|dc.description.abstract||Cerebral palsy (CP) is the most common motor disorder in childhood, with an incidence rate of approximately 2 in 1000 newborns. Although starting from a non‐progressive lesion occurring in the fetal or infant brain, children with CP present with a wide range of progressive primary and secondary impairments. These include increased muscle tone, contractures, muscular deficits and bony deformities, which if untreated may further lead to progressive loss of locomotor function. Due to CP’s multifaceted nature and the patient-specificity of the symptoms, the management of CP is quite complex and standardised treatment plans that suit all patients’ needs cannot be defined. Currently, several non-invasive (e.g. casting or strengthening programs) and invasive (e.g. single-event multilevel surgeries) procedures are performed in the attempt to restore typical muscle and motor function. Nonetheless, roughly 25% of the patients are dissatisfied with the treatment outcomes, which often require a second intervention. This may be due to a poor treatment planning, which is primarily based on information gathered via passive muscle tests and gait analysis assessments. Such tests completely disregard the internal biomechanics, i.e. muscle activations, muscle forces and joint contact forces (JCFs).
Although difficult to measure in vivo, the internal biomechanics may be estimated in silico by employing neuromusculoskeletal (NMSK) models, i.e. virtual digital representations of the human musculoskeletal system. By providing insights on the mechanisms behind the pathology, NMSK models have the potential to improve the management of CP. Moreover, different surgical scenarios may be tested on virtual models prior to entering the operating theatre, allowing for the identification of the most effective and personalised treatment for each patient. Nonetheless, current NMSK models do have limitations. For instance, generic musculoskeletal anatomies, e.g. gait2392 OpenSim model, are commonly employed and scaled with motion capture data to match each participant’s dimensions. However, generic anatomies are based on a limited set of healthy elder adult data. Paediatric bones, presenting with more pronounced torsions or with deformities, may have large deviations from those of an adult; even after linear scaling. Inaccurate bony geometries are associated to mislocation of the joint centres, which may affect external joint angles and joint moment estimates. Moreover, generic muscle attachments on ill-scaled bones may be inaccurately positioned, thereby affecting muscle kinematics and muscle function. In addition, muscle-tendon units (MTUs) are typically represented as Hill-type actuators, whose behaviour is highly dependent on optimal fibre length, tendon slack length and maximal isometric force values. Most commonly, these parameters are linearly scaled along with bones, although this approach has no physiological grounds. As result, MTUs may operate outside the range of physiologically plausible values, and not representing correct muscle function. Furthermore, CP-related MTU abnormalities are often disregarded. Finally, unconstrained static optimisation methods (e.g., static optimisation that minimises muscle activations squared) are typically employed to determine the set of muscle activations to generate the experimental joint moments. However, these methods favour estimation of muscle endurance, minimal muscle activation, the latter also minimising muscle co-contractions, and may not account for muscle dynamics by assuming tendons to be rigid. Static optimisation also estimates similar activation patterns between subjects, and even within subjects when different control is required, which has been proven otherwise. Moreover, abnormal neural solutions, common in observed CP, cannot be generated.
Previous work has focussed on the personalisation of NMSK models to better represent paediatric populations with CP. However, personalisation was mostly introduced using pre-determined factors to scale MTU parameters or based on available literature data. Moreover, only a few features were personalised at once. Therefore, the overarching aim of my thesis was to develop personalised NMSK models of healthy paediatric populations and children with CP, with increasing level of subject-specificity, and to quantify the effect of each personalised feature on the endpoint variables, i.e. muscle excitation patterns and forces, and JCFs estimates.
The first study investigated whether the personalisation of MTU parameters and muscle activation patterns enabled the production of more physiologically plausible internal biomechanics. Two 13 years old identical twin brothers, one typically developing (TD) and one with unilateral spastic CP, were enrolled in the study. For both children, four different NMSK models with incremental level of subject-specificity were generated. The first two models (unCalSO and unCalEMGa) shared the same musculoskeletal anatomy, which was linearly scaled from a simplified gait2392 generic model and featured morphometrically optimised optimal fibre length (OFL) and tendon slack length (TSL) values. Static optimisation in CEINMS was employed to synthesise muscle activation patterns in unCalSO, while unCalEMGa used an electromyography (EMG)-assisted approach. A further two models (CalEMGa and CalEMGaMRI) subsequently calibrated the MTU parameters (±5% of their original value) in CEINMS using experimental EMG data. For the CP child, the initial OFL of selected muscles was decreased by 0.7 before calibration, while the TSL was bound to increase. These alterations were implemented to respectively account for overstretched sarcomeres and longer TSL observed in CP muscles. The last model (CalEMGaMRI) further built on CalEMGa and featured personalised maximal isometric force values scaled with muscle volumes segmented on magnetic resonance imaging (MRI) scans. The use of an EMG-assisted approach had a greater effect on the ability of the models to track experimental data compared to the calibration of MTU parameters. Nonetheless, when OFL and TSL were not calibrated, knee JCFs estimates did not appear physiologically plausible. The results of this study were included in a full paper submitted as Davico G., Pizzolato C, Lloyd D.G., Obst S.P., Walsh H.P.J., Carty C.P. Increasing level of neuromusculoskeletal model personalisation to investigate joint contact forces in cerebral palsy: a twin case study. Clinical Biomechanics.
