Predicting Functional Prognosis in Supratentorial Stroke Using [18F]Fluoro-2-deoxy-D-glucose Positron Emission Tomography Authors: Jong Weon Lee, Suk Hoon Ohn, Yong Won Jang, Won Hyuk Chang, Ki Deok Park, Deog Young Kim Published: 2026 ABSTRACT AbstractPurposeCerebral metabolic activity following stroke plays a crucial role in neurological recovery. This study aimed to investigate whether decreased cerebral metabolic activity, quantified by the asymmetry index (AI) using [18F]fluoro-2-deoxy-D-glucose (FDG) positron emission tomography (PET), predicts functional prognosis at 3 and 6 months in patients with supratentorial stroke.Materials and MethodsForty patients with a first-time supratentorial stroke were enrolled. Neurological status was quantified using the National Institutes of Health Stroke Scale (NIHSS) and the Motricity Index (MI) at baseline (time of PET acquisition), and at 3 and 6 months after stroke onset. Using PET imaging, metabolic AI was calculated for both the cerebral and cerebellar hemispheres. Correlations between AI and neurologic outcomes were analyzed according to the timing of PET acquisition, classified as the acute (1–14 days) or subacute (15–60 days) phases.ResultsIn the subacute group, cerebral AI showed significant correlations with NIHSS scores at baseline (r=0.511, p<0.01) and with both NIHSS (r=0.589, p<0.01) and MI (r=−0.479, p<0.01) at 3 months. These correlations strengthened at 6 months for both NIHSS (r=0.724, p<0.01) and MI (r=−0.744, p<0.01). However, no significant correlations were found between cerebral AI and neurological outcomes in the acute group. Cerebellar AI did not correlate with functional outcomes at any time point.ConclusionCerebral metabolic asymmetry measured during the subacute phase of supratentorial stroke is a significant predictor of long-term neurological outcomes at 3 and 6 months.Keywords: Stroke, prognosis, positron emission tomography, fluorodeoxyglucose F18, recovery of runction INTRODUCTION Accurate prediction of prognosis in the early stages of stroke is essential for establishing realistic rehabilitation goals and optimizing resource allocation.1,2 However, stroke recovery patterns are highly heterogeneous; patients with similar initial lesion volumes often exhibit disparate recovery trajectories.3,4 Consequently, clinical assessment alone is frequently insufficient for precise prognostication during the early post-stroke period. Numerous clinical predictors—including trunk control, motor strength of key muscle groups, evoked potentials, and early functional scales—have been shown to correlate with functional outcomes after stroke.5,6,7,8,9,10 However, most clinical assessments require active patient participation and are therefore difficult to apply reliably during the early phase of severe stroke. In such clinical contexts, advances in neuroimaging have shifted the focus toward understanding the functional reorganization of the brain. Positron emission tomography (PET) using [18F] fluoro-2-deoxy-D-glucose (FDG) provides quantitative measurements of regional cerebral glucose metabolism, offering insight into cerebral activity and network integrity beyond structural damage.11 Previous studies have suggested that FDG PET can predict outcomes by assessing metabolic viability in the penumbra and metabolic connectivity.12,13,14 Specifically, the balance of metabolic activity between the affected and unaffected hemispheres—often quantified as the asymmetry index (AI)—has emerged as a potential marker of functional recovery.15 The AI quantifies metabolic imbalance between cerebral hemispheres, providing a summary measure of hemispheric dysfunction beyond the structural lesion itself. Unlike lesion volumetry, which reflects only the extent of visible tissue damage, metabolic asymmetry captures the broader functional impact of stroke on cortical and subcortical networks, including remote effects such as diaschisis and network-level disruption. This characteristic makes AI a useful indicator of the overall cerebral dysfunction and its potential influence on functional recovery. However, the relationship between AI and clinical outcomes may vary depending on the timing of PET acquisition relative to stroke onset, as the metabolic landscape evolves dynamically during the transition from acute to subacute phases. This study aimed to investigate the predictive value of metabolic asymmetry in supratentorial stroke by examining PET data acquired