Right hemisphere atrophy rate associates with naming recovery after left hemisphere ischaemic stroke Authors: Voss Neal, Andreia V Faria, Wen Zhang, Argye E Hillis, Melissa D Stockbridge Published: 2026 ABSTRACT AbstractNumerous large-scale epidemiological studies investigating the trajectory of brain function after ischaemic stroke have presented data suggesting, on average, a persistent, accelerated decline in cognitive domains, including language, after accounting for the immediate impact of the stroke. We sought to further examine this trend, speculating that the average persistent decline may be a reflection of two subgroups with vastly different prognoses: (i) a minority experiencing decline secondary to neurodegenerative processes such as vascular dementia and Alzheimer’s disease and (ii) a majority without marked progressive brain atrophy who typically see improvement. Our team thus investigated atrophy’s association with language recovery, hypothesizing that declining naming performance in the year after left hemisphere ischaemic stroke would be correlated to atrophy of the contralesional hemisphere. We postulated that volume loss within the lesioned hemisphere would be less informative due to separate confounding processes related to the stroke itself, such as Wallerian degeneration and encephalomalacia.Participants [n = 72; M (SD) age = 60 (11)] in a longitudinal cohort study of language following left hemisphere ischaemic stroke were included if they completed an MRI both acutely and chronically (either 6 or 12 months post-stroke). Naming performance was assessed using the Boston Naming Test; stroke volumes were extracted from acute imaging; and atrophy was measured as the monthly per cent change in hemispheric volume from baseline to chronic scan for each individual. Pearson’s correlations were calculated to determine the relationship between lesion volume and atrophy along with atrophy and change in Boston Naming Test score.Lesion volume negatively correlated with the monthly per cent change in volume of the left (ipsilesional) hemisphere (r = −0.48; P < 0.0001) but did not correlate with rate of right (contralesional) hemisphere volume loss. While there was no clear relationship between atrophy of the left hemisphere and language recovery, we found that volume changes of greater negative magnitude within the right hemisphere (increased atrophy) were associated with worse functional recovery of language (r = 0.38; P = 0.0025).By showing that atrophy of the right hemisphere was not significantly impacted by left hemisphere lesion size, we suggest that accelerated volume loss in the non-lesioned hemisphere after stroke may be indicative of a separate pathology. We then support this claim with behavioural data showing that greater rates of volume loss within the non-lesioned hemisphere were associated with poorer naming recovery. Together, these findings imply that contralesional atrophy after stroke may have negative implications for recovery and could serve as a useful imaging signature for separate neurodegenerative processes. Introduction Aphasia1 is a debilitating potential consequence of stroke, posing substantial challenges in patients’ return to work and prior avocations. It is estimated that almost half of left hemisphere stroke survivors deal with some degree of language impairment, ranging from mild anomia to global aphasia.2 While improvement is possible and a reality for many, stroke recovery is a multifaceted and individual process that is not fully understood. The scale of the issue at hand has led numerous institutions to conduct large epidemiological studies aimed at defining the trajectory of recovery/decline over time following stroke.3-6 These studies employ various outcome measures to assess a wide range of domains (e.g. global cognition, verbal fluency, naming performance, memory, etc.) but paint a similar picture. Specifically, there is a recurring trend in which the average trajectory of recovery for stroke survivors is characterized by persistent, accelerated decline relative to those who remain stroke-free after accounting for the initial, acute impact of the stroke. Of particular relevance to our present linguistic focus, three of the epidemiological studies mentioned earlier assessed verbal fluency as one of their various outcome measures, a metric most critically influenced by language processing.7 In all three cohorts, stroke survivors exhibited accelerated decline in performance relative to participants without incident stroke.3-5 To add to the conclusions drawn from individual population-based studies, a meta-analysis by Lo et al.8 reported a similar trend when examining language as a whole. By aggregating results from multiple studies with a particular focus on fluency and naming [via the Boston Naming Test (BNT)], they showed that the language performance of stroke survivors declined at a faster rate than controls from 1 to 3 years after onset. This consistently reported average decline in language performance, however, may not reflect changes in the majority of individuals who have experienced a stroke and instead could be a result of conflating two distinct