Introduction. Knee arthroplasty (KA), encompassing Total Knee Replacement (TKR) and Unicompartmental Knee Replacement (UKR), is one of the most common orthopedic procedures, aimed at alleviating severe knee arthritis. Postoperative KA management, especially radiographic imaging, remains a substantial financial burden and lacks standardised protocols for its clinical utility during follow-up. Method. In this retrospective multicentre cohort study, data were analysed from January 2014 to March 2020 for adult patients undergoing primary KA at Imperial NHS Trust. Patients were followed over a five-year period. Four machine learning models were developed to evaluate if post-operative X-ray frequency can predict revision surgery. The best-performing model was used to assess the risk of revision surgery associated with different number of X-rays. Result. The study assessed 289 knees with a 2.4% revision rate. The revision group had more X-rays on average than the primary group. The best performing model was Logistic Regression (LR), which indicated that each additional X-ray raised the revision risk by 52% (p<0.001). Notably, having four or more X-rays was linked to a three-fold increase in risk of revision (OR=3.02; p<0.001). Our results align with the literature that immediate post-operative X-rays have limited utility, making the 2nd post-operative X-ray of highest importance in understanding the patient's trajectory. These insights can enhance management by improving risk stratification for patients at higher revision surgery risk. Despite LR being the best-performing model, it is limited by the dataset's significant class imbalance. Conclusion. X-ray frequency can independently predict revision surgery. This study provides insights that can guide surgeons in evidence-based post-operative decision-making. To use those findings and influence post-operative management, future studies should build on this predictive model by incorporating a more robust dataset, surgical indications, and X-ray findings. This will allow early identification of high-risk patients, allowing for personalised post-operative recommendations

Precision health aims to develop personalised and proactive strategies for predicting, preventing, and treating complex diseases such as osteoarthritis (OA). Due to OA heterogeneity, which makes developing effective treatments challenging, identifying patients at risk for accelerated disease progression is essential for efficient clinical trial design and new treatment target discovery and development. To create a reliable and interpretable precision health tool that predicts rapid knee OA progression over a 2-year period from baseline patient characteristics using an advanced automated machine learning (autoML) framework, “Autoprognosis 2.0”. All available 2-year follow-up periods of 600 patients from the FNIH OA Biomarker Consortium were analysed using “Autoprognosis 2.0” in two separate approaches, with distinct definitions of clinical outcomes: multi-class predictions (categorising disease progression into pain and/or radiographic progression) and binary predictions. Models were developed using a training set of 1352 instances and all available variables (including clinical, X-ray, MRI, and biochemical features), and validated through both stratified 10-fold cross-validation and hold-out validation on a testing set of 339 instances. Model performance was assessed using multiple evaluation metrics. Interpretability analyses were carried out to identify important predictors of progression. Our final models yielded higher accuracy scores for multi-class predictions (AUC-ROC: 0.858, 95% CI: 0.856-0.860) compared to binary predictions (AUC-ROC: 0.717, 95% CI: 0.712-0.722). Important predictors of rapid disease progression included WOMAC scores and MRI features. Additionally, accurate ML models were developed for predicting OA progression in a subgroup of patients aged 65 or younger. This study presents a reliable and interpretable precision health tool for predicting rapid knee OA progression. Our models provide accurate predictions and, importantly, allow specific predictors of rapid disease progression to be identified. Furthermore, the transparency and explainability of our methods may facilitate their acceptance by clinicians and patients, enabling effective translation to clinical practice

Anterior approach total hip arthroplasty (AA-THA) has a steep learning curve, with higher complication rates in initial cases. Proper surgical case selection during the learning curve can reduce early risk. This study aims to identify patient and radiographic factors associated with AA-THA difficulty using Machine Learning (ML). Consecutive primary AA-THA patients from two centres, operated by two expert surgeons, were enrolled (excluding patients with prior hip surgery and first 100 cases per surgeon). K- means prototype clustering – an unsupervised ML algorithm – was used with two variables - operative duration and surgical complications within 6 weeks - to cluster operations into difficult or standard groups. Radiographic measurements (neck shaft angle, offset, LCEA, inter-teardrop distance, Tonnis grade) were measured by two independent observers. These factors, alongside patient factors (BMI, age, sex, laterality) were employed in a multivariate logistic regression analysis and used for k-means clustering. Significant continuous variables were investigated for predictive accuracy using Receiver Operator Characteristics (ROC). Out of 328 THAs analyzed, 130 (40%) were classified as difficult and 198 (60%) as standard. Difficult group had a mean operative time of 106mins (range 99–116) with 2 complications, while standard group had a mean operative time of 77mins (range 69–86) with 0 complications. Decreasing inter-teardrop distance (odds ratio [OR] 0.97, 95% confidence interval [CI] 0.95–0.99, p = 0.03) and right-sided operations (OR 1.73, 95% CI 1.10–2.72, p = 0.02) were associated with operative difficulty. However, ROC analysis showed poor predictive accuracy for these factors alone, with area under the curve of 0.56. Inter-observer reliability was reported as excellent (ICC >0.7). Right-sided hips (for right-hand dominant surgeons) and decreasing inter-teardrop distance were associated with case difficulty in AA-THA. These data could guide case selection during the learning phase. A larger dataset with more complications may reveal further factors