at different post-onset intervals (acute vs. subacute) and assessing the relationship between cerebral and cerebellar metabolic asymmetry and outcomes at baseline and at 3 and 6 months after stroke. The study hypothesized that metabolic asymmetry measured during the subacute phase would demonstrate stronger correlations with long-term functional outcomes than acute-phase measurements, as subacute metabolic patterns are more likely to reflect stable, irreversible tissue injury rather than transient metabolic disturbances. MATERIALS AND METHODS Study participants This study retrospectively reviewed the medical records of patients admitted to a single tertiary hospital with a first-ever unilateral supratentorial stroke. Forty patients who underwent FDG PET within 60 days of stroke onset and completed neurological assessments at baseline and at 3 and 6 months after stroke onset were included in the analysis. The diagnosis of stroke was confirmed by computed tomography or magnetic resonance imaging. Participants were stratified into two groups based on the interval between stroke onset and PET acquisition—the acute group (1–14 days) and the subacute group (15–60 days). These time windows were defined based on the pathophysiological evolution of stroke: the acute phase corresponds to the period of ongoing ischemic injury, cerebral edema, and penumbral instability, whereas the subacute phase represents the period following penumbral resolution, during which metabolic patterns are expected to more accurately reflect permanent tissue injury and early reorganization.16,17,18,19,20 No specific National Institutes of Health Stroke Scale (NIHSS), score range was applied as an inclusion criterion, as the study intended to capture the full spectrum of stroke severity encountered in clinical practice to enhance the generalizability of findings. The exclusion criteria included 1) a history of previous stroke or transient ischemic attack; 2) inability to undergo PET because of medical instability (e.g., uncontrolled diabetes or movement disorders); and 3) recurrent stroke during the 6-month follow-up period. All participants received standardized physical and occupational therapy during the follow-up period. The study protocol was approved by the Institutional Review Board of our hospital (IRB No. 4-2025-1644). PET procedure PET imaging was performed using a GE Advance PET scanner (General Electric Medical System, Milwaukee, WI, USA). All medications with potential cerebrovascular effects were withheld for 24 hours prior to scanning. Participants were positioned supine with their eyes closed in a dimly lit, quiet room to minimize sensory stimulation. Thirty minutes after the intravenous injection of 148 MBq (4 mCi) of [18F]FDG, a 10-minute emission scan and a 2-minute transmission scan were acquired in three-dimensional mode. Images were reconstructed using a Hanning filter (cutoff frequency 4.0 mm-1) for transaxial images and a Ramp filter (cutoff frequency 8.5 mm-1) for axial images. The resulting images consisted of 40 transaxial slices with a pixel size of 1.95×1.95 mm and a slice thickness of 3.91 mm. Attenuation correction was performed using a 68Ge pin source. Image analysis Cerebral and cerebellar asymmetries were calculated using a standardized region-of-interest (ROI) method adapted from Pantano, et al.21 ROIs were drawn over the entire ipsilesional and contralesional cerebral hemispheres as well as the bilateral cerebellar hemispheres on all relevant transaxial slices (Fig. 1). The AI was calculated using the following formula: Procedure for drawing regions of interest. where CR and CL represent the summed metabolic activity in the right and left hemispheres, respectively. A higher AI indicates greater metabolic imbalance. Assessment of neurological and motor function Neurological impairment was evaluated using the NIHSS, a standardized 15-item instrument that quantifies stroke severity across domains including consciousness, motor and sensory function, language, and visuospatial attention, with total scores ranging from 0 to 42, where higher scores indicate greater neurological deficit.22 Motor impairment was assessed using the Motricity Index (MI), which grades voluntary muscle strength across three movements of each limb. Each limb is scored on a scale from 0 to 100, with higher scores reflecting better motor function.23 Assessments were conducted at three time points—1) baseline (the time of