populations. It is likely that a minority of stroke survivors show marked decline in language while the larger majority sees improvement or complete recovery.9 Furthermore, the decline experienced by some stroke survivors may be secondary to repeated infarct as well as the development of vascular dementia, Alzheimer’s disease, and/or other neurodegenerative disease, as supported by data from both animal and human studies.10-12 By one estimate, around one in five patients meet criteria for a diagnosis of vascular dementia within 6 months after stroke.13 While the rate of post-stroke incident dementia is significantly greater in cases of repeat stroke,14 the threat is still present in the absence of stroke recurrence. A meta-analysis of 73 studies on the emergence of dementia up to 1 year after stroke reported mean prevalence estimates between 7.4 and 12.0% when examining cases of first-ever stroke specifically.15 As vascular dementia and Alzheimer’s disease compromise language and naming performance as part of their typical progression,16 both have the potential to negatively impact recovery of naming. Previous efforts from our lab have focused on the trajectory of recovery post-stroke as it pertains to language, and the resultant data support our scepticism regarding the generalizability of a central tendency of linguistic decline being ascribed to stroke survivors. In a longitudinal study of the influence of lesion characteristics on changes in picture naming and discourse in the first 6 months after stroke, we found that 71% of people with aphasia showed improvement in picture naming on the BNT and 21% showed decline, with the rest showing no change.9 In another study of trajectory of language deficits after stroke, 51% of participants improved over time, 26% declined, and 23% showed stable performance.17 Despite a great deal of relevant data from extensive epidemiological studies, it is still largely unclear as to why some stroke survivors (15–25%) show persistent decline in cognition and/or language performance, rather than recovery. Given the connection between cerebral volume loss and vascular cognitive impairment,18 some scholars posit that atrophy may be a driver for post-stroke cognitive decline and that associated imaging signatures could be clinically informative.19-21 Data from a multicentre, observational cohort in Australia showed that change in total brain volume was greater in stroke survivors who were cognitively impaired than those without deficits19 and that brain volume loss was greater ipsilesionally than contralesionally when examining both groups as a whole.22 Interestingly, further analysis showed that cortical thinning was greater in the contralesional cortex of cognitively impaired stroke survivors than in the ipsilesional cortex of cognitively normal stroke survivors.23 Other work has shown that atrophy of grey matter in the contralesional cerebellum significantly predicted variability in aphasia severity following left hemisphere ischaemic stroke,24 although this observation may reflect crossed cerebellar diaschisis related to left hemisphere stroke. Taken together, these results highlight the prognostic potential of tracking post-stroke atrophy and suggest that the contralesional hemisphere should not be overlooked. Here, we further investigate the impact of atrophy on language recovery or decline following ischaemic stroke. Our aim is to elucidate neural and functional differences present among stroke survivors that could explain why some experience decline and others experience improvement. Thus, we hypothesized that decline in language, specifically naming performance, in the year after left hemisphere ischaemic stroke correlates with the degree of atrophy in the contralesional hemisphere. We focus on atrophy in the contralesional hemisphere because atrophy in the lesioned hemisphere is primarily driven by Wallerian degeneration and encephalomalacia due to the stroke itself. Confirmation of our hypothesis would support the notion that persistent naming decline after stroke, when it occurs, is due to development or unmasking of neurodegenerative disease which may or may not be independent of the stroke. Materials and methods Participants We analysed data collected from 72 participants [34 female; M (SD) age = 60 (11)] as part of an ongoing longitudinal cohort study following patients’ recovery during the year after acute left hemisphere ischaemic stroke. This study includes a battery of multiple behavioural assessments primarily focusing on language in conjunction with a neuroimaging component. Participants were selected from this study for our analysis if they had completed an MRI acutely in addition to a follow-up MRI at either 6 or 12 months, utilizing the more chronic of the two if both follow-ups were completed. Data collection for the resultant sample occurred between 2008 and 2024. The records reviewed were an exhaustive set who met inclusion criteria and had MRI data at the required timepoints, leading to our study size of 72. We examined participants’ behavioural data along with the imaging analysis, focusing on the BNT (30-item edition) at the timepoints with