The human thalamus is a brain structure composed of numerous highly specific nuclear groups. Different thalamic nuclei have diverse functions (such as processing speed, attention, and executive functions) and are connected to different regions of the cerebral cortex, there is great interest in the neuroimaging community to study their volume, shape, and connectivity (Corona et al., 2020; Iglesias et al., 2018). The thalamus and basal ganglia collectively constitute a vital component of the cortico-striato-thalamo-cortical (CSTC) circuit. The anatomical vulnerability of the lenticulostriate arteries makes capsular infarction the predominant type of ischemic stroke, which usually involves the integrity of the CSTC circuit (Herrero et al., 2002; Yan et al., 2022). In such instances, the thalamus typically experiences secondary degenerative alterations, which not only compromise its structural integrity but also significantly modify its functional state, potentially resulting in motor and cognitive impairments (Ilves et al., 2022; Steiner et al., 2022). For example, some studies have shown that capsular stroke (CS) can lead to frontal dysfunction or memory impairment (Pantoni et al., 2001). The symptoms of frontal dysfunction are thought to be caused by disrupted connections between the thalamus and the frontal cortex (Biesbroek et al., 2017; Tatemichi et al., 1992; Zhang et al., 2021). This phenomenon may be ascribed to secondary degeneration of the thalamus, and this process is not confined solely to the ipsilesional thalamus, with the potential for involvement of the contralateral thalamus as well (Carey et al., 2011; Carmichael et al., 2004; Yassi et al., 2015).

While capsular strokes impact the adjacent thalamus, research often treats the thalamus as a single unit (Liu et al., 2019). However, the thalamus is highly heterogeneous, with distinct subfields differing in anatomical structure and function (Beloozerova and Marlinski, 2020; Lambert et al., 2017; Li et al., 2023). Each subfield has unique connections to the cortex, mediating specific functions. For example, the ventral lateral anterior (VLa) nucleus, part of the motor thalamus, connects with the substantia nigra and prefrontal cortex and functions as motor planning and learning (Kumar et al., 2014); The mediodorsal (MD) nucleus connects with the medial temporal and dorsolateral prefrontal cortex, and pallidum, and is involved in working memory and sensorimotor integration (Lambert et al., 2017). Critically, the integrity of the thalamocortical connection between the MD and dorsolateral prefrontal cortex is crucial for motor recovery post-stroke (Choi et al., 2020). Given this functional divisions of the thalamic subfields within the cortico-striato-thalamo-cortical (CSTC) circuit (Haber and McFarland, 2002; Mair et al., 2022), analyzing the entire thalamus’s volume obscures the specific cognitive and sensorimotor deficits caused by stroke.

Structural covariance network (SCN) is a valuable metric for investigating the brain’s topological organization, as it reflects the coordinated development and synchronous impact of interconnected regions (Liu et al., 2023; Zhang et al., 2025; Zielinski et al., 2010). This SCN is closely linked to cognitive abilities (Khundrakpam et al., 2017), genetic variant (Wen et al., 2023) and expression (Romero-Garcia et al., 2018) and neuropsychiatric disorders including stroke (Liu et al., 2023; Zhang et al., 2025). Therefore, examining both thalamic subfieldal structure and the structural covariance between thalamic subfields and remote cortical areas can improve our ability to identify the specific circuits disrupted by capsular stroke. This, in turn, may provide avenues for targeted rehabilitation.

In this context, we hypothesized that gray matter volume (GMV) covariance patterns in thalamic subfields would be selectively altered in patients with chronic capsular stroke. To test this hypothesis, we enrolled 45 patients with right internal capsule stroke and 93 healthy controls. We investigated whether the GMV covariance of thalamic subfields differed significantly between the two groups. Furthermore, we explored the potential associations between GMV covariance in these thalamic subfields and clinical behavior. This study aims to enhance our understanding of the GMV covariance alterations pattern of different thalamic subfields caused by capsular stroke and its subsequent functional consequences.