PET acquisition), 2) 3 months after stroke onset, and 3) 6 months after stroke onset. Statistical analysis Continuous variables were compared using independent-samples t-tests for normally distributed data or the Mann–Whitney U test for non-normally distributed data. Normality was assessed using the Shapiro–Wilk test. Categorical variables were compared using Fisher’s exact test, as expected cell frequencies were below 5 due to the small sample size of the acute group. Variables that met the normality assumption were analyzed using Pearson’s correlation; otherwise, Spearman’s rank correlation was used. Correlation coefficients were calculated to analyze the relationships between AI (cerebral and cerebellar) and clinical scores (NIHSS and MI) at each time point. To evaluate whether cerebral AI provides independent prognostic information beyond lesion-related variables, multiple linear regression analyses were performed with NIHSS and MI scores at 6 months as dependent variables, entering cerebral AI, age, and baseline NIHSS as independent variables in the subacute group. To account for multiple comparisons across time points and outcome measures, the Benjamini–Hochberg false discovery rate (FDR) procedure was applied to the primary correlation analyses, with a threshold of q<0.01 considered statistically significant. The FDR method was chosen over the Bonferroni correction to avoid excessive conservatism given the correlated nature of the outcome measures and the limited sample size. Post hoc sample size estimation indicated that a sample size of 29 (subacute group) provided 80% power to detect a correlation coefficient of r=0.57 or greater at the α=0.01 significance level (two-tailed). Statistical analyses were performed using SPSS version 28.0 (IBM Corp., Armonk, NY, USA). RESULTS Baseline patient characteristics A total of 40 patients (27 males, 13 females), with an age of 62.5±11.3 years, were enrolled (Table 1). The etiology of stroke was cerebral infarction in 27 patients (67.5%) and intracerebral hemorrhage in 13 patients (32.5%). Lesions were located in the left hemisphere in 24 patients (60.0%) and in the right hemisphere in 16 patients (40.0%). Participants were categorized into acute (n=11) and subacute (n=29) groups based on the timing of PET acquisition. The interval from stroke onset to PET acquisition was 11.3±1.9 days (range: 8–14 days) in the acute group and 35.8±12.6 days (range: 19–55 days) in the subacute group (p<0.001). The two groups did not differ significantly in any demographic or functional characteristics (all p>0.05). Table 1Baseline Characteristics of ParticipantsVariablesTotal participants (n=40)Acute (n=11)Subacute (n=29) p Demographic characteristicsAge (yr)62.5±11.357.4±14.264.4±9.70.115Sex (M: F)27:138:319:100.999*Hypertension27 (67.5)6 (54.5)21 (72.4)0.451*Diabetes mellitus8 (20.0)0 (0.0)8 (27.6)0.080*Stroke type (ischemic:hemorrhagic)27:1310:117:120.068*Affected side (left:right)24:166:518:110.728*Onset to PET interval (days)28.0±15.011.3±1.935.8±12.6<0.001†Functional characteristicsBaseline NIHSS12.1±5.612.3±5.212.0±5.90.8323 months NIHSS8.3±4.67.7±3.48.5±5.00.7736 months NIHSS6.2±4.35.5±2.96.5±4.70.670Baseline MI23.1±24.435.5±27.118.4±22.00.055†3 months MI42.5±27.751.2±25.139.1±28.30.182†6 months MI61.4±25.066.3±14.159.6±28.00.671†PET, positron emission tomography; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index.Values are presented as mean±SD, number (%), or ratio. Continuous variables with normal distribution were compared using independent t-test.*Fisher’s exact test (expected cell count <5); †Mann–Whitney U test (non-normal distribution by Shapiro–Wilk test). Baseline Characteristics of Participants PET, positron emission tomography; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index. Values are presented as mean±SD, number (%), or ratio. Continuous variables with normal distribution were compared using independent t-test. *Fisher’s exact test (expected cell count <5); †Mann–Whitney U test (non-normal distribution by Shapiro–Wilk test). Association of baseline factors and functional outcomes Multiple linear regression analyses were performed to assess the relationship between functional outcomes and baseline independent variables in the subacute group (Table 2). In Model 1, with NIHSS at 6 months as the dependent variable, the overall model was significant (R2=0.86, adjusted R2=0.84, F(3,25)=51.48, p<0.001). Cerebral AI (B=0.17, β=0.38, p<0.001) and