Strategies to optimise machine learning classification performance when using biomechanical features.

Building prediction models using biomechanical features is challenging because such models may require large sample sizes. However, collecting biomechanical data on large sample sizes is logistically very challenging. This study aims to investigate if modern machine learning algorithms can help overcome the issue of limited sample sizes on developing prediction models. This was a secondary data analysis two biomechanical datasets - a walking dataset on 2295 participants, and a countermovement jump dataset on 31 participants. The input features were the three-dimensional ground reaction forces (GRFs) of the lower limbs. The outcome was the orthopaedic disease category (healthy, calcaneus, ankle, knee, hip) in the walking dataset, and healthy vs people with patellofemoral pain syndrome in the jump dataset. Different algorithms were compared: multinomial/LASSO regression, XGBoost, various deep learning time-series algorithms with augmented data, and with transfer learning. For the outcome of weighted multiclass area under the receiver operating curve (AUC) in the walking dataset, the three models with the best performance were InceptionTime with x12 augmented data (0.810), XGBoost (0.804), and multinomial logistic regression (0.800). For the jump dataset, the top three models with the highest AUC were the LASSO (1.00), InceptionTime with x8 augmentation (0.750), and transfer learning (0.653). Machine-learning based strategies for managing the challenging issue of limited sample size for biomechanical ML-based problems, could benefit the development of alternative prediction models in healthcare, especially when time-series data are involved.

Neuromechanical stabilisation of the centre of mass during running.

BACKGROUND: Stabilisation of the centre of mass (COM) trajectory is thought to be important during running. There is emerging evidence of the importance of leg length and angle regulation during running, which could contribute to stability in the COM trajectory The present study aimed to understand if leg length and angle stabilises the vertical and anterior-posterior (AP) COM displacements, and if the stability alters with running speeds. METHODS: Data for this study came from an open-source treadmill running dataset (n = 28). Leg length (m) was calculated by taking the resultant distance of the two-dimensional sagittal plane leg vector (from pelvis segment to centre of pressure). Leg angle was defined by the angle subtended between the leg vector and the horizontal surface. Leg length and angle were scaled to a standard deviation of one. Uncontrolled manifold analysis (UCM) was used to provide an index of motor abundance (IMA) in the stabilisation of the vertical and AP COM displacement. RESULTS: IMAAP and IMAvertical were largely destabilising and always stabilising, respectively. As speed increased, the peak destabilising effect on IMAAP increased from -0.66(0.18) at 2.5 m/s to -1.12(0.18) at 4.5 m/s, and the peak stabilising effect on IMAvertical increased from 0.69 (0.19) at 2.5 m/s to 1.18 (0.18) at 4.5 m/s. CONCLUSION: Two simple parameters from a simple spring-mass model, leg length and angle, can explain the control behind running. The variability in leg length and angle helped stabilise the vertical COM, whilst maintaining constant running speed may rely more on inter-limb variation to adjust the horizontal COM accelerations.

Automatic Variable Selection Algorithms in Prognostic Factor Research in Neck Pain.

This study aims to compare the variable selection strategies of different machine learning (ML) and statistical algorithms in the prognosis of neck pain (NP) recovery. A total of 3001 participants with NP were included. Three dichotomous outcomes of an improvement in NP, arm pain (AP), and disability at 3 months follow-up were used. Twenty-five variables (twenty-eight parameters) were included as predictors. There were more parameters than variables, as some categorical variables had >2 levels. Eight modelling techniques were compared: stepwise regression based on unadjusted p values (stepP), on adjusted p values (stepPAdj), on Akaike information criterion (stepAIC), best subset regression (BestSubset) least absolute shrinkage and selection operator [LASSO], Minimax concave penalty (MCP), model-based boosting (mboost), and multivariate adaptive regression splines (MuARS). The algorithm that selected the fewest predictors was stepPAdj (number of predictors, p = 4 to 8). MuARS was the algorithm with the second fewest predictors selected (p = 9 to 14). The predictor selected by all algorithms with the largest coefficient magnitude was "having undergone a neuroreflexotherapy intervention" for NP (ß = from 1.987 to 2.296) and AP (ß = from 2.639 to 3.554), and "Imaging findings: spinal stenosis" (ß = from -1.331 to -1.763) for disability. Stepwise regression based on adjusted p-values resulted in the sparsest models, which enhanced clinical interpretability. MuARS appears to provide the optimal balance between model sparsity whilst retaining high predictive performance across outcomes. Different algorithms produced similar performances but resulted in a different number of variables selected. Rather than relying on any single algorithm, confidence in the variable selection may be increased by using multiple algorithms.