This study systematically investigates alterations in thalamic subfield structural covariance networks (SCNs) in patients with chronic right capsular stroke (CS), providing novel insights into thalamocortical reorganization mechanisms underlying post-stroke recovery. Utilizing high-resolution thalamic segmentation, we identified three key findings: (1) Subfield-level SCN mapping demonstrates superior sensitivity in detecting network-specific disruptions compared to conventional whole-thalamus approaches; (2) The GMV covariance networks of thalamic subfields are selectively involved in patients with chronic capsular stroke; and (3) contralesional thalamic subfields exhibit adaptive SCN remodeling despite preserved gray matter volume. These findings challenge the traditional whole-thalamus analytical framework and establish the critical role of subfield-specific network reorganization in stroke recovery, offering novel spatial-resolution perspectives for understanding neural plasticity, thereby enhancing our comprehension of the distinct functions played by thalamic subfields in the pathological course of CS.

The thalamus comprises functionally differentiated subfields, with distinct nuclei mediating specialized neural processes (Corona et al., 2020; Iglesias et al., 2018). The fundamental limitation of whole-thalamus analyses stems from the forced homogenization of inherently heterogeneous structures (Yuan et al., 2015), which obscures subfield-level functional antagonism or complementary patterns through artificial neutralization. While this simplification reduces analytical complexity, it critically compromises the understanding of fine-grained thalamic functional architecture and the pathophysiological mechanisms underlying thalamic-related disorder (Antonucci et al., 2021; Lin et al., 2018; McKenna et al., 2023; Yang et al., 2023). The high-resolution thalamic segmentation framework proposed by Iglesias et al. (2018) enables subfield-level structural covariance network (SCN) analysis, substantially enhancing our capacity to detect post-stroke thalamocortical network pathology. In our study, the whole-thalamus analysis identified only ipsilesional ventral visual cortex-thalamus SCN enhancement (Figure. 3B). In contrast, subfield-specific analysis revealed significant GMV covariance alterations in 16/25 ipsilesional and 3/25 contralesional thalamic subfields (Figures 4A, 5A), demonstrating the superior sensitivity of subfield-level approaches—consistent with prior studies showing that subfield decomposition uncovers otherwise hidden network changes (d'Ambrosio et al., 2017; Tung et al., 2022). Analogous methodology in schizophrenia research highlights the value of subfieldal decomposition—amygdala subfield analysis reveals more extensive GMV covariance abnormalities and distinct subfieldal engagement patterns, advancing understanding of limbic pathophysiology (Chang et al., 2024). These convergent findings underscore the critical advantage of subfield-level paradigms in elucidating network-level neuropathology.

While the thalamus does not receive direct vascular supply from the middle cerebral artery (MCA), accumulating evidence suggests that MCA infarction may induce secondary microstructural damage to the ipsilesional thalamus through remote degenerative mechanisms (Herve et al., 2005; Liu et al., 2019). Our findings extend these observations by demonstrating significant volumetric reduction of the ipsilesional thalamus in chronic capsular stroke patients, consistent with the thalamic atrophy patterns reported in prior studies (Liu et al., 2019; Wang et al., 2019). Through high-resolution thalamic parcellation into 25 distinct subfields, we further revealed that this volume loss occurred diffusely across all thalamic subfields compared to healthy controls. This whole thalamic degeneration pattern implies that secondary thalamic damage following capsular stroke may involve more extensive disconnection mechanisms than previously recognized (Herve et al., 2005; Liu et al., 2019; Wang et al., 2019), potentially disrupting both corticothalamic and thalamostriatal pathways (Haber and Calzavara, 2009; Lambert et al., 2017). The specific vulnerability of thalamic nuclei to retrograde degeneration might stem from their pivotal role as multimodal integration hubs within distributed cortical–subcortical networks (Hwang et al., 2017; Sherman, 2016; Tu et al., 2023).

Despite this widespread atrophy of thalamic subfields, SCN enhancements were selective: 16 ipsilesional subfields (e.g., VPL, VM, PuM) showed increased covariance with specific cortical regions, suggesting that structural covariance remodeling is not a random response but a targeted adaptation to preserve critical functions. We speculated that this may be related to the location of the lesions (Guo et al., 2019; Jiang et al., 2017; Wang et al., 2019). We observed that different thalamic subfields could demonstrate covariation with the same cerebral cortex.

