BACKGROUND/OBJECTIVES: Early and accurate prediction of stroke severity is crucial for optimizing guided therapeutic decisions and improving outcomes. This study investigates the predictive value of lesion size and functional connectivity for neurological deficits, assessed by the National Institutes of Health Stroke Scale (NIHSS score), in patients with acute or subacute subcortical ischemic stroke. METHODS: Forty-four patients (mean age: 68.11 years, 23 male, and admission NIHSS score 4.30 points) underwent high-resolution anatomical and resting-state functional Magnetic Resonance Imaging (rs-fMRI) within seven days of stroke onset. Lesion size was volumetrically quantified, while functional connectivity within the motor, default mode, and frontoparietal networks was analyzed using seed-based correlation methods. Multiple linear regression and cross-validation were applied to develop predictive models for stroke severity. RESULTS: Our results showed that lesion size explained 48% of the variance in NIHSS scores (R = 0.48, cross-validated R = 0.49). Functional connectivity metrics alone were less predictive but enhanced model performance when combined with lesion size (achieving an R = 0.71, cross-validated R = 0.73). Additionally, left hemisphere connectivity features were particularly informative, as models based on left-hemispheric connectivity outperformed those using right-hemispheric or bilateral predictors. This suggests that the inclusion of contralateral hemisphere data did not enhance, and in some configurations, slightly reduced, model performance-potentially due to lateralized functional organization and lesion distribution in our cohort. CONCLUSIONS: The findings highlight lesion size as a reliable early marker of stroke severity and underscore the complementary value of functional connectivity analysis. Integrating rs-fMRI into clinical stroke imaging protocols offers a potential approach for refining prognostic models. Future research efforts should prioritize establishing this approach in larger cohorts and analyzing additional biomarkers to improve predictive models, advancing personalized therapeutic strategies for stroke management.