A Comprehensive Machine Learning Benchmark Study for Radiomics-Based Survival Analysis of CT Imaging Data in Patients With Hepatic Metastases of CRC.

OBJECTIVES: Optimizing a machine learning (ML) pipeline for radiomics analysis involves numerous choices in data set composition, preprocessing, and model selection. Objective identification of the optimal setup is complicated by correlated features, interdependency structures, and a multitude of available ML algorithms. Therefore, we present a radiomics-based benchmarking framework to optimize a comprehensive ML pipeline for the prediction of overall survival. This study is conducted on an image set of patients with hepatic metastases of colorectal cancer, for which radiomics features of the whole liver and of metastases from computed tomography images were calculated. A mixed model approach was used to find the optimal pipeline configuration and to identify the added prognostic value of radiomics features. MATERIALS AND METHODS: In this study, a large-scale ML benchmark pipeline consisting of preprocessing, feature selection, dimensionality reduction, hyperparameter optimization, and training of different models was developed for radiomics-based survival analysis. Portal-venous computed tomography imaging data from a previous prospective randomized trial evaluating radioembolization of liver metastases of colorectal cancer were quantitatively accessible through a radiomics approach. One thousand two hundred eighteen radiomics features of hepatic metastases and the whole liver were calculated, and 19 clinical parameters (age, sex, laboratory values, and treatment) were available for each patient. Three ML algorithms-a regression model with elastic net regularization (glmnet), a random survival forest (RSF), and a gradient tree-boosting technique (xgboost)-were evaluated for 5 combinations of clinical data, tumor radiomics, and whole-liver features. Hyperparameter optimization and model evaluation were optimized toward the performance metric integrated Brier score via nested cross-validation. To address dependency structures in the benchmark setup, a mixed-model approach was developed to compare ML and data configurations and to identify the best-performing model. RESULTS: Within our radiomics-based benchmark experiment, 60 ML pipeline variations were evaluated on clinical data and radiomics features from 491 patients. Descriptive analysis of the benchmark results showed a preference for RSF-based pipelines, especially for the combination of clinical data with radiomics features. This observation was supported by the quantitative analysis via a linear mixed model approach, computed to differentiate the effect of data sets and pipeline configurations on the resulting performance. This revealed the RSF pipelines to consistently perform similar or better than glmnet and xgboost. Further, for the RSF, there was no significantly better-performing pipeline composition regarding the sort of preprocessing or hyperparameter optimization. CONCLUSIONS: Our study introduces a benchmark framework for radiomics-based survival analysis, aimed at identifying the optimal settings with respect to different radiomics data sources and various ML pipeline variations, including preprocessing techniques and learning algorithms. A suitable analysis tool for the benchmark results is provided via a mixed model approach, which showed for our study on patients with intrahepatic liver metastases, that radiomics features captured the patients' clinical situation in a manner comparable to the provided information solely from clinical parameters. However, we did not observe a relevant additional prognostic value obtained by these radiomics features.

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.

Combining graph neural networks and spatio-temporal disease models to improve the prediction of weekly COVID-19 cases in Germany.

