Performance of multiple neural networks in predicting lower limb joint moments using wearable sensors.

Joint moment measurements represent an objective biomechemical parameter in joint health assessment. Inverse dynamics based on 3D motion capture data is the current 'gold standard' to estimate joint moments. Recently, machine learning combined with data measured by wearable technologies such electromyography (EMG), inertial measurement units (IMU), and electrogoniometers (GON) has been used to enable fast, easy, and low-cost measurements of joint moments. This study investigates the ability of various deep neural networks to predict lower limb joint moments merely from IMU sensors. The performance of five different deep neural networks (InceptionTimePlus, eXplainable convolutional neural network (XCM), XCMplus, Recurrent neural network (RNNplus), and Time Series Transformer (TSTPlus)) were tested to predict hip, knee, ankle, and subtalar moments using acceleration and gyroscope measurements of four IMU sensors at the trunk, thigh, shank, and foot. Multiple locomotion modes were considered including level-ground walking, treadmill walking, stair ascent, stair descent, ramp ascent, and ramp descent. We show that XCM can accurately predict lower limb joint moments using data of only four IMUs with RMSE of 0.046 ± 0.013 Nm/kg compared to 0.064 ± 0.003 Nm/kg on average for the other architectures. We found that hip, knee, and ankle joint moments predictions had a comparable RMSE with an average of 0.069 Nm/kg, while subtalar joint moments had the lowest RMSE of 0.033 Nm/kg. The real-time feedback that can be derived from the proposed method can be highly valuable for sports scientists and physiotherapists to gain insights into biomechanics, technique, and form to develop personalized training and rehabilitation programs.

Smooth and accurate predictions of joint contact force time-series in gait using over parameterised deep neural networks.

Alterations in joint contact forces (JCFs) are thought to be important mechanisms for the onset and progression of many musculoskeletal and orthopaedic pain disorders. Computational approaches to JCFs assessment represent the only non-invasive means of estimating in-vivo forces; but this cannot be undertaken in free-living environments. Here, we used deep neural networks to train models to predict JCFs, using only joint angles as predictors. Our neural network models were generally able to predict JCFs with errors within published minimal detectable change values. The errors ranged from the lowest value of 0.03 bodyweight (BW) (ankle medial-lateral JCF in walking) to a maximum of 0.65BW (knee VT JCF in running). Interestingly, we also found that over parametrised neural networks by training on longer epochs (>100) resulted in better and smoother waveform predictions. Our methods for predicting JCFs using only joint kinematics hold a lot of promise in allowing clinicians and coaches to continuously monitor tissue loading in free-living environments.

Predicting the Internal Knee Abduction Impulse During Walking Using Deep Learning.

Knee joint moments are commonly calculated to provide an indirect measure of knee joint loads. A shortcoming of inverse dynamics approaches is that the process of collecting and processing human motion data can be time-consuming. This study aimed to benchmark five different deep learning methods in using walking segment kinematics for predicting internal knee abduction impulse during walking. Three-dimensional kinematic and kinetic data used for the present analyses came from a publicly available dataset on walking (participants n = 33). The outcome for prediction was the internal knee abduction impulse over the stance phase. Three-dimensional (3D) angular and linear displacement, velocity, and acceleration of the seven lower body segment's center of mass (COM), relative to a fixed global coordinate system were derived and formed the predictor space (126 time-series predictors). The total number of observations in the dataset was 6,737. The datasets were split into training (75%, n = 5,052) and testing (25%, n = 1685) datasets. Five deep learning models were benchmarked against inverse dynamics in quantifying knee abduction impulse. A baseline 2D convolutional network model achieved a mean absolute percentage error (MAPE) of 10.80%. Transfer learning with InceptionTime was the best performing model, achieving the best MAPE of 8.28%. Encoding the time-series as images then using a 2D convolutional model performed worse than the baseline model with a MAPE of 16.17%. Time-series based deep learning models were superior to an image-based method when predicting knee abduction moment impulse during walking. Future studies looking to develop wearable technologies will benefit from knowing the optimal network architecture, and the benefit of transfer learning for predicting joint moments.

Femoral neck strain prediction during level walking using a combined musculoskeletal and finite element model approach.

Recently, coupled musculoskeletal-finite element modelling approaches have emerged as a way to investigate femoral neck loading during various daily activities. Combining personalised gait data with finite element models will not only allow us to study changes in motion/movement, but also their effects on critical internal structures, such as the femur. However, previous studies have been hampered by the small sample size and the lack of fully personalised data in order to construct the coupled model. Therefore, the aim of this study was to build a pipeline for a fully personalised multiscale (body-organ level) model to investigate the strain levels at the femoral neck during a normal gait cycle. Five postmenopausal women were included in this study. The CT and MRI scans of the lower limb, and gait data were collected for all participants. Muscle forces derived from the body level musculoskeletal models were used as boundary constraints on the finite element femur models. Principal strains were estimated at the femoral neck region during a full gait cycle. Considerable variation was found in the predicted peak strain among individuals with mean peak first principal strain of 0.24% ± 0.11% and mean third principal strain of -0.29% ± 0.24%. For four individuals, two overall peaks of the maximum strains were found to occur when both feet were in contact with the floor, while one individual had one peak at the toe-off phase. Both the joint contact forces and the muscular forces were found to substantially influence the loading at the femoral neck. A higher correlation was found between the predicted peak strains and the gluteus medius (R2 ranged between 0.95 and 0.99) than the hip joint contact forces (R2 ranged between 0.63 and 0.96). Therefore, the current findings suggest that personal variations are substantial, and hence it is important to consider multiple subjects before deriving general conclusions for a target population.