For instance, ipsilesional thalamic subfields (VPL, VM, AV, VLa, LGN, and VLp) exhibited strengthened SCN with the superior occipital gyrus (SOG)—a region central to visual processing and spatial attention (Hageter et al., 2023; Reinhold et al., 2023). These subfields are key nodes in somatosensory, cognitive, visual, and motor pathways (Choi et al., 2022; Lambert et al., 2017), their enhanced covariance with the SOG likely reflects cross-modal compensation: as ascending sensory pathways are disrupted by capsular lesions (Herrero et al., 2002), residual connections between thalamic subfields and the SOG may be reinforced to maintain visual-somatosensory integration. Similarly, the ipsilesional OFCant also exhibited increased SCN with several thalamic subregions, such as VPL, VM, PuM, PuI, CM, and Pt. As research conducted by Bechara et al. (2000) has demonstrated, the OFC plays a vital role in decision-making and emotional regulation. This pattern suggests adaptive remodeling of thalamocortical loops supporting cognitive-emotional functions, consistent with prior findings that thalamo-cortical connectivity preserves affective stability post-stroke (Steiner et al., 2022). Furthermore, the strengthened SCN between the ipsilesional sensorimotor cortex (SMC) and the ipsilesional LD and MDm is also notable. According to previous studies, the MDm are associated with motor control and integration of sensory and motor information (Choi et al., 2020; Stockert et al., 2023).

In striking contrast to the ipsilesional thalamus, contralesional subfields showed no significant volume changes but exhibited enhanced SCN in 3/25 subfields—specifically with the ipsilesional OFCant, thalamus, and striatum. This dissociation between structural volume and covariance dynamics highlights a unique form of neural plasticity: the contralesional thalamus, though structurally intact, reorganizes its network connections to compensate for ipsilesional deficits. This pattern is best explained by the “diaschisis” effect—a phenomenon where focal brain lesions induce functional or structural changes in remote, interconnected regions (Carmichael et al., 2004; Seitz et al., 1999). Previous study (Yassi et al., 2015) demonstrated that contralesional thalamic microstructure is altered post-stroke, linked to functional disconnection of ipsilateral cortical regions; our findings extend this by showing that such changes manifest as enhanced structural covariance with the ipsilateral OFCant. Specifically, the contralesional thalamus may leverage interhemispheric projections (e.g., via the corpus callosum) to assume a “relay” role: as direct pathways from the ipsilesional thalamus to OFCant are damaged, contralesional subfields strengthen indirect connections to sustain critical cognitive-emotional functions. This is supported by Dermon and Barbas (1994) who identified robust contralateral thalamic projections to orbital cortices in non-human primates, providing an anatomical substrate for such cross-hemispheric compensation.

Previous animal studies have demonstrated that the majority of thalamic nuclei project into one or more well-defined cortical areas (Jones, 1990; Wise et al., 1996). The connection between the thalamus and the cortex is not a simple one-to-one relationship; rather, multiple cortical regions receive input from a single thalamic nucleus and transmit information back to different thalamic nuclei (Percheron et al., 1996). Hwang’s research also reveals significant functional connections between each nucleus and a minimum of three or more functional networks, and numerous cortical functional networks also demonstrate robust functional connections with multiple thalamic nuclei (Hwang et al., 2017). In our study, we found that some thalamic subfields exhibited increased GMV covariance with different cortical regions, which not only belonged to different functional networks, but also spanned to the contralesional hemisphere. For example, the strengthened SCN between the ipsilesional VPL and VM and bilateral SOG and ipsilesional OFCant. This may be the fact that thalamic nuclei play important roles in various cortical systems, and there seems to be overlapping patterns of connections among certain nuclei (Sherman and Usrey, 2024; Yuan et al., 2015). This overlap results in heterogeneity in the functional characteristics of individual nuclei, further compounding the complexity of thalamic nuclei functional representations (Sherman, 2007, 2016). Overall, these findings highlight the differential vulnerability and plasticity of thalamocortical circuits at the subfield level, providing valuable insights into the neural mechanisms underlying the observed structural changes.