During 2020, the infection rate of COVID-19 has been investigated by many scholars from different research fields. In this context, reliable and interpretable forecasts of disease incidents are a vital tool for policymakers to manage healthcare resources. In this context, several experts have called for the necessity to account for human mobility to explain the spread of COVID-19. Existing approaches often apply standard models of the respective research field, frequently restricting modeling possibilities. For instance, most statistical or epidemiological models cannot directly incorporate unstructured data sources, including relational data that may encode human mobility. In contrast, machine learning approaches may yield better predictions by exploiting these data structures yet lack intuitive interpretability as they are often categorized as black-box models. We propose a combination of both research directions and present a multimodal learning framework that amalgamates statistical regression and machine learning models for predicting local COVID-19 cases in Germany. Results and implications: the novel approach introduced enables the use of a richer collection of data types, including mobility flows and colocation probabilities, and yields the lowest mean squared error scores throughout the observational period in the reported benchmark study. The results corroborate that during most of the observational period more dispersed meeting patterns and a lower percentage of people staying put are associated with higher infection rates. Moreover, the analysis underpins the necessity of including mobility data and showcases the flexibility and interpretability of the proposed approach.

Machine learning versus logistic regression for prognostic modelling in individuals with non-specific neck pain.

PURPOSE: Prognostic models play an important clinical role in the clinical management of neck pain disorders. No study has compared the performance of modern machine learning (ML) techniques, against more traditional regression techniques, when developing prognostic models in individuals with neck pain. METHODS: A total of 3001 participants suffering from neck pain were included into a clinical registry database. Three dichotomous outcomes of a clinically meaningful improvement in neck pain, arm pain, and disability at 3 months follow-up were used. There were 26 predictors included, five numeric and 21 categorical. Seven modelling techniques were used (logistic regression, least absolute shrinkage and selection operator [LASSO], gradient boosting [Xgboost], K nearest neighbours [KNN], support vector machine [SVM], random forest [RF], and artificial neural networks [ANN]). The primary measure of model performance was the area under the receiver operator curve (AUC) of the validation set. RESULTS: The ML algorithm with the greatest AUC for predicting arm pain (AUC = 0.765), neck pain (AUC = 0.726), and disability (AUC = 0.703) was Xgboost. The improvement in classification AUC from stepwise logistic regression to the best performing machine learning algorithms was 0.081, 0.103, and 0.077 for predicting arm pain, neck pain, and disability, respectively. CONCLUSION: The improvement in prediction performance between ML and logistic regression methods in the present study, could be due to the potential greater nonlinearity between baseline predictors and clinical outcome. The benefit of machine learning in prognostic modelling may be dependent on factors like sample size, variable type, and disease investigated.

The mechanical energetics of walking across the adult lifespan.

PURPOSE: Understanding what constitutes normal walking mechanics across the adult lifespan is crucial to the identification and intervention of early decline in walking function. Existing research has assumed a simple linear alteration in peak joint powers between young and older adults. The aim of the present study was to quantify the potential (non)linear relationship between age and the joint power waveforms of the lower limb during walking. METHODS: This was a pooled secondary analysis of the authors' (MT, KD, JJ) and three publicly available datasets, resulting in a dataset of 278 adults between the ages of 19 to 86 years old. Three-dimensional motion capture with synchronised force plate assessment was performed during self-paced walking. Inverse dynamics were used to quantity joint power of the ankle, knee, and hip, which were time-normalized to 100 stride cycle points. Generalized Additive Models for location, scale and shape (GAMLSS) was used to model the effect of cycle points, age, walking speed, stride length, height, and their interaction on the outcome of each joint's power. RESULTS: At both 1m/s and 1.5 m/s, A2 peaked at the age of 60 years old with a value of 3.09 (95% confidence interval [CI] 2.95 to 3.23) W/kg and 3.05 (95%CI 2.94 to 3.16), respectively. For H1, joint power peaked with a value of 0.40 (95%CI 0.31 to 0.49) W/kg at 1m/s, and with a value of 0.78 (95%CI 0.72 to 0.84) W/kg at 1.5m/s, at the age of 20 years old. For H3, joint power peaked with a value of 0.69 (95%CI 0.62 to 0.76) W/kg at 1m/s, and with a value of 1.38 (95%CI 1.32 to 1.44) W/kg at 1.5m/s, at the age of 70 years old. CONCLUSIONS: Findings from this study do not support a simple linear relationship between joint power and ageing. A more in-depth understanding of walking mechanics across the lifespan may provide more opportunities to develop early clinical diagnostic and therapeutic strategies for impaired walking function. We anticipate that the present methodology of pooling data across multiple studies, is a novel and useful research method to understand motor development across the lifespan.

