一种基于DeepLabCut算法的便捷式步态分析系统的建立及其在中枢神经系统疾病模型中的应用  

Establishment of a Convenient Gait Tracking and Analyzing System Based on the DeepLabCut Algorithm and Its Application in Central Nervous System Disease Models

作　　者：娄嫣云 贺玉琴 刘幸华[2] 郝嘉欢 余颖[1] 汪明欢[1] 吴莹莹 LOU Yanyun;HE Yuqin;LIU Xnghua;HAO Jiahuan;YU Ying;WANG Minghua;WU Yingying(Department of Neurology,Tongji Medical College,Huazhong University of Science and Technology,Wuban 430030,China;Department of Emergency and Trauma Surgery,Tongji Medical College,Huazhong University of Science and Technology,Wuban 430030,China;Department of Oncology,Tongji Hospital,Tongji Medical College,Huazhong University of Science and Technology,Wuban 430030,China)

机构地区：[1]华中科技大学同济医学院附属同济医院神经内科,武汉430000 [2]华中科技大学同济医学院附属同济医院创伤外科,武汉430000 [3]华中科技大学同济医学院附属同济医院肿瘤科,武汉430000

出　　处：《神经损伤与功能重建》2024年第12期700-705,共6页Neural Injury and Functional Reconstruction

基　　金：国家自然科学基金项目mLVs-Gly mphatic系统在急性脑梗塞后神经血管单元损伤与功能重塑中的作用及机制研究,No.82271355;TREM2介导小胶质细胞活化状态在慢性低灌注脑白质缺血后结构与功能重塑中的作用及机制研究,No.82101404;抑制CCL8-CCR2调控PVM介导脑血管内皮细胞间质转化及动脉硬化的机制,No.82301509。

摘　　要：目的:开发一种基于深度学习技术的便捷式步态跟踪系统用于检测实验鼠步态细节,并初步检测其在野生型小鼠和多种中枢神经系统疾病小鼠模型中的应用。方法:搭建简便的步态走廊,将小鼠放入走廊内自由行走4 min,从腹侧记录小鼠步行视频。从小鼠自由运动的视频中抽取120帧,使用DeepLabCut分析动物的运动,标记36个身体部位用于神经网络训练。应用该系统及网络对1、3、6和18月龄的野生型小鼠、APP/PS1小鼠(6月龄,阿尔茨海默病模型)、社会孤立(social isolation,SI)小鼠(3月龄,焦虑抑郁模型)、双侧颈动脉狭窄(bilateral carotid artery stenosis,BCAS)小鼠(3月龄,慢性脑缺血模型)和手术造模后1、3、7天的脓毒血症相关性脑病(sepsis-associated encephalopathy,SAE)小鼠(2月龄)的步态进行分析。结果:利用DeepLabCut可以在所有动物视频追踪中,展现出很高的准确性。3月龄野生型小鼠相比其他月龄小鼠运动速度最快,步幅提高。APP/PS1小鼠运动速度显著高于同龄对照,并伴有步幅增加和站立时间减少。SI小鼠步幅缩短,左前爪脚趾展开度和脚趾展开角度减小,提示存在脚爪姿势改变。BCAS小鼠在步幅上没有显著改变,但后肢脚趾展开度显著增大,脚趾展开角减小。SAE小鼠在术后1、3天运动速度下降,伴有步幅缩短和站立时间延长;术后7天运动速度低于对照小鼠但无显著差异,后肢脚趾展开度和脚趾展开角度小于对照组。结论:本研究搭建了基于深度学习的、便捷、低成本的步态分析设备,只需少量工作即可标记感兴趣的身体部位,比以往的步态分析方法更节省成本。应用这一设备描述了野生型小鼠各年龄组的步态特征,并证明阿尔茨海默病、焦虑抑郁状态、慢性脑缺血和脓毒血症相关性脑病模型小鼠表现出步态缺陷。Objective:To develop a convenient and low-cost gait tracking system based on deep learning technology for detecting gait details in experimental mice,and to preliminarily test its application in wild-type mice and various central nervous system disease mouse models.Methods:A simple gait corridor was built,and mice were allowed to walk freely inside the corridor for 4 minutes while their walking videos were recorded from the ventral side.From the free movement videos of the mice,120 frames were extracted and analyzed using DeepLabCut to label 36 body parts for neural network training.The system and network were applied to analyze the gait of wild-type mice at ages 1,3,6,and 18 months,APP/PS1 mice(6 months old,Alzheimer's disease model),social isolation(SI)mice(3 months old,anxiety and depression model),bilateral carotid artery stenosis(BCAS)mice(3months old,chronic cerebral ischemia model),and sepsis-associated encephalopathy(SAE)mice at postoperative days 1,3,and 7(2 months old).Results:DeepLabCut demonstrated high accuracy in all animal video tracking.Three-month-old wild-type mice had the fastest movement speed and increased stride length compared to other age groups.APP/PS1 mice showed significantly higher movement speed than age-matched controls,accompanied by increased stride length and decreased standing time.SI mice exhibited shortened stride length,reduced toe spread and toe angle of the left front paw,indicating foot posture changes.BCAS mice showed no significant change in stride length but had significantly increased hind limb toe spread and decreased toe angle.SAE mice showed reduced movement speed with shortened stride length and extended standing time on postoperative days 1and 3.By day 7 post-operation,SAE mice had lower movement speed than control mice but without significant difference,and had smaller hind limb toe spread and toe angle compared to the control group.Conclusion:This study established a convenient,low-cost gait analysis device based on deep learning,requiring minimal effort to label

关 键 词：步态分析 DeepLabCut 阿尔茨海默病 抑郁 慢性脑缺血 脓毒血症相关性脑病 

分 类 号：R741[医药卫生—神经病学与精神病学] R741.02[医药卫生—临床医学] R742R743R749