completed MRIs as a proxy for naming and language function. Participants (or their legally authorized representatives in cases of incapacitation resulting from the stroke) were approached if they had experienced a left hemisphere ischaemic stroke outside of the brainstem/cerebellum, were over the age of 18, were proficient English speakers, and had no history of hearing/vision impairment. Study team members screened medical charts to identify and exclude patients with prior neurological disorders including, but not limited to, dementia and mild cognitive impairment. Medical charts were also used to compile data for vascular risk factors (diabetes and hypertension) and cognitive function at the time of hospitalization per the Short Blessed Test.25 We sought to eliminate collection bias within the population by approaching all patients meeting eligibility criteria. Those who consented to participate then completed acute testing within 5 days of symptom onset. In cases where the patient was intubated over the course of their hospitalization, the acute window was extended to include the 5 days post-extubation. Following discharge, members of the study team reached out to patients and/or family members via phone call to schedule follow-up visits at 3, 6, and 12 months after hospitalization, though data collected at 3-month follow-ups was not included in this analysis. To minimize loss to follow-up, multiple contact attempts were made per timepoint window along with offering transportation assistance within reasonable distance. Participants were encouraged to complete all clinically indicated rehabilitation including speech-language pathology to address deficits in communication and were not barred from participating in other interventional or longitudinal research that might influence their results. The study was conducted at Johns Hopkins Hospital and Johns Hopkins Bayview Medical Center in accordance with the Johns Hopkins Medicine Institutional Review Board (IRB Protocol NA00042097), and participant consent was obtained according to the Declaration of Helsinki. Naming assessment Participants completed a 30-item version of the BNT26,27 during the acute phase of their recovery and again at either 6 or 12 months post-symptom onset, using the more chronic scores for analysis. Administrators of the BNT presented an illustrated list of 30 common objects to name. This metric is particularly sensitive for detecting mild anomia, a common residual deficit in persons with aphasia.28 Of the 72 participants with acute and chronic imaging data, 62 had associated BNT scores at both timepoints. The 10 who did not have available data at either one or both timepoints were thus unable to yield a change in BNT score and were not included when analysing the correlation between atrophy and naming recovery but were included in our analysis regarding the correlation between lesion size and atrophy. Missing data for these participants were a result of the inability to complete acute testing prior to hospital discharge, omission of the assessment at follow-up, or the completion of testing during a brief window when the BNT was not included in the behavioural protocol. Healthy controls As part of a preliminary analysis intended to contextualize the atrophy seen in our population of stroke survivors, we also examined rates of volume loss associated with healthy aging for neurotypical controls (n = 11) who had an initial MRI followed by a subsequent scan 1 year later. All controls were without a diagnosis of any neurological disease and had Montreal Cognitive Assessment scores >26. Imaging analysis Acquisition and processing This is a retrospective analysis of prospectively collected data in an ongoing longitudinal observational trial. Our sample is made up of participants from this trial who had sufficient acute and follow-up MRI during the first year after their stroke. The baseline image is the scan completed at the time of admission for their first symptomatic stroke on a 3 Tesla scanner. All the patients had diffusion-weighted imaging (DWI), which was used to delineate the stroke core and obtain its volume. Acute Stroke Detection and Segmentation software (ADS; https://www.nitrc.org/projects/ads) was used for stroke core segmentation. The ADS algorithm29 was trained and tested on over 2000 DWIs from patients with acute ischaemic stroke and achieved performance comparable to, or slightly better than that of intra-rater agreement. All segmentations were quality controlled, and minor corrections were performed when needed by two evaluators: a neuroimaging researcher and a neuroradiologist with over 20 years of experience. A subset of 43 patients had high-resolution T1-weighted imaging (T1-WI), specifically magnetization-prepared rapid gradient echo (MPRAGE), with resolution of 1 × 1 × 1 mm at baseline; 29 had regular T1-WI with high in-plane resolution (1 × 1 mm) but large slice thickness varying from 4 to 6 mm. All the patients had T1-WI (MPRAGE) at follow-up, with 56 patients having scans available from their 1-year follow-up and 16 patients having scans available from their 6-month follow-up. The 16 patients without 12-month follow-up