Markerless analysis of hindlimb kinematics in spinal cord-injured mice through deep learning

Highlights

• A deep learning based markerless 2D kinematic analysis in SCI mice was developed.

• Markerless approach reduced skin error from soft tissue movement on the knee joint.

• 30 extracted kinematic parameters distinguished intact and SCI locomotion.

Abstract

Rodent models are commonly used to understand the underlying mechanisms of spinal cord injury (SCI). Kinematic analysis, an important technique to measure dysfunction of locomotion after SCI, is generally based on the capture of physical markers placed on bony landmarks. However, marker-based studies face significant experimental hurdles such as labor-intensive manual joint tracking, alteration of natural gait by markers, and skin error from soft tissue movement on the knee joint. Although the pose estimation strategy using deep neural networks can solve some of these issues, it remains unclear whether this method is adaptive to SCI mice with abnormal gait. In the present study, we developed a deep learning based markerless method of 2D kinematic analysis to automatically track joint positions. We found that a relatively small number (< 200) of manually labeled video frames was sufficient to train the network to extract trajectories. The mean test error was on average 3.43 pixels in intact mice and 3.95 pixels in SCI mice, which is comparable to the manual tracking error (3.15 pixels, less than 1 mm). Thereafter, we extracted 30 gait kinematic parameters and found that certain parameters such as step height and maximal hip joint amplitude distinguished intact and SCI locomotion.

Introduction

Potential treatments for spinal cord injury commonly occur in rodent models. In recent decades, various interventions have been developed, including cell replacement therapy (Kajikawa et al., 2020; Kamata et al., 2021; Kojima et al., 2019; Nori et al., 2011; Okubo et al., 2018; Tsuji et al., 2019), a combination of multiple treatments (Ito et al., 2018; Maier et al., 2009), and genetically modified animals (Takeoka et al., 2014), and consequently, gait patterns after interventions have become more diverse and complex. Thus, these rodent studies require reliable, consistent, and objective behavioral assessments of gait function to accurately determine the effects of these treatments.

Kinematic analyses using reflective markers attached on animals’ skin have been used for evaluating locomotor deficits; however, this marker-based approach has some issues to consider in both recording and analysis. First, the attached markers are potentially distracting to animals (Pérez-Escudero et al., 2014). Moreover, it is empirically known that SCI mice tend to have low motivation for stable walking, causing them to fail to complete their walking task. Second, reflective markers placed on distal joints such as the toe or the metatarsophalangeal (MTP) joint may affect the natural gait of mice. Notably, this effect is even more pronounced in SCI mice that have difficulty locomotion. Finally, the marker positions are often manually digitized afterward, which is a labor-intensive and time-consuming procedure that constrains the number of groups, mice, and evaluations that can be included within the experimental plan.

Besides the recording and analysis issues described above, reflective markers placed on the skin surface relative to the bone are a primary factor limiting the resolution of detailed joint movements. Soft tissue movement around the knee joint is the principal source of error during locomotion in rodents (Bauman and Chang, 2010). Previous kinematic analyses have either neglected to account for skin error or utilized a triangulation algorithm for estimating the knee joint position (Filipe et al., 2006).

Developing markerless-based analyses would avoid the drawbacks of reflective marker-based analyses (Sheets et al., 2013). Recently, machine learning-based motion capture has become common. In the present study, we demonstrate that markerless kinematic analysis using deep learning can quantitatively assess locomotion in intact and SCI mice. Using this method, we present a variety of gait parameters and show that these parameters change robustly after SCI.