基于深度学习的多关节同步跟踪，客观量化大鼠后肢运动运动学
April 3rd, 2026
该方案利用无标记跑步机测试和深度学习驱动的多关节轨迹自动标记，建立了大鼠后肢运动学评估流程，实现可重复的运动量化。
我们开发了一台由人工智能驱动的跑步机，用于追踪老鼠的运动，帮助我们精确评估脊髓损伤的恢复情况。与传统足迹方法导致复杂的关节经济不同，我们的模拟AI系统直接追踪多关节的运动。首先，打开工业电脑并启动视频采集与分析软件。
将相机镜头垂直于跑步机纵轴，获得严格横向视角，保持距离跑步机腰带平面15厘米的水平工作距离。接下来，用电子秤记录每只老鼠的体重。只包含体重相匹配的大鼠，以减少体型影响。
戴上手套，轻轻抓住老鼠。将弹性胸带绕在前腋下，并连接到可调节滑轨上。然后调整绑带，使得插入一根手指时不会限制呼吸。
在触摸控制屏幕上，将速度设置为150毫米每秒，倾斜度为零度。进行10分钟的适应训练，帮助大鼠适应体重支持水平，同时监测其压力和疲劳迹象，如不愿移动和长时间排便。当老鼠保持连续且均匀的步伐至少60秒且无拖爪且尾巴自然垂挂时，确认成功适应。
排除那些在最大适应期后仍未达到这些标准的动物。在触摸屏上输入正式实验的参数。依次输入所需的速度和坡度，然后选择跑步机皮带的方向。
稳定后，选择开始录制开始数据采集。持续捕捉至少五个完整的步态周期。每次试验结束时，将速度降至零毫米每秒。
解开胸带，把老鼠放回对应的笼子里。提取约10秒的视频片段，包含至少10个稳定步态周期的目标步态。登记大鼠信息，包括识别编号、组别和实验条件。
将MOV文件导入分析软件。将每个步态周期从零归一化到100%，以标准化周期长度。生成代表性的图像来展示步态动态。
将每个关节随时间变化的空间位置坐标导出为csv文件以便进一步分析。完成动物建模后，植入电生理记录装置。为了记录脑部信号，可以在颅骨表面、硬膜外空间或大脑皮层放置电极以记录脑部信号。
对于脊髓记录，将记录电极插入椎间孔硬膜外空间。在肌电图记录中，将双极银线埋入目标肌肉中以记录肌肉电活动。着床后应让动物恢复五到七天。
每天检查伤口和步态，确保没有感染或疼痛迹象，然后再进行跑步机检测。在同步实验前准备好设备和动物。然后将电生理数据采集与运动视频同步，确保两者共享相同的时间戳。
分析过程中将神经信号与相应的视频帧对齐，以可视化不同步态阶段的电生理模式。脊髓损伤大鼠表现出显著的摆动丧失和髂动脉位移曲线不规则。脊髓损伤大鼠的关节角度波动较健康大鼠增加。
关节活动范围热图显示，脊髓损伤大鼠的颜色鳞度明显下降，而轨迹活动图则显示每个关节的活动范围显著减少，运动连续性受损。对于脊髓损伤大鼠，点云图显示扩散增加且质心向左移动。高度瀑布图显示了各组沿x轴的双峰分布，运动中间有一般的差值下降。
对于纵轴，运动开始时会观察到差值下降。整体差异分布混乱，表明脊髓损伤大鼠的运动阶段异常，与健康大鼠相比。脊髓损伤大鼠的脚趾垂直高度峰值分布向右移动。
速度范围图显示，每个帧的综合速度范围在受伤组中被缩小。此外，脊髓损伤大鼠的相位面面积、估计的峰值推进力指数和运动平滑度指标也有所减少。我们的系统使研究人员能够实时精确测量多关节轨迹、力分布和运动平滑度。
一个重要的考虑是彻底让老鼠适应跑步机系统，以确保步态稳定且能分析。我们可以将系统与肌电图或脑电图同步，解码背后机制的信号。
View the full transcript and gain access to thousands of scientific videos
Overview
This article presents a markerless, treadmill-based gait analysis system for rodents that leverages custom deep learning algorithms to enable real-time, multidimensional tracking of lower-limb joint kinematics. The system provides objective, high-throughput quantification of gait parameters under various experimental conditions, and is validated using spinal cord injury (SCI) models to demonstrate its sensitivity and utility in neuromuscular research.
Key Study Components
Area of Science
- Neuroscience
- Kinesiology
- Behavioral analysis
- Biomedical engineering
Background
- Accurate assessment of rodent gait is critical for studying neuromuscular and neurological disorders.
- Traditional footprint-based methods are limited in capturing complex, multi-joint kinematics.
