为实现脑卒中患者手功能康复情况的自动、精准定量评估,本文提出一种基于手部骨骼的手势识别与功能评估方法。首先,利用MediaPipe框架提取手部关键点并连接形成手骨骼模型,将传统的RGB视频数据集转化为手骨骼数据集。然后,通过C3D模型...为实现脑卒中患者手功能康复情况的自动、精准定量评估,本文提出一种基于手部骨骼的手势识别与功能评估方法。首先,利用MediaPipe框架提取手部关键点并连接形成手骨骼模型,将传统的RGB视频数据集转化为手骨骼数据集。然后,通过C3D模型进行训练,实现手功能动作的识别。最后,在正确识别的基础上进一步评估,采用动态时间规整(dynamic time warping,DTW)算法,在实现时序对齐的同时引入空间对齐机制,通过计算患者健侧手与患侧手完成同一动作的DTW距离,量化动作执行的相似度,为每个动作找到最佳阈值作为定量评估的标准。实验结果表明,用骨骼数据代替传统视频数据,使手势识别的准确率提升至99.01%,缩短了训练时间,并结合DTW算法,实现了手功能康复情况的自动评估。展开更多
目的探讨上肢动作研究测试(action research arm test,ARAT)量表应用于亚急性期缺血性卒中患者上肢及手功能障碍评定的效度、信度和敏感性。方法本研究招募2020年1月—2022年5月于首都医科大学附属北京天坛医院等17家医院住院的300例缺...目的探讨上肢动作研究测试(action research arm test,ARAT)量表应用于亚急性期缺血性卒中患者上肢及手功能障碍评定的效度、信度和敏感性。方法本研究招募2020年1月—2022年5月于首都医科大学附属北京天坛医院等17家医院住院的300例缺血性卒中患者,并由两名评定者分别对受试者进行ARAT量表评估。同时以Fugl-Meyer运动功能评定量表上肢部分(Fugl-Meyer assessment-upper extremity,FMA-UE)作为效标。分别采用Spearman相关分析及验证性因子分析评价ARAT量表的效标效度及结构效度;采用Cronbach’sα系数及组内相关系数评价ARAT量表的内部信度、敏感性及外部信度。结果以FMA-UE为效标,ARAT量表总分与其总分呈正相关(r=0.946,P<0.001)。结构效度检验中,验证性因子分析结果表明:因子间相关系数范围为0.79~0.92,模型整体拟合指标为χ^(2)/df=7.011,P=0.001,比较拟合指数=0.906,增值拟合指数=0.907,规范拟合指数=0.895,近似误差均方根=0.055。各因子的组合信度值在0.964~0.983,各因子的平均变异抽取量在0.940~0.952。内部一致性检验中,评分等级的Cronbach’sα系数为0.987,敏感性分析中,Cronbach’sα系数在0.986~0.987。外部一致性检验中,两名评定者间的组内相关系数最高为0.999,95%CI集中于0.961~0.999。结论ARAT量表具有良好的效标效度、结构效度、信度和敏感性,适用于亚急性期缺血性卒中患者偏瘫侧上肢及手功能的评估。展开更多
Objective:To reveal the neural network of active and passive hand movements. Method:Seven healthy aged people were checked, and acquired functional magnetic resonance imaging data on a 1.5T scanner. Active movement co...Objective:To reveal the neural network of active and passive hand movements. Method:Seven healthy aged people were checked, and acquired functional magnetic resonance imaging data on a 1.5T scanner. Active movement consisted of repetitive grasping and loosening of hand; passive movement involved the same movement performed by examiner. Both types of hand movements were assessed separately. These data were analysed by Statistical Parametric Mapping Microsoft. Result:The main activated brain areas were the contralateral supplemental motor area, primary motor area, primary sensory area and the ipsilateral cerebellum when subjects gripped right hands actively and passively. The supplemental area was less active in passive hand movement than active hand movement. The activated brain areas were mainly within Brodmann area 4 during active hand movement; in the contrast, the voxels triggered by passive movement were mainly within Brodmann areas 3,1,2 areas. Conclusion:The results suggest that the neural networks of passive and active tasks spared some common areas, and the passive movement could be as effective as active movement to facilitate the recovery of limbs motor function in patients with brain damage.展开更多
