Abstract / Summary
Importance: Brain MRIs are routinely acquired following moderate or severe traumatic brain injury (TBI) because of their high sensitivity for detecting traumatic brain lesions. However, there is no standard framework for their interpretation, and their clinical utility is unproven. Objective: Here, we test whether MRI-based quantification of traumatic brain lesions improves outcome prediction relative to the best tools currently available to clinicians. Design, Setting and Participants: We analyzed hemorrhagic contusions and axonal injury in four prospective studies of moderate or severe TBI, validating results in a fifth held-out dataset. Exposure: Contusion volume was measured via manual segmentation. Axonal injury volume was quantified with diffusion tensor imaging, by measuring the volume of skeletonized fractional anisotropy significantly lower than that of healthy controls. Main Outcomes and Measures: Functional outcomes were assessed at 6 months with the Glasgow Outcome Scale-Extended; values [≤] 4 were considered unfavorable. Results: Among 201 patients with MRI (mean age: 41, median Glasgow Coma Scale (GCS) score on admission: 7, MRI obtained median 14 days post-injury), MRI-measured contusion volume discriminated outcomes better than CT Marshall score (AUC 0.81 vs 0.64; p=0.001), and remained associated with unfavorable outcomes controlling for multiple commonly-measured clinical variables. Quantitative axonal injury volume discriminated outcomes better than manual MRI grading (AUC 0.82 vs 0.62; p<0.001) and remained associated with unfavorable outcomes independent of other clinical predictors. The prognostic discrimination of traumatic lesion volumes was not improved by accounting for the location or normative connectivity of either lesion type. In a multivariate model, contusion and axonal injury volumes were independently predictive, with age-dependent volume thresholds for 90% probability of unfavorable outcome, and strong discrimination in an external cohort (N=27; AUC 0.83, p<0.001). Replacing all CT-based measures with MRI lesion volumes in the IMPACT model significantly improved out-of-sample prognostic performance (AUC 0.85 vs 0.64; p<0.05). Conclusions and Relevance: MRI-based lesion quantification improves prognostic discrimination relative to measures currently available to clinicians treating patients with TBI. Global functioning at 6-months is driven more by the volume than the location or connectivity of traumatic lesions. These findings support integrating MRI-based lesion quantification into clinical workflows and updated TBI classification frameworks.