MRI Brain Age After Traumatic Brain Injury: Systematic Review Finds Mixed, Hard-to-Compare Evidence

Adults who sustain a traumatic brain injury sometimes carry brains that look older on MRI than their birth certificates suggest, but a new systematic review concludes that the evidence behind that striking idea is far messier than the growing number of published papers implies. The review, published on 1 October 2026 in Discover Neuroscience, analyzed 15 reports drawn from 13 provisional study families and found that some studies showed older-appearing brains after injury, others reported null or weak findings, and almost none could be compared directly with one another. The authors deliberately avoided a meta-analysis, arguing that pooling the available effect sizes would have produced a precise but misleading number.

The review team, led by Anoosh Maazi and Saro Bavani, searched PubMed, Scopus, and Web of Science from inception to 16 July 2026, supplemented by extensive hand-searching completed the following day. Of 338 database records, 162 duplicates were removed and 20 full texts were assessed, yielding 15 included reports. Hand-searching of reference lists and citation records turned up no additional eligible studies. The review was prospectively registered in PROSPERO (CRD420261451858) and reported according to PRISMA 2020.

Brain-age models use machine-learning or statistical algorithms trained on neuroimaging data from healthy reference samples to estimate a person’s chronological age from their brain scans. The difference between predicted and chronological age is called the brain-age gap, and a positive gap is conventionally read as an older-appearing brain. The measure has obvious appeal in traumatic brain injury research, because injury is increasingly viewed as a chronic and evolving condition rather than a time-limited insult, and because mechanisms such as diffuse axonal injury, neuroinflammation, vascular dysfunction, and protein aggregation could plausibly leave a structural signature that accumulates over years.

Yet the review found that the published literature does not speak with one voice. Four T1-weighted case-control reports provided raw group summaries: Cole observed a higher mean gray-matter predicted-age difference in adults with chronic, mostly moderate-to-severe civilian injury; Savjani found a higher cortical-thickness-derived measure in predominantly mild chronic military TBI; Zhang reported a higher multifeature brain-age gap after acute mild TBI; while Coetzee found only a small uncorrected difference between veterans with and without TBI in a study with a small control group. Adjusted estimates were equally mixed. Spitz reported a difference of 4.51 years (95% confidence interval 2.32 to 6.70), whereas Posis reported an adjusted estimate of minus 0.10 years after direction harmonization (95% confidence interval minus 1.75 to 1.56). The ENIGMA mega-analysis reported an adjusted difference of 4.99 years with p less than .001, but an exact standard error or confidence interval could not be recovered from the report.

Stratifying the evidence by age-bias correction, injury severity, and post-injury epoch did not resolve the inconsistency. Positive findings occurred both in studies that applied explicit post-prediction age-bias correction and in those that did not, but null or weak findings and analytically non-comparable outcomes also appeared in both categories. Because the correction method was confounded with population, imaging modality, model type, and reporting format, the reviewers concluded that the evidence does not support attributing between-study differences to age-bias handling alone. Military and veteran cohorts with predominantly mild or retrospectively ascertained injury were particularly inconsistent, while chronic moderate-to-severe civilian and multicohort samples more often described older-appearing brains, though with potential cohort overlap limiting independent replication.

The longitudinal evidence was even thinner. Seven reports included repeated imaging or a longitudinal component, yet none provided a set of at least two independent, directly compatible brain-age change estimates with valid within-person variance. Gan reported a small, nonsignificant mean increase from baseline to 6 to 12 months without a direct time coefficient. Amgalan reported similar mean values at approximately 7 days and 6 months but analyzed repeated observations with an independent-samples test. Dennis included 201 repeat scans without a complete longitudinal coefficient and standard error, Mayer reported a nonsignificant group-by-visit interaction without the coefficient and covariance needed for synthesis, and Rodrigues reported only p values for change between two weeks and six months. The review concluded that the available longitudinal evidence could not establish whether the brain-age gap recovers, remains stable, or progressively diverges after injury, and that only a larger group-by-time increase relative to longitudinally scanned controls would support progressive divergence.

Cognitive associations showed promise but stopped short of confirmation. Cole reported that increased gray-matter brain age was generally associated with decreased processing speed and impaired memory and executive functioning, and Gan and Zhang both found inverse relationships between the brain-age gap and information-processing speed, though potentially overlapping samples prevented those findings from being treated as independent. Spitz found an adjusted relationship with a delayed verbal memory factor but not with other cognitive factors, while Dennis and colleagues reported no clear association with the tested cognitive measures, although PTSD, depressive symptoms, and alcohol misuse were associated with the age-deviation score. The ENIGMA mega-analysis found no conclusive cognitive evidence and produced mainly p values incompatible with the coefficients reported elsewhere. Because effect types, cognitive domains, time points, covariate sets, and sample independence all differed, the reviewers characterized the cognitive evidence as suggestive rather than confirmatory.

The review also sounded a cautionary note about what an elevated brain-age gap actually means biologically. A model trained on healthy anatomy can assign an older predicted age to focal lesions, edema, registration error, or other anomalies simply because those images depart from its training distribution, meaning the measure may index model sensitivity to abnormal anatomy rather than a genuine aging process. In Cole’s study, contusions were identified in 52 of 99 participants, but the predicted-age difference did not differ between participants with and without contusions, and the contrast persisted in a lesion-free moderate-to-severe subset, reducing but not eliminating the concern. The reviewers also emphasized that cross-sectional gaps conflate pre-injury vulnerability, acute injury effects, chronic consequences, and comorbidity, so the results should be read as associations rather than causal contrasts.

Four recurring obstacles emerged as the review’s central contribution: absent or inconsistent age-bias correction, clinical heterogeneity of injury phenotype and post-injury epoch, overlap between published cohorts, and incomplete reporting of outcome precision and longitudinal covariance. The authors identified four known or potentially overlapping report clusters, including the Gan/Zhang and Spitz/ENIGMA pairs, and treated reports within a cluster as related data rather than independent replication. Sample sizes in the included studies ranged from 50 athletes to 822 military participants, and most studies used T1-weighted MRI, although two applied diffusion MRI. The breadth of the literature, the authors warned, should not be mistaken for independent replication of a single estimand.

The review’s conclusion is measured: MRI-derived brain-age measures can identify structural deviation from normative aging in some adult TBI populations, but they do not yet define a common effect or demonstrate accelerated longitudinal aging. Until a harmonized minimum reporting standard, independent external validation, and longitudinal studies with appropriately scanned controls are in place, the authors argue that brain age should be treated as a family of research measures of deviation from normative anatomy rather than a validated stand-alone biomarker of post-injury neurodegeneration, prognosis, or clinical status.

Subject of Research: A systematic review of MRI-derived brain age differences in adults after traumatic brain injury.

Article Title: A systematic review finds variable MRI derived brain age differences after adult traumatic brain injury and limited evidence on longitudinal change and cognitive associations

Article References: Maazi, A., & Bavani, S. (2026). A systematic review finds variable MRI derived brain age differences after adult traumatic brain injury and limited evidence on longitudinal change and cognitive associations. Discover Neuroscience, 21(1), Article 110. https://doi.org/10.1186/s13064-026-00339-5

Image Credits: AI Generated

DOI: 10.1186/s13064-026-00339-5

Keywords: traumatic brain injury, brain age, brain-age gap, MRI, systematic review, neuroimaging, mild TBI, concussion, neurodegeneration, biomarker, longitudinal imaging, cognitive outcomes