Predicting Cervical Spine Compression and Shear in Helicopter Helmeted Conditions Using Artificial Neural Networks

Abstract

Introduction: Military helicopter pilots around the globe experience a high prevalence of neck pain. The requirement for pilots to use night vision goggles (NVGs) has been linked to increases in neck pain and injury prevalence. As a result, next generation helmet designs aim to offset or mitigate NVG-related consequences on cervical spine loading. However, in vivo human-participant experiments are currently required to collect necessary data (e.g., electromyography) to estimate joint contact forces on the cervical spine associated with unique helmet designs. This is costly, and inefficient. Thus, a more time and resource-efficient approach is required. A digital human modelling approach wherein multi-body dynamics (MBD) models, which provide inverse dynamics, are combined with artificial neural networks (ANNs) can provide a surrogate for more costly musculoskeletal joint modeling to predict joint contact forces.

Objective: To develop ANNs to predict cervical spine compression and shear, given inputs available through MBD modelling, with enough sensitivity to differentiate between compression and shear exposures associated with different helicopter helmet designs.

Methods: ANNs with systematically varied inputs and parameters were developed to predict cervical spine compression and shear given head-trunk kinematics and C6-C7 neck joint moments, data readily available from digital human models. ANN development was driven by a previously collected and processed dataset. Motion capture and electromyography data were collected from 26 participants who performed flight-relevant reciprocal head movements about pitch and yaw axes while donning one of four helmet configurations. These data were input into an electromyography-driven musculoskeletal model of the neck to generate time series C6-C7 compression and shear outputs. ANNs were trained to predict the electromyography-driven model compression and shear outputs given only the head-trunk kinematics and C6-C7 moments as inputs.

Results: Rotation-specific (i.e., yaw and pitch) ANNs yielded stronger predictive performance than ANNs that generalized to both pitch and yaw axes of rotation. ANNs for pitch rotations accurately predicted peak and cumulative compression and shear outputs with an absolute error that was lower than absolute differences in joint contact forces between relevant helmet conditions. ANNs for yaw rotations were similarly successful in predicting cumulative C6-C7 compression and shear where absolute error was lower than corresponding differences between relevant helmet conditions. However, they were unable to do so for peak C6-C7 compression and shear.

Conclusions: When combined with biomechanical data readily available from digital human modeling software, use of an ANN surrogate for joint musculoskeletal modeling can permit evaluation of joint contact forces in the cervical spine associated with novel helmet design concepts during upstream design. Improved consideration of joint contact forces during a computer-aided helmet design process will assist in identifying helmet designs that reduce the biomechanical exposures of the cervical spine during helicopter flight.