In this study, after applying multiple comparison corrections, no significant correlation was identified between altered GMV covariance and clinical metrics in CS patients and the NC group—likely due to limited clinical variability: CS patients exhibited generally good functional recovery and mild symptoms, while NC group had no cognitive impairments, and this reduces the range of clinical measures needed to detect strong correlations. However, adopting a less stringent threshold (p < 0.05, uncorrected) revealed several intriguing yet weak associations. In CS patients, S_ACC showed the highest number of these weak associations with thalamic SCN: 19 in ipsilesional thalamic subfields and 16 in contralesional subfields, followed by RAVLT_SR with 8 associations in ipsilesional subfields and 16 in contralesional ones. This asymmetry reflects stroke-induced reliance on the ipsilesional thalamus for spatial working memory, while contralesional subfields contribute more to verbal memory. At the subfield level, the ipsilesional L_Sg was the top node (N = 6), followed by the ipsilesional MGN (N = 5), while the contralesional MGN, LD, and Pt each had 4 associations—indicating that CS patients rely disproportionately on ipsilesional subfields to sustain cognitive function, with contralesional subfields playing a secondary compensatory role. The NC group exhibited S_ACC had the weak associations with right thalamic SCN (N = 15), followed by N_ACC (N = 14), with left thalamic subfields also showing weak working memory links. Unlike CS group, NCs exhibited a right-weighted but non-lateralized pattern. Additional, the NC group also shared weak associations with CS patients for key metrics (e.g., right thalamic SCN with RAVLT_LR, working memory indices; left thalamic SCN with RAVLT_LR), indicating preserved fundamental thalamic-cognitive links. These cross-group findings collectively suggest that while no associations reached strict statistical significance, certain cognitive measures (especially spatial working memory) exhibit subtle relationships with specific thalamic subfields in both groups. The nuanced pattern of weak associations highlights the complexity of the relationship between thalamic subfields GMV covariance and behavior in CS patients, underscoring the need for more refined analytical approaches to fully elucidate these interactions. Future studies should explore potential confounding factors and consider longitudinal designs to better understand the temporal dynamics of these associations.

Several limitations of our study should be considered. First, our patient cohort exhibits high homogeneity, characterized by nearly complete motor recovery (median FMA = 100, interquartile range [82.5, 100]) and mild cognitive deficits. This is likely due to practical barriers in recruiting chronic stroke patients with moderate-to-severe deficits, who often cannot complete MRI scans or standardized assessments. The low variability in clinical measures weakened our ability to detect associations with SCN changes, as statistical power for correlation analyses depends on sufficient range in both variables. Furthermore, this sample bias limits generalizability: our findings may specifically reflect SCN remodeling in well-recovered chronic capsular stroke, rather than the broader stroke population (including those with persistent motor/cognitive impairments). Thus, our results should be interpreted with caution and validated in cohorts encompassing diverse recovery outcomes. Second, this is a cross-sectional study, which constraining its capacity to monitor the temporal variations in the covariance patterns of GMV across thalamic subfields. Longitudinal studies are crucial for observing the volume changes in thalamic subfields in stroke patients, as well as the progression of covariance between these subfields and cortical structures. Moreover, they are also essential for establishing clearer causal relationships between thalamic subfield structural covariance patterns and clinical outcomes. Additional validation in cohorts encompassing greater symptom severity heterogeneity is warranted to confirm these findings. Finally, our current research focuses on the structural covariance characteristics of thalamic subfields, but the study of the causal relationship with the structural covariance of the cerebral cortex is still shallow. In the future, we will deepen the research on the causal structural covariance of thalamic subfields to more accurately reveal the mechanism of its structural covariance with the cerebral cortex after stroke. This is crucial for understanding the interaction between them and the impact of stroke.

In conclusion, our research investigated the SCN of thalamic subfields in patients with capsular stroke for the first time. The GMV covariance networks of thalamic subfields, which are selectively involved in patients with chronic capsular stroke, affect not only the ipsilesional thalamic subfields but also the contralesional ones. These subfield-specific SCN disruptions likely reflect neurodegenerative cascades and compensatory plasticity mechanisms post-stroke, with implications for both motor and cognitive functional reorganization. Future research should prioritize longitudinal designs to delineate dynamic GMV covariance patterns across thalamic subfields and validate their clinical translatability as biomarkers of post-stroke.