Brain MRI–based prognostication after cardiac arrest: qualitative assessment outperforms variable voxel-wise ADC thresholds

Purpose: Quantitative apparent diffusion coefficient (ADC) analysis is increasingly studied as a prognostic tool to predict neurological outcomes after cardiac arrest. Notably, however, optimal thresholds for poor outcome prediction differ widely between studies, limiting consistent clinical application. The aim was to investigate the prognostic value of voxel-wise ADC thresholds for neurological outcome prediction in the entire cohort and specific subgroups after cardiac arrest, and to compare quantitative thresholds with qualitative MRI visual assessments. Methods: This cohort study examined brain MRI scans from 261 comatose patients who were resuscitated post-cardiac arrest and treated with targeted temperature management at a single tertiary care centre. Subgroup analyses considered arrest aetiology, MRI timing, and acquisition protocol. The primary outcome was defined as poor neurological outcome (a Cerebral Performance Category score of 3–5). Results: The percentage of brain voxels (PV) at 400 and 450 × 10^–6 mm²/s exhibited the strongest discriminative performance (AUC 0.86 [95 % CI, 0.82–0.90]). PV 450 × 10^–6 mm²/s values exceeding 3.1 % predicted poor outcomes with 68.6 % sensitivity and 96.5 % specificity; a threshold above 11.8 % achieved 40.6 % sensitivity and 100.0 % specificity. Qualitative visual MRI assessment achieved the highest AUC (0.91 [95 % CI, 0.88–0.93]), yielding perfect specificity (100 %) and superior sensitivity (81.1 %). This approach also demonstrated the highest sensitivity and 100 % specificity when used in combination with other modalities. Further analysis identified substantial variation in ADC values across subgroups. The highest AUC for cardiac aetiology was noted at PV 400 (0.85 [95 % CI 0.79–0.91]), whereas in non-cardiac aetiology, PV 500 and adjacent thresholds (PV 450–550) demonstrated similarly high peak discriminatory performance (0.89 [95 % CI, 0.83–0.95]). Modification of the scanning protocol (transition to diffusion-tensor imaging-based diffusion scheme with 12 directions) shifted ADC distributions upward without altering other brain injury markers. Conclusions: Qualitative visual assessment remained a robust predictor, both independently and as part of multimodal prognostication, whereas quantitative ADC thresholds showed substantial variability across aetiologies and MRI protocols, underscoring the limitations of a universal threshold.