Automated Gait Analysis Through Hues and Areas (AGATHA): A Method to Characterize the Spatiotemporal Pattern of Rat Gait

Abstract

While rodent gait analysis can quantify the behavioral consequences of disease, significant methodological differences exist between analysis platforms and little validation has been performed to understand or mitigate these sources of variance. By providing the algorithms used to quantify gait, open-source gait analysis software can be validated and used to explore methodological differences. Our group is introducing, for the first time, a fully-automated, open-source method for the characterization of rodent spatiotemporal gait patterns, termed Automated Gait Analysis Through Hues and Areas (AGATHA). This study describes how AGATHA identifies gait events, validates AGATHA relative to manual digitization methods, and utilizes AGATHA to detect gait compensations in orthopaedic and spinal cord injury models. To validate AGATHA against manual digitization, results from videos of rodent gait, recorded at 1000 frames per second (fps), were compared. To assess one common source of variance (the effects of video frame rate), these 1000 fps videos were re-sampled to mimic several lower fps and compared again. While spatial variables were indistinguishable between AGATHA and manual digitization, low video frame rates resulted in temporal errors for both methods. At frame rates over 125 fps, AGATHA achieved a comparable accuracy and precision to manual digitization for all gait variables. Moreover, AGATHA detected unique gait changes in each injury model. These data demonstrate AGATHA is an accurate and precise platform for the analysis of rodent spatiotemporal gait patterns.

Keywords: Automated gait analysis; Behavioral analysis; Meniscus; Osteoarthritis; Rodent; Spatiotemporal gait patterns; Spinal cord contusion; Spinal cord injury.

Figure 1. Deriving Temporal Variables using Automated Gait Analysis Through Hues and Areas (AGATHA)

a) Original video frame. b) The transformed image frame in which the animal is isolated and shows as white, while the background is negated and shows as black. c) Graphical representation of the row of pixels identified as the floor in the transformed image. Within this row of pixels, portions of the animal in contact with the floor show up as white. d) Consecutive pixel rows representing the floor are stacked and the paw print objects can be represented 2-dimensionally for calculation of foot-strike and toe-off. Red shapes identify foot-strike and blue shapes represent toe-off. Triangles are used for fore-limbs and circles are used for hind-limbs.

Figure 2. Deriving Spatial Variables using AGATHA

a) Original video frame. b) AGATHA’s isolated paw image from the ventral view. c) AGATHA’s filtered representation of the paw print. d) A sequence of hind paw prints from a full trial.

Figure 3. Validating AGATHA Against Manual Digitization

Paired scatterplots of raw data from the 1000 fps root videos of the naïve Lewis rat cohort. AGATHA reproduces data digitized manually for symmetry, duty factor imbalance, and stride length, but was slightly different for velocity, hind limb duty factor, and step width. Still, differences between manual digitization and AGATHA were much smaller than inter-trial variability.

Figure 4. Effects of Video Frame Rate on Gait Variable Calculation for AGATHA and Manual Digitization

Raw data scatter plots of each gait variable for each digitization method at all video frame rates. Low video frame rate appears to affect AGATHA's accuracy more than manual digitization, especially for temporal variables. Spatial variables seem largely unaffected by video frame rate.

Figure 5. Effects of Video Frame Rate on Estimation of Foot-Strike and Toe-off

The temporal residual of foot-strike and toe-off are summarized in panel a) and b), respectively. The foot-strike and toe-off residuals are in reference to the matching gait event from each video’s respective 1000 fps root video. To directly compare each video frame rate, the residuals are also represented in raw time (seconds) instead of frames.

Figure 6. AGATHA's Results for the Orthopaedic Model

a) Overall, MCLT sham animals walked faster than naive animals. b) A graphical representation of spatial variable residuals by surgical group. Normalized to the left hind foot placement, each ellipse center is located at its group's average stride length and step width. The major and minor axes of each ellipse correspond with one standard deviation of the respective residual (in this case, horizontal axes represent stride length residual deviations while vertical axes represent step width residual deviations). c) A Hildebrand graph summarizing the residualized temporal differences between groups during a step cycle. The left side of each colored box indicates foot-strike, the horizontal length of each colored box indicates duration of stance, and the right side of each colored box indicates toe-off. The white space in between represents the duration and temporal location during the step cycle of the swing phase for that limb. d)–i) Animal average scatter plots of gait results. *: indicates p<0.05.

Figure 7. AGATHA's Results for the Spinal Cord Injury Models

a) Overall, there were no velocity differences between laminectomy animals, cervical contusion animals, and their respective naive group, but the C2-hemisection animals walked significantly slower than their respective naive group. b) A graphical representation of spatial variable residuals by surgical group for each spinal cord injury model. Normalized to the left hind foot placement, each ellipse center is located at its group's average stride length and step width. The major and minor axes of each ellipse correspond with one standard deviation of the respective residual (in this case, horizontal axes represent stride length residual deviations while vertical axes represent step width residual deviations). For clarity, fore paw data are not presented in the C2 hemisection model. For paw data could not be obtained for cervical contusion due to the lack of foot-strike in the injured paw of this model. c) Hildebrand graphs summarizing the residualized temporal differences between spinal cord injury groups during a step cycle. The left side of each colored box indicates foot-strike, the horizontal length of each colored box indicates duration of stance, and the right side of each colored box indicates toe-off. The white space in between represents the duration and temporal location during the step cycle of the swing phase for that limb. d)–i) Animal average scatter plots of gait results. *: indicates p<0.05.