- Objective, high-dimensional gait analysis tools are needed to reduce bias and improve data richness.
- Integration of neural signal recording with behavioral analysis can elucidate central-peripheral control mechanisms.
Purpose of Study
- To develop a markerless, AI-driven treadmill system for real-time, multi-joint gait analysis in rodents.
- To enable objective quantification of gait parameters under variable speed, incline, and weight support.
- To validate the system's sensitivity in distinguishing healthy from spinal cord injured rodents.
Methods Used
- Setup of a programmable treadmill with lateral video capture for markerless joint tracking.
- Application of deep learning algorithms for automatic extraction of joint trajectories and gait parameters.
- Experimental modulation of treadmill speed (0–300 mm/s), incline (±30°), and weight support (0–500 g).
- Integration of electrophysiological recordings (EEG, EMG, spinal cord signals) synchronized with gait video.
- Analysis of gait cycles, joint kinematics, force distribution, and movement smoothness.
Main Results
- The system enables real-time, high-throughput extraction of multi-joint trajectories and gait cycle parameters.
- SCI rats exhibited significant loss of swing, irregular iliac displacement, increased joint angle fluctuations, and reduced range of motion compared to healthy controls.
- Heat maps and trajectory analyses revealed impaired movement continuity and altered force distribution in SCI rats.
- Synchronization with neural recordings allowed visualization of electrophysiological patterns across gait phases.
Conclusions
- The markerless treadmill system provides objective, multidimensional gait analysis, overcoming limitations of traditional methods.
- It is sensitive to neuromuscular deficits and suitable for disease discrimination and grading.
- The platform supports integration with neurophysiological recordings, enabling advanced studies of motor control and neuromodulation.
Frequently Asked Questions
Executive Industry Relevance
Objective, high-throughput quantification of rodent hindlimb kinematics is critical for translational neuroscience and neuromuscular drug discovery. This deep learning-based, markerless gait analysis system enables multidimensional, bias-free assessment of locomotor function, supporting robust target validation and mechanistic de-risking in preclinical models. Its integration of synchronized neural and biomechanical data advances predictive confidence at key discovery and preclinical inflection points.
Strategic Applications in Biopharma R&D
Early Discovery & Target Validation
- Enables rigorous interrogation of neuromuscular function in disease-relevant rodent models.
- Supports functional target validation by quantifying joint trajectories and force outputs.
- Facilitates mechanistic de-risking through objective, multidimensional behavioral readouts.
- Improves predictive confidence for early-stage portfolio triage in neurotherapeutic pipelines.
Screening & Assay Development
- Provides validated, reproducible kinematic outputs for downstream screening workflows.
- Standardizes gait cycle and joint trajectory measurements for assay comparability.
- Enables high-throughput, automated data collection for scalable compound evaluation.
- Reduces subjective bias and low-dimensionality limitations of traditional footprint assays.
Translational & Preclinical Research
- Aligns preclinical functional endpoints with disease-relevant neuromuscular phenotypes.
- Supports continuity from discovery through preclinical validation with synchronized neural and biomechanical data.
- Enables risk-adjusted advancement decisions based on sensitive, quantitative locomotor metrics.
- Facilitates biomarker development for neuromodulatory and neuroregenerative strategies.
Pipeline & Workflow Integration
This system bridges early discovery, lead identification, and preclinical validation by providing objective, quantitative locomotor phenotyping in rodent models.
- Discovery Biology: Supports hypothesis testing and pathway clarification in neuromuscular and neurodegenerative disease models.
- Screening: Delivers reproducible, high-dimensional kinematic outputs for assay readiness and compound triage.
- Analytics: Enables export of joint position, force, and movement smoothness data for robust statistical comparison.
- Translational Research: Integrates neural and biomechanical signals to align preclinical endpoints with clinical relevance.
- Enterprise Reuse: Functions as a reusable platform for diverse disease models and mechanistic studies.
Operational & Enterprise Impact
- Scientific Value: Increases predictive confidence and reduces mechanistic ambiguity in neuromuscular target validation.
- Operational Value: Standardizes and automates gait analysis for reproducibility and scalability.
- Strategic Value: Enables informed go/no-go decisions and capital-efficient portfolio management.
- Portfolio Impact: Supports risk-adjusted prioritization and advancement of neurotherapeutic candidates.
Implementation Considerations
- Requires expertise in rodent handling, deep learning-based video analysis, and neurophysiological data integration.
- Needs programmable treadmill, high-speed video capture, and synchronized electrophysiological recording infrastructure.
- Demands cross-team standardization of acclimation, data acquisition, and analysis protocols.
- Adaptable across central and peripheral nerve injury, neurodegeneration, and musculoskeletal disease models.
- Stable gait pattern acquisition depends on thorough animal acclimation and stress minimization.
