reading condition using odds ratios (ORs) with 95% confidence intervals.

Results: Interrater agreement was fair to moderate (κ = 0.34-0.44). Performance differed across reading conditions (global comparison, p < 0.001). The highest diagnostic performance was observed for one neurosurgeon and for unanimous voting (balanced accuracy 0.84 and 0.85; AUC-ROC 0.90 and 0.85, respectively). In these two conditions, an MRI-positive interpretation was associated with improved adenoma identification at surgery and greater 3-month remission compared with MRI-negative interpretations (PLR 4.38 and 15.35; NLR 0.16 and 0.32; OR for remission 12.38 and 10.07, respectively).

Conclusions: In a surgically selected cohort with confirmed Cushing's disease, MRI localization of corticotroph microadenomas is strongly reader dependent. Consensus interpretation improves performance, and negative MRI are associated with lower remission rates, supporting optimization of preoperative imaging.