The second study examined the best methods to accurately reconstruct paediatric lower limb bones for use in NMSK modelling. Medical imaging and motion capture (MOCAP) data from 18 TD children collected in the past five years at Queensland Children’s Hospital were used in the study. Ten different combinations of morphing and mesh fitting techniques to reconstruct pelvis, femurs and tibiofibular bones were developed and tested in the open-source software Musculoskeletal Atlas Project (MAP) Client. To determine the minimum required amount of data to achieve acceptable reconstructions, different levels of medical image data incompleteness were provided. The resulting bone reconstructions were compared to the corresponding MRI segmentations using three metrics of similarity: Jaccard index, root mean squared surface-to-surface distance error, and Hausdorff distance. In addition, for each reconstructed pelvis, hip joint centres (HJCs) locations and HJC distance were extracted and compared to the corresponding MRI measurements. The HJC distance was also compared to a clinical MOCAP based measurement, i.e. Harrington regression equation. Our results suggested that non-linear scaling methods should not be used to reconstruct the lower limb bones of children smaller than 145 cm, which would be abnormally shaped. Secondly, the use of medical imaging data, even if incomplete, should be preferred to generate highly accurate bony geometries. Moreover, in small children, HJC-distance may largely differ between MOCAP and reconstruction-based calculations. All research findings from this study were detailed in the invited paper: Davico G., Pizzolato C., Killen B.A., Barzan M., Suwarganda E., Lloyd D.G., Carty C.P. Best methods and data to reconstruct paediatric lower limb bones for musculoskeletal modelling. Biomechanics and Modeling in Mechanobiology, 2019. doi:10.1007/s10237-019-01245-y.
The third study had two main aims. First, was to develop a highly personalised NMSK paediatric model, and second, to determine the individual effects of personalised anatomies, muscle activation patterns and MTU parameters on muscle excitation patterns and forces, and JCFs estimates. Six different NMSK models with incremental levels of subject-specificity were generated for each of the six children (3 TD, 3 with CP) enrolled in the study. For the first time a model generated via the MAP Client was developed and tested. This included personalised bony geometries, and physiologically and physically plausible MTU kinematics (i.e. MTU lengths and moment arms). In addition to the MAP generated anatomies and generically scaled anatomies (simplified gait2392 OpenSim model) were employed for biomechanical simulations of gait. Following the steps of study one, both anatomies were progressively personalised by (1) calibrating MTU parameters and (2) replacing static optimisation methods that minimised muscle activations squared with EMG-assisted approaches to synthesise muscle activations. The calibrated EMG-assisted MAP generated model produced the most physiologically plausible estimates, as (i) it well tracked both external moments and muscle excitations (i.e. EMG linear envelopes), (ii) featured subject-specific bones and (iii) estimated non-zero loading in swing phase. Among all, the use of EMG-assisted methods and personalised musculoskeletal anatomies appeared to have the greatest impact on the endpoint estimates. Nonetheless, the neural solution substantially affected lateral JCF profiles. The paper describing these results will be submitted as Davico G., Killen B.A., Carty C.P., Lloyd D.G., Devaprakash D., Pizzolato C. Developing the new generation of personalised neuromusculoskeletal models to investigate cerebral palsy. IEEE Transactions on Biomedical Engineering.
In conclusion, this thesis rigorously assessed what the effects of personalisation are on the endpoint estimates of a NMSK model and provided guidelines to develop more physiologically plausible paediatric musculoskeletal anatomies and models. Particularly, the studies highlighted the weakness and strengths of common clinical measurements and associated methods that may be used to improve the personalisation of NMSK models. The future use of personalised NMSK modelling simulations has the potential to provide knowledge of the internal biomechanics and substantial benefit to the CP paediatric population, by better informing clinical management and by enabling the development of personalised treatments for each patient.||