Abstract
OBJECTIVE: The aim of this study was to investigate the characteristics of brain activity changes in patients with post-stroke balance dysfunction and their relationship with clinical assessment, and to construct a classification model based on the extreme Gradient Boosting (XGBoost) algorithm to discriminate between stroke patients and healthy controls (HCs). METHODS: In the current study, twenty-six patients with post-stroke balance dysfunction and twenty-four HCs were examined by resting-state functional magnetic resonance imaging (rs-fMRI). Static amplitude of low frequency fluctuation (sALFF), static fractional ALFF (sfALFF), static regional homogeneity (sReHo), dynamic ALFF (dALFF), dynamic fALFF (dfALFF) and dynamic ReHo (dReHo) values were calculated and compared between the two groups. The values of the imaging metrics for the brain regions with significant differences were used in Pearson correlation analyses with the Berg Balance Scale (BBS) scores and as features in the construction of the XGBoost model. RESULTS: Compared to HCs, the brain regions with significant functional abnormalities in patients with post-stroke balance dysfunction were mainly involved bilateral insula, right fusiform gyrus, right lingual gyrus, left thalamus, left inferior occipital gyrus, left inferior temporal gyrus, right calcarine fissure and surrounding cortex, left precuneus, right median cingulate and paracingulate gyri, right anterior cingulate and paracingulate gyri, bilateral supplementary motor area, right putamen, and left cerebellar crus II. XGBoost results show that the model constructed based on static imaging features has the best classification prediction performance. CONCLUSION: In conclusion, this study provided evidence of functional abnormalities in local brain regions in patients with post-stroke balance dysfunction. The results suggested that the abnormal brain regions were mainly related to visual processing, motor execution, motor coordination, sensorimotor control and cognitive function, which contributed to our understanding of the neuropathological mechanisms of post-stroke balance dysfunction. XGBoost is a promising machine learning method to explore these changes.