Automatic Detection of Musculoskeletal Deformities Using Gait Biomechanics and Deep Learning: A Comprehensive Review

Abstract

Human gait can be analyzed as an essential point of view of musculoskeletal and nervous systems functionality. Gait abnormalities are the initial measurable factors of pathology including Cerebral Palsy (CP) and a spinal cord injury. The gait analysis has historically moved forward, replacing the subjective visual observation of the gait to an objectively quantitative lab test comparing lab standards of lab-based kinematic and kinetic measurements. But, the recent intersection of wearable sensor technology and Deep Learning (DL), has triggered a paradigm shift, where automatic, non-invasive and continuous detection of musculoskeletal deformities is enabled. The current paper is a thorough review of the literature aimed at analyzing the progression of the models of traditional kinematic algorithms and force-plate methodologies to the advanced ones based on Convolutional Neural Networks (CNNs) and neuro-fuzzy systems. Combining the information of all the studies used through the overground and treadmill locomotion, we assess the effectiveness of several biomechanical parameters, including the timing of the heel-strike and the toe-off, in the diagnosis of clinical conditions. The review focuses on the evolution of threshold-based heuristic algorithms to data-driven deep learning models that can extract latent features at the inertial measurement units (IMUs), and concludes that the combination of deep learning and biomechanical profiling can be the future of accuracy in orthopedic and rehabilitation diagnostic procedures.