baseline NIHSS (B=0.51, β=0.64, p<0.001) were significant independent predictors, whereas age was not (B=0.03, β=0.06, p=0.423). In Model 2, with MI at 6 months as the dependent variable, the model was also significant (R2=0.73, adjusted R2=0.70, F(3,25)=22.53, p<0.001). Cerebral AI (B=−1.18, β=−0.46, p=0.001) and baseline NIHSS (B=−2.19, β=−0.46, p=0.001) were significant independent predictors, while age was not (B=−0.41, β=−0.14, p=0.200). All variance inflation factors were below 1.5, confirming the absence of multicollinearity. Table 2Multiple Linear Regression Analysis of Factors Associated with Neurological Outcomes at 6 Months in the Subacute Group (n=29)VariablesBSEβ p 95% CIVIFModel 1: NIHSS at 6 months (R2=0.861, adjusted R2=0.844, F(3,25)=51.483, p<0.001)Cerebral AI0.1650.0390.384<0.0010.085, 0.2441.448Age0.0310.0380.0630.423-0.047, 0.1081.078Baseline NIHSS0.5140.0720.640<0.0010.366, 0.6631.451Model 2: MI at 6 months (R2=0.73, adjusted R2=0.698, F(3,25)=22.529, p<0.001)Cerebral AI-1.1750.320-0.4590.001-1.835, -0.5161.448Age-0.4120.313-0.1420.200-1.057, 0.2321.078Baseline NIHSS-2.1850.599-0.4570.001-3.419, -0.9511.451B, unstandardized coefficient; SE, standard error; β, standardized coefficient; CI, confidence interval; VIF, variance inflation factor; AI, asymmetry index; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index. Multiple Linear Regression Analysis of Factors Associated with Neurological Outcomes at 6 Months in the Subacute Group (n=29) B, unstandardized coefficient; SE, standard error; β, standardized coefficient; CI, confidence interval; VIF, variance inflation factor; AI, asymmetry index; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index. Cerebral AI and functional outcomes In the subacute group, cerebral AI demonstrated significant correlations with NIHSS scores at baseline (r=0.511, FDR adjusted p<0.01) and with both NIHSS (r=0.589) and MI (r=−0.479) at 3 months (FDR adjusted p<0.01) (Table 3). Higher correlation coefficients were observed at 6 months for both NIHSS (r=0.724) and MI (r=−0.744), both of which were statistically significant (FDR adjusted p<0.01). In contrast, no significant correlations were observed between cerebral AI and neurological outcomes (NIHSS or MI) at any time point in the acute group. Table 3Correlations between Asymmetry Index and Neurological Outcomes by Stroke StageGroupRegionBaseline3 months6 monthsNIHSSMINIHSSMINIHSSMIAcute (n=11)Cbr AI0.452-0.2770.322-0.2590.541-0.444Cbll AI-0.140-0.405-0.234-0.275-0.109-0.373Subacute (n=29)Cbr AI0.511*†-0.2670.589*†-0.479*†0.724*†-0.744*†Cbll AI0.343-0.0700.261-0.1230.320-0.288Cbr AI, cerebral asymmetry index; Cbll AI, cerebellar asymmetry index; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index.Values represent Pearson’s correlation coefficients.*p<0.01; †Remained significant after Benjamini–Hochberg false discovery rate correction (adjusted p<0.01). Correlations between Asymmetry Index and Neurological Outcomes by Stroke Stage Cbr AI, cerebral asymmetry index; Cbll AI, cerebellar asymmetry index; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index. Values represent Pearson’s correlation coefficients. *p<0.01; †Remained significant after Benjamini–Hochberg false discovery rate correction (adjusted p<0.01). Subgroup analysis by stroke type When stratified by stroke type within the subacute group, significant correlations were observed primarily in the ischemic group (n=17). In patients with ischemic stroke, cerebral AI correlated with NIHSS scores at baseline (r=0.711, FDR adjusted p<0.01) and at 3 months (r=0.686, FDR adjusted p<0.01) (Table 4). At 6 months post-stroke, cerebral AI demonstrated significant correlations with both NIHSS (r=0.759) and MI (r=−0.813), both of which were statistically significant (FDR adjusted p<0.01). The hemorrhagic stroke group (n=12) did not show statistically significant correlations with neurological outcome measures. Table 4Correlations between Asymmetry Index and Neurological Outcomes by Stroke Type in the Subacute StageStroke typeRegionBaseline3 months6 monthsNIHSSMINIHSSMINIHSSMIHemorrhagic (n=12)Cbr AI0.436-0.3430.594-0.4970.526-0.391Cbll AI-0.0110.0050.109-0.1780.022-0.019Ischemic (n=17)Cbr AI0.711*†-0.3930.686*†-0.5740.759*†-0.813*†Cbll AI0.514-0.1340.296-0.1090.321-0.288Cbr AI, cerebral asymmetry index; Cbll AI, cerebellar asymmetry index; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index.Values represent Pearson’s correlation coefficients.