Video-Based Gait Analysis for Spinal Deformity

Abstract

In this paper, we explore the area of classifying spinal deformities unintrusively using machine learning and RGB cameras. We postulate that any changes to posture due to spinal deformity can induce specific changes in people’s gait. These changes are not limited to the spine’s bending but manifest in the movement of the entire body, including the feet. Thus, spinal deformities such as Kyphosis and Lordosis can be classified much more effectively by observing people’s gait. To test our claim, we present a bidirectional long short-term memory (BiLSTM) based neural network that operates using the key points on the body to classify the deformity. To evaluate the system, we captured a dataset containing 29 people simulating Kyphosis, Lordosis and their normal gait under the supervision of an orthopaedic surgeon using an RGB camera. Results suggest that gait is a better indicator of spinal deformity than spine angle.