文摘为实现脑卒中患者手功能康复情况的自动、精准定量评估,本文提出一种基于手部骨骼的手势识别与功能评估方法。首先,利用MediaPipe框架提取手部关键点并连接形成手骨骼模型,将传统的RGB视频数据集转化为手骨骼数据集。然后,通过C3D模型进行训练,实现手功能动作的识别。最后,在正确识别的基础上进一步评估,采用动态时间规整(dynamic time warping,DTW)算法,在实现时序对齐的同时引入空间对齐机制,通过计算患者健侧手与患侧手完成同一动作的DTW距离,量化动作执行的相似度,为每个动作找到最佳阈值作为定量评估的标准。实验结果表明,用骨骼数据代替传统视频数据,使手势识别的准确率提升至99.01%,缩短了训练时间,并结合DTW算法,实现了手功能康复情况的自动评估。
文摘目的探讨上肢动作研究测试(action research arm test,ARAT)量表应用于亚急性期缺血性卒中患者上肢及手功能障碍评定的效度、信度和敏感性。方法本研究招募2020年1月—2022年5月于首都医科大学附属北京天坛医院等17家医院住院的300例缺血性卒中患者,并由两名评定者分别对受试者进行ARAT量表评估。同时以Fugl-Meyer运动功能评定量表上肢部分(Fugl-Meyer assessment-upper extremity,FMA-UE)作为效标。分别采用Spearman相关分析及验证性因子分析评价ARAT量表的效标效度及结构效度;采用Cronbach’sα系数及组内相关系数评价ARAT量表的内部信度、敏感性及外部信度。结果以FMA-UE为效标,ARAT量表总分与其总分呈正相关(r=0.946,P<0.001)。结构效度检验中,验证性因子分析结果表明:因子间相关系数范围为0.79~0.92,模型整体拟合指标为χ^(2)/df=7.011,P=0.001,比较拟合指数=0.906,增值拟合指数=0.907,规范拟合指数=0.895,近似误差均方根=0.055。各因子的组合信度值在0.964~0.983,各因子的平均变异抽取量在0.940~0.952。内部一致性检验中,评分等级的Cronbach’sα系数为0.987,敏感性分析中,Cronbach’sα系数在0.986~0.987。外部一致性检验中,两名评定者间的组内相关系数最高为0.999,95%CI集中于0.961~0.999。结论ARAT量表具有良好的效标效度、结构效度、信度和敏感性,适用于亚急性期缺血性卒中患者偏瘫侧上肢及手功能的评估。
基金supported by the Key Projects of Shanghai Science and Technology on Biomedicine(NO.10DZ1950800)the Major Project of Shanghai Zhabei District Health Bureau (No. 2011ZD01)
文摘Objective:To reveal the neural network of active and passive hand movements. Method:Seven healthy aged people were checked, and acquired functional magnetic resonance imaging data on a 1.5T scanner. Active movement consisted of repetitive grasping and loosening of hand; passive movement involved the same movement performed by examiner. Both types of hand movements were assessed separately. These data were analysed by Statistical Parametric Mapping Microsoft. Result:The main activated brain areas were the contralateral supplemental motor area, primary motor area, primary sensory area and the ipsilateral cerebellum when subjects gripped right hands actively and passively. The supplemental area was less active in passive hand movement than active hand movement. The activated brain areas were mainly within Brodmann area 4 during active hand movement; in the contrast, the voxels triggered by passive movement were mainly within Brodmann areas 3,1,2 areas. Conclusion:The results suggest that the neural networks of passive and active tasks spared some common areas, and the passive movement could be as effective as active movement to facilitate the recovery of limbs motor function in patients with brain damage.