imaging data were either lost to follow-up after numerous contact attempts or declined an MRI at the final timepoint, as imaging was offered as a supplemental portion of the longitudinal protocol related to language recovery. We addressed the discrepancy that arose from grouping 6- and 12-month scans by calculating the rate of atrophy as a monthly percentage change, which is described in further detail later on. Follow-up scans were performed in the F.M. Kirby Research Center at Kennedy Krieger Institute and completed at the same time point as the BNT. The images were segmented using SynthSeg30 (from FreeSurfer version 7.4.1 with the options ‘-parc -robust’); in this study, we analysed the hemisphere volumes (Fig. 1). These volumes were calculated on synthetic MPRAGEs generated for each case with SynthSR31 to minimize the differences in the image protocols. In addition, we utilized the 4-point Fazekas visual assessment on Fluid Attenuation Inversion Recovery sequences for all participants to assess burden of white matter hyperintensities as a relevant risk factor.32 Quality control for all the steps was done by a radiologist. Image analysis for the controls was done at baseline and again after 6 and 12 months using T1-WI MPRAGE in an identical post-processing pipeline as used for the patients. Schematic representation of the imaging analysis. Lesion was segmented from acute diffusion-weighted imaging (DWI) scans using the Acute Stroke Detection and Segmentation (ADS)24 tool, as denoted by green overlay. Synthetic magnetization-prepared rapid gradient echo (MPRAGE) scans were created from T1-weighted images using SynthSR26 and subsequently segmented using SynthSeg,25 producing multi-region labels (coloured overlay). These regional labels are shown to illustrate segmentation accuracy and were not analysed individually. All analyses were performed at the hemispheric level, with volumes measured acutely and at chronic follow-up to quantify atrophy. Atrophy quantification Atrophy was measured by comparing the volumes of the hemispheres at baseline scans to follow-up scans for each individual, as (follow-up—baseline)/baseline. This allowed us to minimize variation between head volumes and normalize the range of values. Furthermore, we divided this ratio by the number of months between the time points. In summary, the resultant index is a percentage change from the initial volume, per month, with negative numbers representing atrophy (i.e. follow-up < baseline). In order to make the volumetric changes comparable to the naming changes, changes in naming scores were also divided by the time interval between them, representing the monthly rate as well. Statistical analysis Group differences in structural volumes (e.g. ventricles, hemispheres) were assessed using t-tests. Pearson’s correlation was used to assess relationships between atrophy, BNT scores, and stroke volumes. Statistical tests, completed using MATLAB 2024, were two-tailed and utilized a significance level of α = 0.05 to denote the probability of a type I error. Each contrast was tested sequentially in the a priori order that it is presented. Missing data were not imputed prior to analysis. Results We reviewed 732 records of consented participants in our longitudinal study of language recovery following left hemisphere ischaemic stroke. Of this group, 72 participants were included in this sample having met eligibility criteria with sufficient MRI at both the acute and chronic timepoints. Table 1 shows the profile of the 72 participants along with summary statistics regarding lesion volume, naming performance, atrophy rates for both hemispheres, cognition during hospitalization, and vascular risk factors. Controls used in our preliminary analysis are also included and had age and gender distribution not significantly different from the participants. Table 1Demographic and summary statisticsStroke survivors (n = 72)Controls (n = 11) M (SD)Range M (SD)RangeAge (years)60 (11)28 to 8556 (15)39 to 77Education (years)14 (3)9 to 20––Lesion volume (cc)18.89 (33.64)0.20 to 181.07––Naming performance (n = 62) Acute BNT score (/30)20 (9)0 to 30–– Chronic BNT score (/30)24 (7)0 to 30–– Change in BNT score4 (6)−5 to 26––Cognition (n = 38; unable to assess 12 participants due to severity) Acute SBT score6 (6.3)0 to 21––Atrophy Change in LH volume (overall %)−3.49% (2.65%)−11.01 to 0.74%−0.85% (0.50%)−1.71 to −0.11% Change in RH volume (overall %)−1.84% (1.85%)−7.21 to 3.16%−0.84% (0.66%)−1.77 to 0.08% Change in LH volume (monthly %)−0.35% (0.29%)−1.31 to 0.06%–– Change in RH volume (monthly %)−0.18% (0.21%)−0.81 to 0.28%––Vascular risk factors WMHs (via Fazekas scores)1.26 (1.6)0 to 6–– Hypertension63/72 (87.5%)– Diabetes26/72 (36.1%)–Sex (male:female)38:343:8Race (Black:White:other)44:27:10:11:0Ethnicity (non-Hispanic:Hispanic)71:1–Handedness (left:right)9:630:11BNT, Boston Naming Test—Shortened26,27; SBT, Short Blessed Test25; LH, left hemisphere; RH, right hemisphere; WMHs, white matter hyperintensities. Demographic and summary statistics BNT, Boston Naming Test—Shortened26,27; SBT, Short Blessed Test25; LH, left hemisphere; RH, right hemisphere; WMHs, white