*p<0.01; †Remained significant after Benjamini–Hochberg false discovery rate correction (adjusted p<0.01). Correlations between Asymmetry Index and Neurological Outcomes by Stroke Type in the Subacute Stage Cbr AI, cerebral asymmetry index; Cbll AI, cerebellar asymmetry index; NIHSS, National Institutes of Health Stroke Scale; MI, Motricity Index. Values represent Pearson’s correlation coefficients. *p<0.01; †Remained significant after Benjamini–Hochberg false discovery rate correction (adjusted p<0.01). Cerebellar AI and functional outcomes Cerebellar AI did not show significant correlations with NIHSS or MI scores at any time point, regardless of the timing of PET acquisition or stroke subtype (Tables 3 and 4). DISCUSSION This study demonstrated that cerebral metabolic asymmetry measured during the subacute phase—but not during the acute phase—is a significant predictor of long-term functional recovery after supratentorial stroke. Specifically, a higher cerebral AI, reflecting greater hypometabolism in the affected hemisphere, was associated with more severe neurological impairment and worse motor function at 3 and 6 months after stroke onset. Notably, these associations strengthened over time, suggesting that subacute metabolic asymmetry reflects more stable and enduring pathophysiological processes that influence recovery trajectories rather than transient metabolic disturbances present during the acute phase. Alterations in cerebral metabolism after stroke and their relationship to neurological recovery have been investigated using PET. Marchal, et al.13,24 demonstrated that cerebral hypometabolism on PET and hypoperfusion on SPECT during the acute stage were associated with neurological outcomes at 2 months. Heiss, et al.25 also reported that preserved glucose metabolism in non-infarcted brain regions correlated with better recovery. However, these studies focused primarily on hyperacute to early acute-stage imaging, during which metabolic measurements may be influenced by transient factors such as ongoing ischemia, cerebral edema, and inflammatory responses, potentially limiting their prognostic specificity.20 The apparent discrepancy between the present findings and prior studies reporting acute-phase prognostic value may reflect several methodological differences, including the imaging time window (within 24 hours in Marchal, et al.13,24 vs. 1–14 days in our acute group), outcome assessment timing (approximately 2 months in previous studies versus 6 months in the present study), and the relatively small size of the acute group (n=11), which may have been underpowered to detect moderate correlations. This study directly compared the prognostic value of PET acquired during the acute and subacute phases within the same cohort and demonstrated that metabolic asymmetry assessed in the subacute phase provides superior prognostic information. This finding aligns with the observations of Iglesias, et al.,17 who reported that delayed intrahemispheric hypometabolism correlated with early recovery, and supports the concept that subacute metabolic imaging more accurately reflects persistent functional impairment and recovery potential rather than transient metabolic disturbances. The superior prognostic value of subacute FDG PET compared with acute FDG PET may be explained by the temporal evolution of the ischemic penumbra. During the acute phase (days 1–14), the penumbra—defined as hypoperfused but potentially salvageable tissue—may demonstrate heterogeneous metabolic patterns influenced by dynamic changes in cerebral perfusion, spontaneous or treatment-induced reperfusion, and early inflammatory responses.18,19 Consequently, metabolic snapshots obtained during this unstable period may not reliably represent the ultimate extent of irreversible tissue injury. By the subacute phase (after 2 weeks), the ischemic penumbra has largely resolved into either infarcted or salvaged tissue, rendering regional metabolic deficits more stable and reflective of permanent neuronal loss and residual functional capacity.17 Furthermore, within the 15–60 day window used in this study, the initial inflammatory cascade (microglial activation, cytokine release) has largely subsided,26 edema has resolved,27 and collateral flow patterns have stabilized,17,18 thereby creating metabolic conditions that more faithfully represent the permanent structural and functional landscape of the post-stroke