In this dissertation we focus on utilizing computer-aided engineering techniques to improve our understanding of gait in cerebral palsy (CP). CP is the most common motor disability in children and arises from a non-progressive brain injury at or near the time of birth which alters control (i.e., poor coordination and increased muscle co-contraction). Additionally, individuals with CP often develop secondary, progressive impairments like weakness and contracture. Current treatments to improve mobility in CP primarily target secondary impairments but functional outcomes are inconsistent, leaving treatment efficacy at around 50%. To improve treatment efficacy, clinicians need a better understanding of the complex interactions between, and relative effects of, multimodal neuromuscular impairments on gait. However, eliciting interactions between, and relative effects of, neuromuscular impairments on gait is difficult or even impossible to do clinically and experimentally. Thus, the goal of this dissertation was to utilize in silico techniques to improve the understanding of gait in CP. Specifically, we use physics-based (i.e., musculoskeletal) modeling, optimal control (i.e., neuromuscular simulation), and data-driven modeling (i.e., machine learning) to investigate the interactions between, and relative effects of, altered control, muscle weakness, and contracture on gait and predict and understand gait energetics in CP which can be used to improve treatment efficacy.

The effects of altered motor control on gait are poorly understood because altered control persists post-intervention and its relative effects are difficult to discern amidst secondary impairments, like weakness and contracture. Prior studies have investigated the impacts of weakness, contracture, and altered control on gait, but they have yet to be investigated together. Thus, in this dissertation we sought to understand the effects of, and interactions between, neuromuscular impairments during gait by utilizing a musculoskeletal model and neuromuscular simulation framework. We simulated nondisabled (ND) gait and then perturbed each simulation with altered control, weakness, and contracture of varying severities. We found that altered control exacerbated the restrictions imposed by secondary impairments:​ ND gait was less robust to, and required more muscle activation to adapt to, weakness and contracture with altered control when compared to unaltered control (Chapter 3). These findings highlight the inimical effects of altered control on gait and emphasize the advantages of in silico techniques to identify specific impairments, such as altered control, that should take treatment precedence (in silico-informed interventions). However, it is unclear if these conclusions extend to different gait patterns like those in CP.

Abnormal gait patterns are common for individuals with CP;​ the most inimical and common of which is crouch gait. Crouch gait is characterized by excessive knee flexion, which increases knee extensor demand while reducing the knee extensor's ability to extend the knee making it inefficient and disadvantageous. In Chapter 4, we extended our prior computational methods to simulate crouch gait of varying severities. By simulating both crouch and ND gait, and incorporating machine learning (ML), we investigated if the interactions between, and relative effects of, neuromuscular impairments are gait pattern-specific. We determined that the interactions between, and relative effects of, neuromuscular impairments are gait pattern-specific highlighting advantages and disadvantages of walking in crouch. Thus, by combining computational techniques like modeling, simulation, and machine learning we elicited rationale for why individuals may select non-normative gait patterns and emphasized the utility of in silico techniques to parse and identify impairments primarily affecting function in CP which could then be used to inform treatment.

Individuals with CP consume on average 2x the energy of their ND peers while walking;​ the origin of which remains unknown. Elevated energy consumption persists post-intervention making it a primary complaint among patients and objective of research in the CP community. We sought to accurately predict and understand energetics in CP with modeling, simulation, and machine learning to reduce clinical collection burden on patients and caregivers and improve identification of effective treatment methods for reducing energetics in CP. In the final study of this dissertation, we first used our modeling and simulation framework to generate and perturb walking simulations from gait data from the largest database of walking data for individuals with CP. Generated simulations then acted as synthetic data within a machine learning algorithm to complement existing clinical data and attempt to improve predictions of energetics in CP. Using simulations generated for 240 children with cerebral palsy we analyzed the energetic discrepancy—difference between measured and predicted—to identify primary mechanisms elevating energetics in CP (Chapter 5). Synthetic data generated from gait simulations marginally improve prediction accuracy of energetics in CP, but augmented discrepancy models—energetic predictions with the reconstructed discrepancy—improved modeling of CP energetics, identifying kinematics at initial contact and contracture as primary mechanisms elevating walking energy in CP. Utilizing in silico techniques can provide additional synthetic data (i.e., data augmentation) to reduce data collection burdens on patients, caregivers, and clinicians while eliciting additional insight in causal mechanisms affecting gait and function.

This dissertation supports in silico informed interventions by improving our understanding of gait in CP. By utilizing modeling, simulation, and machine learning we examined the interactions between, and effects of, neuromuscular impairments on gait in both ND and CP individuals and how that information could better predict and understand energetics in CP. This work provides a foundation to utilize modeling, simulation, and machine learning to rapidly evaluate causal mechanisms impacting gait, probe and parse complex relationships between neuromuscular impairments, and incorporate synthetic data to better inform machine learning algorithms and clinical decision making. In conclusion, the work we have completed over the last 4 years highlights the benefits of in silico techniques to understand gait in CP, seeking to support the creation and implementation of in silico informed interventions for individuals with CP.