matter hyperintensities. Atrophy associated with healthy aging compared to post-stroke Both controls and patients showed reduction in the brain volume over time, as well as ventricle enlargement, as expected with aging. However, the atrophy was more severe in patients than in controls (P = 0.028, effect size by Cohen’s d = 0.545, power = 0.80; Supplementary Fig. 1). Lesion volume and subsequent hemispheric atrophy Volumetric reduction in the left hemisphere correlated to stroke volumes (r = −0.48; P < 0.0001; Fig. 2A), indicating that it might be the direct result of the stroke as well as secondary Wallerian degeneration. However, atrophy of the right hemisphere did not correlate with stroke volume (Fig. 2B), suggesting a minimal direct effect of the stroke on progressive atrophy in the non-lesioned hemisphere. Scatter plots of stroke volume versus hemispheric atrophy. The relationship between log(stroke volume (mL)) and hemispheric atrophy as a monthly percentage was assessed using Pearson’s correlation; n = 72. Each dot represents a single patient, with lines representing the fitted regression. Stroke volume was correlated to atrophy of the left hemisphere (A; r = −0.48; P < 0.0001), but not of the right hemisphere (B; r = 0.05; P = 0.67). Lesion volume and naming recovery When examining the relationship between stroke volume and BNT scores, an interesting phenomenon emerges. Patients with larger strokes appear to recover more (r = 0.41; P = 0.001; Fig. 3A). However, this might be explained by a ‘ceiling effect’—patients with high baseline scores (BNT ≥ 25) are performing at or near the maximum possible value of 30 correctly named items and thus have less room for improvement. It is difficult to determine the true effect of lesion size on naming recovery in this context, so we performed an exploratory analysis in which the high performing patients at baseline were excluded. With this adjustment, no clear relationship between stroke volume and recovery is observed (Fig. 3B). BNT scores and demographic profiles for the group that was excluded can be found in Supplementary Table 1. Scatter plots of stroke volume versus change in naming. The relationship between log(stroke volume (mL)) and monthly change in naming score on the Boston Naming Test was assessed using Pearson’s correlation. Each dot represents a single patient, with lines representing the fitted regression. Patients with large strokes counterintuitively appear to recover more (A; r = 0.41; P = 0.001; n = 62). When patients close to naming ceiling at baseline are removed—those with less room to recover—this relation disappears (B; r = 0.25; P = 0.14; n = 34). Hemispheric atrophy and naming recovery The relationship between language recovery and atrophy in the left hemisphere was not very informative, as its volume is heavily influenced by stroke volume, as previously noted. In fact, a similar counterintuitive pattern is observed: patients with greater left hemisphere atrophy (which we saw was correlated to larger strokes) appear to recover more (Fig. 4A). Again, this phenomenon could be explained by a ceiling effect and disappears when excluding patients with high naming scores at baseline. In contrast, in the right hemisphere—where atrophy is not influenced by stroke volume—there is a clearer relationship: patients with greater atrophy show less functional recovery (r = 0.38; P = 0.0025; Fig. 4B). This relation persists when only considering patients with lower naming scores (BNT < 25; r = 0.36; P = 0.039; Supplementary Fig. 2A) or larger strokes (log(stroke volume (mL)) > 0.9; r = 0.48; P = 0.0014; Supplementary Fig. 2B) at baseline. Scatter plots of hemispheric atrophy versus change in naming. The relationship between hemispheric atrophy as a monthly percentage and monthly change in naming score on the Boston Naming Test was assessed using Pearson’s correlation; n = 62. Each dot represents a single patient, with lines representing the fitted regression. Atrophy of the left hemisphere, highly influenced by the stroke volume, was not correlated to naming recovery (A; r = −0.21; P = 0.11), while atrophy of the right hemisphere was associated with worse recovery of naming (B; r = 0.38; P = 0.0025). To confirm the association between decline in naming and right hemisphere atrophy, we also replaced monthly change with absolute change in BNT score. As expected, results were very similar (Supplementary Fig. 3). Discussion Given the growing societal impact of language decline following stroke and the importance of better understanding the trajectory of recovery for stroke survivors, we set out to examine the role that atrophy plays in post-stroke naming impairment, one important element of cognitive dysfunction due to stroke. Specifically, our team used longitudinal behavioural and imaging data to characterize the effect of left hemisphere ischaemic stroke and lesion volume on rates of hemispheric atrophy in the year after stroke and then further explored the significance of atrophy in relation to naming recovery, as measured by performance on the BNT. As expected, participants with larger strokes saw greater rates of atrophy of the left, lesioned