The effect of boundary and loading conditions on patient classification using finite element predicted risk of fracture.

BACKGROUND: Osteoporotic proximal femoral fractures associated to falls are a major health burden in the ageing society. Recently, bone strength estimated by finite element models emerged as a feasible alternative to areal bone mineral density as a predictor of fracture risk. However, previous studies showed that the accuracy of patients' classification under their risk of fracture using finite element strength when simulating posterolateral falls is only marginally better than that of areal bone mineral density. Patients tend to fall in various directions: since the predicted strength is sensitive to the fall direction, a prediction based on certain fall directions might not be fully representative of the physical event. Hence, side fall boundary conditions may not be completely representing the physical event. METHODS: The effect of different side fall boundary and loading conditions on a retrospective cohort of 98 postmenopausal women was evaluated to test models' ability to discriminate fracture and control cases. Three different boundary conditions (Linear, Multi-point constraints and Contact model) were investigated under various anterolateral and posterolateral falls. FINDINGS: The stratification power estimated by the area under the receiver operating characteristic curve was highest for Contact model (0.82), followed by Multi-point constraints and Linear models with 0.80. Both Contact and MPC models predicted high strains in various locations of the proximal femur including the greater trochanter, which has rarely reported previously. INTERPRETATION: A full range of fall directions and less restrictive displacement constraints can improve the finite element strength ability to classify patients under their risk of fracture.

A multiscale model to predict current absolute risk of femoral fracture in a postmenopausal population.

Osteoporotic hip fractures are a major healthcare problem. Fall severity and bone strength are important risk factors of hip fracture. This study aims to obtain a mechanistic explanation for fracture risk in dependence of these risk factors. A novel modelling approach is developed that combines models at different scales to overcome the challenge of a large space-time domain of interest and considers the variability of impact forces between potential falls in a subject. The multiscale model and its component models are verified with respect to numerical approximations made therein, the propagation of measurement uncertainties of model inputs is quantified, and model predictions are validated against experimental and clinical data. The main results are model predicted absolute risk of current fracture (ARF0) that ranged from 1.93 to 81.6% (median 36.1%) for subjects in a retrospective cohort of 98 postmenopausal British women (49 fracture cases and 49 controls); ARF0 was computed up to a precision of 1.92 percentage points (pp) due to numerical approximations made in the model; ARF0 possessed an uncertainty of 4.00 pp due to uncertainties in measuring model inputs; ARF0 classified observed fracture status in the above cohort with AUC = 0.852 (95% CI 0.753-0.918), 77.6% specificity (95% CI 63.4-86.5%) and 81.6% sensitivity (95% CI 68.3-91.1%). These results demonstrate that ARF0 can be computed using the model with sufficient precision to distinguish between subjects and that the novel mechanism of fracture risk determination based on fall dynamics, hip impact and bone strength can be considered validated.

Investigating the mechanical response of paediatric bone under bending and torsion using finite element analysis.

Fractures of bone account 25% of all paediatric injuries (Cooper et al. in J Bone Miner Res 19:1976-1981, 2004. https://doi.org/10.1359/JBMR.040902 ). These can be broadly categorised into accidental or inflicted injuries. The current clinical approach to distinguish between these two is based on the clinician's judgment, which can be subjective. Furthermore, there is a lack of studies on paediatric bone to provide evidence-based information on bone strength, mainly due to the difficulties of obtaining paediatric bone samples. There is a need to investigate the behaviour of children's bones under external loading. Such data will critically enhance our understanding of injury tolerance of paediatric bones under various loading conditions, related to injuries, such as bending and torsional loads. The aim of this study is therefore to investigate the response of paediatric femora under two types of loading conditions, bending and torsion, using a CT-based finite element approach, and to determine a relationship between bone strength and age/body mass of the child. Thirty post-mortem CT scans of children aged between 0 and 3 years old were used in this study. Two different boundary conditions were defined to represent four-point bending and pure torsional loads. The principal strain criterion was used to estimate the failure moment for both loading conditions. The results showed that failure moment of the bone increases with the age and mass of the child. The predicted failure moment for bending, external and internal torsions were 0.8-27.9, 1.0-31.4 and 1.0-30.7 Nm, respectively. To the authors' knowledge, this is the first report on infant bone strength in relation to age/mass using models developed from modern medical images. This technology may in future help advance the design of child, car restrain system, and more accurate computer models of children.