机器人外科学杂志（中英文）2025，Vol.6Issue(11)：1902-1908,7.DOI:10.12180/j.issn.2096-7721.2025.11.015
基于康复机器人辅助训练诱导的脑网络响应特征构建意识障碍患者预后预测模型
Prognostic prediction model for patients with disorders of consciousness based on brain network response characteristics induced by robot-assisted rehabilitation training
摘要
Abstract
Objective:To analyze the characteristics of brain network responses induced by robot-assisted rehabilitation training in patients with disorders of consciousness(DoC)and construct a prognostic prediction model.Methods:80 DoC patients who were treated at Beijing Rehabilitation Hospital Affiliated to Capital Medical University from March 2022 to March 2025 were enrolled.Patients were randomly divided into the control group(n=40,receiving conventional rehabilitation)and the observation group(n=40,receiving robot-assisted training combined with conventional rehabilitation)using the computer-generated randomization.Both groups received 8 weeks of continuous intervention.Coma recovery scale-revised(CRS-R)scores before and after intervention,functional connectivity(FC)strength of key brain networks(PCC-mPFC,DMN-SN,S1/M1-SMA),high-density EEG(hdEEG)microstate parameters(microstate C coverage,microstate D duration,transition Shannon entropy),and CRS-R improvement values across various etiologies(traumatic,hypoxic-ischemic,cerebrovascular accident)were compared between the two groups.An optimized XGBoost model was developed using baseline features at T0,brain network response changes at T2,and rehabilitation parameters to identify key features.Model performance was evaluated via ROC curves and AUC.Results:After 8 weeks of intervention,CRS-R scores increased in both groups,with significantly higher scores in the observation group(P<0.05).Key FC values(PCC-mPFC,DMN-SN,S1/M1-SMA)increased after intervention,with the observation group showing higher values than the control group(P<0.05).hdEEG microstate parameters(microstate C coverage,microstate D duration,transition entropy)increased in both groups after 8 weeks of intervention,but they were significantly higher in the observation group(P<0.05).In the observation group,CRS-R improvement was significantly lower in cerebrovascular accident patients than that in traumatic brain injury patients(P<0.05).No significant differences were observed between other etiology pairs(P>0.05).The XGBoost model identified the top 6 SHAP-ranked features:alterations in PCC-mPFC connectivity after intervention,alterations in microstate C coverage after intervention,baseline total CRS-R,mean robot-assisted training engagement,alterations in alpha power after intervention,and etiology-cerebrovascular accident.The model achieved an AUC of 0.928 for prognostic prediction.Conclusion:Robot-assisted rehabilitation training can effectively improve prognosis in DoC patients,primarily by remodeling functional connectivity in key networks(e.g.,DMN,SMN),optimizing brain network topology,enhancing neural oscillations and microstate dynamics,with etiology-dependent effects.The multimodal brain network prognostic model optimized using XGBoost model has high clinical utility.关键词
意识障碍/机器人辅助康复训练/功能连接/微状态参数/预后/XGBoost模型Key words
Disorder of Consciousness/Robot-assisted Rehabilitation Training/Functional Connectivity/Microstate Parameters/Prognosis/XGBoost Model分类
医药卫生引用本文复制引用
甄颐,刘畅,安霞,潘化杰,杨帆..基于康复机器人辅助训练诱导的脑网络响应特征构建意识障碍患者预后预测模型[J].机器人外科学杂志（中英文）,2025,6(11):1902-1908,7.基金项目
国家重点研发计划项目(2020YFC2004303) National Key Research and Development Program(2020YFC2004303) （2020YFC2004303）