hemisphere (likely due to Wallerian degeneration and encephalomalacia), but we found no significant relationship between lesion volume and atrophy of the right, contralesional hemisphere. Thus, we surmised that accelerated rates of contralesional atrophy in stroke survivors may indicate a separate neurodegenerative process, such as vascular dementia or Alzheimer’s disease. To expand upon that observation, we investigated contralesional atrophy as a potential imaging signature for cognitive decline and sought to determine if volumetric reductions in the non-lesioned hemisphere manifested behaviourally in a way that was associated with naming recovery. Results aligned with our hypothesis, showing that participants with greater rates of atrophy in the right hemisphere exhibited worse recovery of naming. In contrast, patients showing minimal right hemisphere atrophy tended to see improvement, rather than continued decline, in naming over the first year after stroke. Involvement of the contralesional hemisphere in the recovery process after stroke has been documented, with prior research indicating that increased contralesional activity, as a compensatory mechanism, contributes to motor recovery following stroke33 and may be increasingly important for patients with larger lesions.34 Heiss35 drew similar conclusions with respect to aphasia, stating that although the greatest potential for recovery comes from reorganization of the ipsilesional hemisphere, compensatory utilization of the contralesional hemisphere can still lead to improvement (although to a lesser degree) in cases of larger lesions where the language network is sufficiently compromised. The higher utility of maintaining left hemisphere functionality as opposed to right hemisphere compensation is reflected with work by Keser et al.,36 who utilized diffusion tensor imaging acutely and chronically along with the BNT in left hemisphere stroke survivors. They concluded that greater recruitment of the right hemisphere’s language homologues actually were associated with poorer language outcomes, presumably as it reflects difficulty with utilization of the dominant hemisphere. We can consolidate and add further nuance to these findings via the bimodal balance-recovery model,37 a framework for stroke recovery in which contralesional influence and the resulting interhemispheric balance are contingent upon structural reserve, a parameter denoting the lesion’s impact upon critical neural pathways. Specifically, it suggests that support from the contralesional hemisphere is expected in cases of greater ipsilesional damage, and inhibition is expected with greater structural reserve. The absence of correlation between lesion volume and right hemisphere atrophy indicates that it is likely to be independent of structural injury and thus an independent predictor of recovery versus decline. From a conceptual standpoint, there are two main reasons right hemisphere atrophy is related to naming decline. First, hemispheric atrophy could be related to naming recovery because naming recovery may depend on intact homologous right hemisphere regions, at least in some people.38 Alternatively, right hemisphere atrophy may just be a biomarker that the individual has the beginnings of dementia (for example, vascular dementia or Alzheimer’s disease, which would likely impede recovery). Our finding that left hemisphere stroke volume was unrelated to right hemisphere atrophy indicates that accelerated contralesional atrophy may be a harbinger of separate neurodegenerative processes leading to the decline seen in naming performance. Although the development of contralesional hemisphere atrophy may be causally related to stroke, when it is due to vascular dementia or Alzheimer’s dementia, it is not an inevitable consequence. It is important to note that the atrophy highlighted in this study can have various origins such as prior strokes and small vessel disease. See Table 1 for data on hypertension, diabetes, and white matter hyperintensity burden in our patients. Future studies are essential (and many are ongoing) to identify why some individuals develop decline in language or dementia after stroke. Previous literature examining the longitudinal impact of stroke provides invaluable insight regarding the average trajectory of recovery for stroke survivors, largely analysing their samples as a whole. Here we propose a further granulation of this population along the lines of contralesional atrophy to help identify groups within the population that may experience vastly different trajectories of recovery. Our team contributes a novel lens to the literature by pairing longitudinal MRI with behavioural data to examine the correlation between atrophy of the right (contralesional) hemisphere and naming accuracy during the year after left hemisphere ischaemic stroke. Through this lens, we hope to partially address the uncertainty regarding why some individuals experience continued decline after stroke and others see improvement. These results shine light on the potential implications of post-stroke contralesional atrophy and its downstream effects on language recovery. While this has merit