Comparing shallow, deep, and transfer learning in predicting joint moments in running.

Joint moments are commonly calculated in biomechanics research and provide an indirect measure of muscular behaviors and joint loads. However, joint moments cannot be easily quantified clinically or in the field, primarily due to challenges measuring ground reaction forces outside the laboratory. The present study aimed to compare the accuracy of three different machine learning (ML) techniques - functional regression [ MLfregress ], a deep neural network (DNN) built from scratch [ MLDNN ], and transfer learning [ MLTL ], in predicting joint moments during running. Data for this study came from an open-source dataset and two studies on running with and without external loads. Three-dimensional (3D) joint moments of the hip, knee, and ankle, were derived using inverse dynamics. 3D joint angle, velocity, and acceleration of the three joints served as predictors for each of the three ML techniques. Prediction performance was generally the best using MLDNN, and the worse using MLfregress. Absolute predictive performance was the best for sagittal plane moments, which ranged from a RMSE of 0.16 Nm/kg at the ankle using MLDNN, to a RMSE of 0.49Nm/kg at the knee using MLfregress. MLDNN resulted in the greatest improvement in relative prediction performance (relRMSE) by 20% compared to MLfregress for the ankle adduction-abduction moment. DNN with or without transfer learning was superior in predicting joint moments using kinematic inputs compared to functional regression. Synergizing ML with kinematic inputs has the potential to solve the constraints of obtaining high fidelity biomechanics data normally only possible during laboratory studies.

Machine learning approaches applied in spinal pain research.

The purpose of this narrative review is to provide a critical reflection of how analytical machine learning approaches could provide the platform to harness variability of patient presentation to enhance clinical prediction. The review includes a summary of current knowledge on the physiological adaptations present in people with spinal pain. We discuss how contemporary evidence highlights the importance of not relying on single features when characterizing patients given the variability of physiological adaptations present in people with spinal pain. The advantages and disadvantages of current analytical strategies in contemporary basic science and epidemiological research are reviewed and we consider how analytical machine learning approaches could provide the platform to harness the variability of patient presentations to enhance clinical prediction of pain persistence or recurrence. We propose that machine learning techniques can be leveraged to translate a potentially heterogeneous set of variables into clinically useful information with the potential to enhance patient management.

A novel metric of reliability in pressure pain threshold measurement.

The inter-session Intraclass Correlation Coefficient (ICC) is a commonly investigated and clinically important metric of reliability for pressure pain threshold (PPT) measurement. However, current investigations do not account for inter-repetition variability when calculating inter-session ICC, even though a PPT measurement taken at different sessions must also imply different repetitions. The primary aim was to evaluate and report a novel metric of reliability in PPT measurement: the inter-session-repetition ICC. One rater recorded ten repetitions of PPT measurement over the lumbar region bilaterally at two sessions in twenty healthy adults using a pressure algometer. Variance components were computed using linear mixed-models and used to construct ICCs; most notably inter-session ICC and inter-session-repetition ICC. At 70.1% of the total variance, the source of greatest variability was between subjects ([Formula: see text] = 222.28 N2), whereas the source of least variability (1.5% total variance) was between sessions ([Formula: see text] = 4.83 N2). Derived inter-session and inter-session-repetition ICCs were 0.88 (95%CI: 0.77 to 0.94) and 0.73 (95%CI: 0.53 to 0.84) respectively. Inter-session-repetition ICC provides a more conservative estimate of reliability than inter-session ICC, with the magnitude of difference being clinically meaningful. Quantifying individual sources of variability enables ICC construction to be reflective of individual testing protocols.

Clinical predictive modelling of post-surgical recovery in individuals with cervical radiculopathy: a machine learning approach.

