Summary
A bioRxiv preprint of 47 patients links phase-specific brain activity and connectivity patterns to early and later language recovery after ischemic stroke.
A longitudinal study of 47 people with aphasia after ischemic stroke found that different patterns of brain activity and connectivity predicted language recovery at different stages. The findings, reported in a bioRxiv preprint posted on 20 September 2026, point to phase-specific brain signals rather than a single marker of recovery.
Aphasia is a language disorder that can affect speaking, understanding, reading or writing after brain injury. Recovery depends not only on the damaged tissue but also on how remaining language-related and broader cognitive networks reorganise. The researchers examined activity in the left-hemisphere language network and in bilateral multiple-demand networks, which support demanding tasks such as attention, control and problem-solving.
Brain predictors shifted across recovery phases
The participants were assessed longitudinally during the acute, subacute and chronic phases after stroke. During functional neuroimaging, they completed a sentence-comprehension task. The analysis examined both task-related activity and functional connectivity. In this context, connectivity refers to coordinated activity between regions; it is a statistical measure of interaction, not a direct measurement of physical nerve fibres.
The prediction pattern depended on when recovery was measured. Activity and connectivity recorded in the acute and subacute phases independently predicted early recovery. More specifically, subacute connectivity running from multiple-demand regions to language regions was associated with early improvement.
For longer-term outcomes, activity in the acute multiple-demand network was the relevant predictor identified by the analysis. Later recovery spanning the subacute to chronic phases was instead predicted by activity in bilateral inferior frontal regions. The results therefore distinguish between network interactions that help predict earlier improvement and regional activity associated with later gains.
In the authors' model, “independently” means that these neural measures added predictive information after accounting for lesion-related information, age and aphasia severity. It does not describe a separate treatment effect.
A multivariate test of recovery signals
The researchers used confound-controlled multivariate regularised regression with cross-validation. A multivariate model evaluates several features together, while regularisation limits overly complex relationships that could fit the study sample too closely. Cross-validation tests whether those relationships retain predictive value across different subsets of the data. Feature ablation was used to assess how much specific predictor groups contributed to the results.
This approach allowed the team to compare broad predictor families and identify region-specific signals. The study was designed around prediction of recovery over time, not an intervention. Its findings therefore nominate multiple-demand and inferior frontal regions as candidate targets for future phase-adapted neurostimulation research. Neurostimulation would need to be evaluated separately to determine whether changing activity in those regions improves language outcomes.
The work supports a more time-sensitive view of post-stroke recovery: the brain features associated with early language improvement may not be the same as those associated with later recovery. The report is a bioRxiv preprint based on a cohort of 47 patients, and clinical use will require validation of the predictors in new patient groups and settings.