on its own, our research may provide greater predictive value if considered as a singular component for future multifactorial models. Lesion characteristics like infarct size and location,39 overall brain health based on factors such as white matter hyperintensities40,41 and brain age,42 and other predictive factors including demographics43 and social determinants of health44 are all potential contributors to future models, though this list is by no means exhaustive. Better predictive tools in the hands of clinicians would in turn improve the efficacy of targeted therapies and lifestyle modification to optimize patient outcomes and help maintain function. Intervention could take on various forms, with the most established practices promoting vascular risk management45 and cognitive engagement.46 Subsequent studies will hopefully support additional methods for improving patient outcomes post-stroke when paired with more accurate, individualized means of prediction. Areas of interest include, but are not limited to, anti-inflammatory therapies,47 gene-based approaches,48 and non-invasive brain stimulation (NIBS) techniques leveraging the brain’s neuroplastic potential, such as transcranial magnetic stimulation49 and transcranial direct current stimulation.50 It is worthwhile to note as an example of synergistic developments that the bimodal balance-recovery model from our earlier discussion is promising as a means to facilitate effective incorporation of NIBS based on predicted patterns of interhemispheric balancing and response to therapeutic intervention.37 The long-term ramifications of our findings are limited by the time frame in which data collection occurred, as no follow-ups were completed beyond the 1-year mark, nor did we have any data prior to participants’ hospital admission to determine cognitive trajectory before their stroke. Other larger scale epidemiological studies of post-stroke cognitive decline have followed their cohort for years before and after stroke, though typically with a trade-off in the ability to attain repeated imaging. Our findings would be strengthened if follow-up data were available at time points beyond the 1-year mark, and future work may be beneficial to determine how the trends presented here hold up when assessed further along the course of stroke recovery. Results from the ongoing DISCOVERY study (NCT04916210) will likely provide insights. There are other limitations to our study as well. We must be careful not to overstate the generalizability of our findings to overall cognitive decline, as the correlation presented is between hemispheric atrophy and naming performance after left hemisphere ischaemic strokes specifically. There are certainly other domains of language that may be impacted and even more domains of cognition that may be impacted following stroke. Therefore, it will be valuable going forward to examine if greater contralesional atrophy is similarly correlated to decline in other areas such as memory and executive function. It will also be pertinent to examine if this trend persists when hemispheres are reversed (in cases of right hemisphere lesions), a distinction that is not easily examined with language due to its mainly left hemisphere-dominant nature. Furthermore, we examined atrophy of each hemisphere as a whole in our analysis. A more granular approach incorporating lesion load and volume loss of specific regions and pathways that are associated with naming recovery (e.g. the left posterior superior temporal gyrus and superior longitudinal fasciculus/arcuate fasciculus)9 might add value to our conclusions and will be a key future direction. However, we note that per cent damage to these areas significantly correlates with total lesion volume in people with left middle cerebral artery strokes (as in this sample). Finally, plasma or cerebrospinal fluid biomarkers of specific dementias, such as Alzheimer’s disease, would be useful to confirm the presence of independent neurodegenerative diseases. In conclusion, we utilized longitudinal and imaging data collected both acutely and chronically (either 6- or 12-months) from 72 individuals after left hemisphere ischaemic stroke to investigate the relationships between (i) stroke size and hemispheric atrophy and (ii) hemispheric atrophy and recovery of naming performance. Our analysis revealed a significant correlation between stroke volume and the rate of atrophy in the lesioned (left) hemisphere, but not of the contralesional (right) hemisphere, implying that accelerated atrophy of the non-infarcted hemisphere may indicate separate pathologies. We then saw that atrophy of the contralesional hemisphere had a significant negative impact on change in naming accuracy, but the same negative impact was not correlated with atrophy of the lesioned hemisphere. Combining these two findings led us to conclude that atrophy of the contralesional hemisphere may have negative implications for language recovery and could serve as a potential imaging signature for the onset of post-stroke cognitive decline caused by neurodegenerative processes such as vascular dementia and/or Alzheimer’s disease. Supplementary Material