brain. Accordingly, cerebral metabolic asymmetry assessed during this phase is more likely to capture clinically meaningful information relevant to long-term recovery. The progressive strengthening of the correlation between cerebral AI and functional outcomes over time (from baseline to 6 months) further supports the role of network reorganization in post-stroke recovery. This observation is consistent with the concept of “interhemispheric rivalry,” whereby an imbalance in inhibitory and excitatory interactions between the affected and unaffected hemispheres influences cortical plasticity.28,29 Persistent hypometabolism in the affected hemisphere may reflect not only local structural injury but also sustained disruption of interhemispheric connectivity required for neurological recovery.30 Recent studies examining metabolic and functional connectivity have reinforced the notion that hemispheric asymmetry, particularly within the middle cerebral artery territory, is a key determinant of long-term outcomes in ischemic and cryptogenic stroke.31,32 These findings underscore the importance of evaluating hemispheric metabolic balance when interpreting the prognostic significance of post-stroke PET imaging. In this study, cerebellar metabolic asymmetry did not demonstrate a significant association with functional outcomes at any time point, despite the well-established phenomenon of crossed cerebellar diaschisis (CCD). Although CCD reflects disruption of cortico-ponto-cerebellar pathways secondary to supratentorial lesions, its prognostic significance remains controversial.33,34,35 Serrati, et al.33 reported that CCD persisting on later PET imaging (13–56 days after stroke onset) was associated with neurological recovery assessed using the Mathew and Orgogozo scales, whereas CCD detected on very early PET imaging (5–30 hours after onset) showed no prognostic relevance. In contrast, Chen, et al.35 suggested that CCD does not independently predict long-term functional outcomes when stroke volume and initial severity are taken into account. In this study, the absence of significant correlations between cerebellar AI and functional outcomes in both the acute and subacute phases, regardless of stroke subtype, supports the notion that cerebral cortical metabolic integrity, rather than remote cerebellar metabolic changes, plays a more central role in determining functional prognosis after supratentorial stroke. One possible explanation for the lack of prognostic relevance of cerebellar AI in this cohort is that the supratentorial lesions may not have sufficiently involved the cortico-ponto-cerebellar pathway to produce clinically significant CCD. In this study, cerebral AI in the subacute stage was significantly associated with NIHSS scores at baseline, whereas association with MI emerged at 3 months. Moreover, cerebral AI in ischemic stroke showed significant associations at all time points, whereas an association with MI was observed only at 6 months. The NIHSS22 captures a broad range of neurological domains, including consciousness, language, motor impairment, visuospatial function, ataxia, and sensory deficits, which are closely linked to distributed cortical network integrity and global hemispheric metabolism, whereas MI23 primarily reflects focal motor strength and may be more susceptible to compensatory strategies and task-specific recovery. Accordingly, cerebral AI, which represents whole-hemisphere metabolic imbalance, may correspond more directly to global neurological impairment than to isolated motor performance. The stronger associations observed in ischemic stroke likely reflect the pathophysiology of ischemia, in which metabolic dysfunction extends beyond the infarct core to surrounding and remote cortical regions, making FDG PET particularly sensitive to ischemia-related network disruption.11 In contrast, hemorrhagic stroke involves additional factors such as mass effect, perihematomal edema, and hematoma resolution, which may obscure the relationship between metabolic asymmetry and functional outcome.36 This study has certain limitations. First, the sample sizes of each subgroup, stratified by imaging timing (acute vs. subacute) and stroke subtype (ischemic vs. hemorrhagic), were relatively small, which may have limited statistical power. Second, the subacute imaging window (15–60 days post-stroke) was relatively broad. Given that cerebral metabolic activity evolves dynamically during this