• Brain age is an indicator of brain health • Presence of stroke lesions results in underestimation of brain age • Healing images by filling the lesion results in improved brain age estimates • Different methods of healing have different strengths and limitations Post-stroke cognitive deficits are linked not only to the focal brain lesion but also to the condition of the surrounding brain structures, which have complementary roles in influencing the deficits’ recovery. Stroke survivors commonly have cardiovascular risk factors, which are determinants of cumulative brain tissue damage and premature brain aging processes. Brain age is a machine learning-based neuroimaging estimate of premature structural changes associated with aging. Nonetheless, existing brain age methodological approaches may be distorted by the presence of lesions and new strategies are important to guide the use and interpretation of brain age estimation in stroke survivors. To address this challenge, we developed a pipeline that allows the assessment of brain age in individuals with stroke while avoiding the distortion related to the stroke lesion. We compare brain age estimates via brainageR and their association with post-stroke language deficits (aphasia) when calculated without accounting for lesion, versus enantiomorphic lesion correction or AI-based synthetic SynthSR lesion correction. Enantiomorphic healing yielded brain age estimates that were, on average, 4.5 years older than unhealed estimates and more strongly correlated with chronological age (r = 0.72 vs. r = 0.55 for unhealed; p < 0.001). SynthSR-based healing produced estimates that were 3.1 years younger than unhealed, with a chronological age correlation (r = 0.54), statistically similar to the unhealed approach. After lesion correction, the spurious association between lesion volume and brain age was eliminated (p = 0.65), whereas it remained significant in unhealed images (p = 0.002). Both lesion correction methods produced brain age estimates that added explanatory variance to aphasia severity (WAB-AQ) beyond lesion quantification, while unhealed brain age did not. Lesion correction produces brain age estimates that are free of lesion artifact, well-calibrated to chronological age, and can augment the explanatory variance for aphasia severity. Each lesion adjustment is unique, and AI-based methods (SynthSR) can inadvertently normalize ventricular morphology and underestimate aging. This pipeline provides a tested, reproducible framework for integrating brain age into stroke research and cognitive neuroscience, as well as for interpreting results.
Jeakle et al. (Fri,) studied this question.