Prognostic models play an important role in the clinical management of cervical radiculopathy (CR). No study has compared the performance of modern machine learning techniques, against more traditional stepwise regression techniques, when developing prognostic models in individuals with CR. We analysed a prospective cohort dataset of 201 individuals with CR. Four modelling techniques (stepwise regression, least absolute shrinkage and selection operator [LASSO], boosting, and multivariate adaptive regression splines [MuARS]) were each used to form a prognostic model for each of four outcomes obtained at a 12 month follow-up (disability-neck disability index [NDI]), quality of life (EQ5D), present neck pain intensity, and present arm pain intensity). For all four outcomes, the differences in mean performance between all four models were small (difference of NDI < 1 point; EQ5D < 0.1 point; neck and arm pain < 2 points). Given that the predictive accuracy of all four modelling methods were clinically similar, the optimal modelling method may be selected based on the parsimony of predictors. Some of the most parsimonious models were achieved using MuARS, a non-linear technique. Modern machine learning methods may be used to probe relationships along different regions of the predictor space.

Classifying individuals with and without patellofemoral pain syndrome using ground force profiles - Development of a method using functional data boosting.

BACKGROUND: Predictors of recovery in patellofemoral pain syndrome (PFPS) currently used in prognostic models are scalar in nature, despite many physiological measures originally lying on the functional scale. Traditional modelling techniques cannot harness the potential predictive value of functional physiological variables. RESEARCH QUESTION: What is the classification performance of PFPS status of a statistical model when using functional ground reaction force (GRF) time-series? METHODS: Thirty-one individuals (control = 17, PFPS = 14) performed maximal countermovement jumps, on two force plates. The three-dimensional components of the GRF profiles were time-normalized between the start of the eccentric phase and take-off, and used as functional predictors. A statistical model was developed using functional data boosting (FDboost), for binary classification of PFPS statuses (control vs PFPS). The area under the Receiver Operating Characteristic curve (AUC) was used to quantify the model's ability to discriminate the two groups. RESULTS: The three predictors of GRF waveform achieved an average out-of-bag AUC of 93.7 %. A 1 % increase in applied medial force reduced the log odds of being in the PFPS group by 0.68 at 87 % of jump cycle. In the AP direction, a 1 % reduction in applied posterior force increased the log odds of being classified as PFPS by 1.10 at 70 % jump cycle. For the vertical GRF, a 1 % increase in applied force reduced the log odds of being classified in the PFPS group by 0.12 at 44 % of the jump cycle. SIGNIFICANCE: Using simple functional GRF variables collected during functionally relevant task, in conjunction with FDboost, produced clinically interpretable models that retain excellent classification performance in individuals with PFPS. FDboost may be an invaluable tool to be used in longitudinal cohort prognostic studies, especially when scalar and functional predictors are collected.

Interpretable machine learning models for classifying low back pain status using functional physiological variables.

PURPOSE: To evaluate the predictive performance of statistical models which distinguishes different low back pain (LBP) sub-types and healthy controls, using as input predictors the time-varying signals of electromyographic and kinematic variables, collected during low-load lifting. METHODS: Motion capture with electromyography (EMG) assessment was performed on 49 participants [healthy control (con) = 16, remission LBP (rmLBP) = 16, current LBP (LBP) = 17], whilst performing a low-load lifting task, to extract a total of 40 predictors (kinematic and electromyographic variables). Three statistical models were developed using functional data boosting (FDboost), for binary classification of LBP statuses (model 1: con vs. LBP; model 2: con vs. rmLBP; model 3: rmLBP vs. LBP). After removing collinear predictors (i.e. a correlation of > 0.7 with other predictors) and inclusion of the covariate sex, 31 predictors were included for fitting model 1, 31 predictors for model 2, and 32 predictors for model 3. RESULTS: Seven EMG predictors were selected in model 1 (area under the receiver operator curve [AUC] of 90.4%), nine predictors in model 2 (AUC of 91.2%), and seven predictors in model 3 (AUC of 96.7%). The most influential predictor was the biceps femoris muscle (peak [Formula: see text] = 0.047) in model 1, the deltoid muscle (peak [Formula: see text] = 0.052) in model 2, and the iliocostalis muscle (peak [Formula: see text] =  0.16) in model 3. CONCLUSION: The ability to transform time-varying physiological differences into clinical differences could be used in future prospective prognostic research to identify the dominant movement impairments that drive the increased risk. These slides can be retrieved under Electronic Supplementary Material.