period through compensatory mechanisms and neural reorganization, variability in PET timing within this window may have influenced the observed associations. Third, this study relied primarily on correlation analyses without adjustment for potential confounding factors such as lesion volume, lesion location, or rehabilitation intensity. Fourth, the use of asymmetry indices, while robust and clinically practical, may have obscured region-specific metabolic alterations within functionally critical cortical or subcortical networks. To address these limitations, future studies should include larger, prospectively enrolled cohorts with narrower imaging intervals (e.g., 2–3 week windows) and employ multivariate regression models adjusting for age, baseline stroke severity, lesion volume, lesion location, and rehabilitation intensity. The incorporation of voxel-based morphometry or region-specific ROI analyses targeting motor cortex and language areas would help determine whether AI provides independent prognostic value and identify which specific cortical networks drive the observed associations. Additionally, machine learning-based predictive modeling could be applied to evaluate the true predictive accuracy of metabolic asymmetry measures. In conclusion, FDG PET performed during the subacute stage of supratentorial stroke may serve as a clinically meaningful prognostic tool. Cerebral metabolic asymmetry was significantly associated with neurological recovery, with stronger correlations observed at long-term follow-up. However, cerebellar AI did not demonstrate prognostic relevance. These findings suggest that subacute FDG PET-derived measures of cerebral metabolic integrity may complement conventional clinical assessments by assisting clinicians in estimating recovery potential and in formulating individualized rehabilitation goals. Footnotes The authors have no potential conflicts of interest to disclose. AUTHOR CONTRIBUTIONS: Conceptualization: Suk Hoon Ohn and Deog Young Kim.Data curation: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim.Formal analysis: Jong Weon Lee and Suk Hoon Ohn.Funding acquisition: Deog Young Kim.Investigation: all authors.Methodology: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim.Project administration: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim.Resources: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim.Software: Jong Weon Lee.Supervision: Suk Hoon Ohn and Deog Young Kim.Validation: Jong Weon Lee and Suk Hoon Ohn.Visualization: Jong Weon Lee.Writing—original draft: Jong Weon Lee and Suk Hoon Ohn.Writing—review & editing: Deog Young Kim.Approval of final manuscript: all authors. Conceptualization: Suk Hoon Ohn and Deog Young Kim. Conceptualization: Suk Hoon Ohn and Deog Young Kim. Data curation: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Data curation: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Formal analysis: Jong Weon Lee and Suk Hoon Ohn. Formal analysis: Jong Weon Lee and Suk Hoon Ohn. Funding acquisition: Deog Young Kim. Funding acquisition: Deog Young Kim. Investigation: all authors. Investigation: all authors. Methodology: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Methodology: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Project administration: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Project administration: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Resources: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Resources: Jong Weon Lee, Suk Hoon Ohn, and Deog Young Kim. Software: Jong Weon Lee. Software: Jong Weon Lee. Supervision: Suk Hoon Ohn and Deog Young Kim. Supervision: Suk Hoon Ohn and Deog Young Kim. Validation: Jong Weon Lee and Suk Hoon Ohn. Validation: Jong Weon Lee and Suk Hoon Ohn. Visualization: Jong Weon Lee. Visualization: Jong Weon Lee. Writing—original draft: Jong Weon Lee and Suk Hoon Ohn. Writing—original draft: Jong Weon Lee and Suk Hoon Ohn. Writing—review & editing: Deog Young Kim. Writing—review & editing: Deog Young Kim. Approval of final manuscript: all authors. Approval of final manuscript: all authors. DATA SHARING STATEMENT The data associated with the paper are not publicly available but may be obtained from the corresponding author upon reasonable request. References Associated Data Data Availability Statement The data associated with the paper are not publicly available but may be obtained from the corresponding author upon reasonable request.