Establishment of a novel deep learning-based method for gait analysis after peripheral nerve injury in pigs

Peripheral nerve injury with deficits has poor functional prognosis, making motor function assessment during nerve regeneration crucial. Recently, pigs have been used as research animals for peripheral nerve regeneration from a translational perspective. However, the established methods for evaluating motor function remain insufficient, and the development of motor function assessment protocols for evaluating paralysis in medium-to large-sized animals is necessary.

Bilateral video recordings of healthy pigs (n = 4) walking straight ahead were obtained. A common peroneal nerve injury model was established on the left side of the animal (n = 3), and similar videos were recorded at 1 and 3 months post-operatively. Stride length, stance phase duration, swing phase duration, gait cycle duration, maximum heel height, joint angles, and foot velocity during one gait cycle were evaluated before and after surgery. Deep learning for posture estimation was employed to analyze joint angles and foot velocities.

The post-operative stance phase duration on the intact side was prolonged and the maximum heel height on the impaired side was significantly higher after surgery. The accuracy of the posture estimated using deep learning was comparable to that estimated using manual human labeling. The ankle angle on the impaired side increased post-operatively and significant changes in foot velocity were observed at the end of the swing phase. Changes in walking patterns of the animals caused by nerve regeneration were also captured.

Deep learning-based gait analysis enabled the quantitative and objective identification of the characteristics of common peroneal nerve palsy and compensatory movements during one gait cycle. This analytical method is a potentially useful technique for studying recovery from paralysis associated with nerve regeneration.