Classifying neck pain status using scalar and functional biomechanical variables - development of a method using functional data boosting.

BACKGROUND: Individuals with neck pain have different movement and muscular activation (collectively termed as biomechanical variables) patterns compared to healthy individuals. Incorporating biomechanical variables as covariates into prognostic models is challenging due to the high dimensionality of the data. RESEARCH QUESTION: What is the classification performance of neck pain status of a statistical model which uses both scalar and functional biomechanical covariates? METHODS: Motion capture with electromyography assessment on the sternocleidomastoid, splenius cervicis, erector spinae, was performed on 21 healthy and 26 individuals with neck pain during walking over three gait conditions (rectilinear, curvilinear clockwise (CW) and counterclockwise (CCW)). After removing highly collinear variables, 94 covariates across the three conditions were used to classify neck pain status using functional data boosting (FDboost). RESULTS: Two functional covariates trunk lateral flexion angle during CCW gait, and trunk flexion angle during CW gait; and a scalar covariate, hip jerk index during CCW gait were selected. The model achieved an estimated AUC of 80.8 %. For hip jerk index, an increase in hip jerk index by one unit increased the log odds of being in the neck pain group by 0.37. A 1° increase in trunk lateral flexion angle throughout gait alone reduced the probability of being in the neck pain group from 0.5 to 0.15. A 1° increase in trunk flexion angle throughout gait alone increased the probability of being in the neck pain group from 0.5 to 0.9. SIGNIFICANCE: Interpreting the physiological significance of the extracted covariates, with other biomechanical variables, suggests that individuals with neck pain performed curvilinear walking using a stiffer strategy, compared to controls; and this increased the risk of being in the neck pain group. FDboost can produce clinically interpretable models with complex high dimensional data and could be used in future prognostic modelling studies in neck pain research.

Giardiosis and other enteropathogenic infections: a study on diarrhoeic calves in Southern Germany.

BACKGROUND: Diarrhoea induces massive problems in the rearing of calves. The aim of the study was to obtain current data about the frequency of Giardia spp., Cryptosporidium spp. and Eimeria spp. in diarrhoeic calves in Southern Germany with the particular focus on giardiosis. RESULTS: 1564 samples were analysed for the three pathogens using microscopical methods. Giardia spp. was detectable in 112/1564 samples (7.2%). The mean age was 46.5 days and the odds of being infected with Giardia spp. increased slowly up to 8 times from about 12 days to 30 days of age. There appeared to be no seasonal influence on the frequency of Giardia spp. A mono-infection with Giardia spp. was diagnosed in 46 calves (2.9%) whereas 15 calves (1.0%) had a mixed-infection with Cryptosporidium spp. and 51 calves (3.3%) with Eimeria spp. Cryptosporidium spp. and Eimeria spp. could be detected in 646/1564 samples (41.3%) and 208/1564 samples (13.3%), respectively, with a mean age of 11.3 and 55.0 days, respectively. The odds of being infected with Cryptosporidium spp. increased up to 4.5 times until an age of 10 days. After that the odds decreased continuously and was approaching zero at about 30 days. The odds of being infected with Eimeria spp. increased continuously up to 30 times from about 20 days to 60 days of age. There appeared to be no significant seasonal influence on the frequency of Cryptosporidium spp.; but there was one for Eimeria spp.: the odds of being infected with Eimeria spp. in March and April decreased by about half and increased up to 2.3 times between July and September. Additionally, as requested by the veterinarians, 1282 of those samples were analysed for E. coli, Rota-, Coronavirus and Cryptosporidium spp. using an ELISA. Obtained frequencies for these pathogens were 0.9%, 37.8%, 3.4% and 45.3% with a mean age of 24.8 days, 12.1 days, 9.0 days and 12.1 days, respectively. CONCLUSIONS: The results indicate that in Southern Germany in addition to Eimeria spp., Giardia spp. seems to play a contributing role in diarrhoea in older calves, whereas Cryptosporidium spp. and Rotavirus are mostly relevant in young calves.