Abstract. Introduction. Precision health aims to develop personalised and proactive strategies for predicting, preventing, and treating complex diseases such as osteoarthritis (OA), a degenerative joint disease affecting over 300 million people worldwide. Due to OA heterogeneity, which makes developing effective treatments challenging, identifying patients at risk for accelerated disease progression is essential for efficient clinical trial design and new treatment target discovery and development. Objectives. This study aims to create a trustworthy and interpretable precision health tool that predicts rapid knee OA progression based on baseline patient characteristics using an advanced automated machine learning (autoML) framework, “Autoprognosis 2.0”. Methods. All available 2-year follow-up periods of 600 patients from the FNIH OA Biomarker Consortium were analysed using “Autoprognosis 2.0” in two separate approaches, with distinct definitions of clinical outcomes: multi-class predictions (categorising patients into non-progressors, pain-only progressors, radiographic-only progressors, and both pain and radiographic progressors) and binary predictions (categorising patients into non-progressors and progressors). Models were developed using a training set of 1352 instances and all available variables (including clinical, X-ray, MRI, and biochemical features), and validated through both stratified 10-fold cross-validation and hold-out validation on a testing set of 339 instances. Model performance was assessed using multiple evaluation metrics, such as AUC-ROC, AUC-PRC, F1-score, precision, and recall. Additionally, interpretability analyses were carried out to identify important predictors of rapid disease progression. Results. Our final models yielded high accuracy scores for both multi-class predictions (AUC-ROC: 0.858, 95% CI: 0.856–0.860; AUC-PRC: 0.675, 95% CI: 0.671–0.679; F1-score: 0.560, 95% CI: 0.554–0.566) and binary predictions (AUC-ROC: 0.717, 95% CI: 0.712–0.722; AUC-PRC: 0.620, 95% CI: 0.616–0.624; F1-score: 0.676, 95% CI: 0.673–0679). Important predictors of rapid disease progression included the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scores and MRI features. Our models were further successfully validated using a hold-out dataset, which was previously omitted from model development and training (AUC-ROC: 0.877 for multi-class predictions; AUC-ROC: 0.746 for binary predictions). Additionally, accurate ML models were developed for predicting OA progression in a subgroup of patients aged 65 or younger (AUC-ROC: 0.862, 95% CI: 0.861–0.863 for multi-class predictions; AUC-ROC: 0.736, 95% CI: 0.734–0.738 for binary predictions). Conclusions. This study presents a reliable and interpretable precision health tool for predicting rapid knee OA progression using “Autoprognosis 2.0”. Our models provide accurate predictions and offer insights into important predictors of rapid disease progression. Furthermore, the transparency and interpretability of our methods may facilitate their acceptance by clinicians and patients, enabling effective utilisation in clinical practice. Future work should focus on refining these models by increasing the sample size, integrating additional features, and using independent datasets for external validation. Declaration of Interest. (b) declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported:I declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research project

Access to health care, including physiotherapy, is increasingly occurring through virtual formats. At-home adherence to physical therapy programs is often poor and few tools exist to objectively measure low back physiotherapy exercise participation without the direct supervision of a medical professional. The aim of this study was to develop and evaluate the potential for performing automatic, unsupervised video-based monitoring of at-home low back physiotherapy exercises using a single mobile phone camera. 24 healthy adult subjects performed seven exercises based on the McKenzie low back physiotherapy program while being filmed with two smartphone cameras. Joint locations were automatically extracted using an open-source pose estimation framework. Engineered features were extracted from the joint location time series and used to train a support vector machine classifier (SVC). A convolutional neural network (CNN) was trained directly on the joint location time series data to classify exercises based on a recording from a single camera. The models were evaluated using a 5-fold cross validation approach, stratified by subject, with the class-balanced accuracy used as the performance metric. Optimal performance was achieved when using a total of 12 pose estimation landmarks from the upper and lower body, with the SVC model achieving a classification accuracy of 96±4% and the CNN model an accuracy of 97±2%. This study demonstrates the feasibility of using a smartphone camera and a supervised machine learning model to effectively assess at-home low back physiotherapy adherence. This approach could provide a low-cost, scalable method for tracking adherence to physical therapy exercise programs in a variety of settings

Total knee and hip arthroplasty (TKA and THA) are the most commonly performed surgical procedures, the costs of which constitute a significant healthcare burden. Improving access to care for THA/TKA requires better efficiency. It is hypothesized that this may be possible through a two-stage approach that utilizes prediction of surgical time to enable optimization of operating room (OR) schedules.

Data from 499,432 elective unilateral arthroplasty procedures, including 302,490 TKAs, and 196,942 THAs, performed from 2014-2019 was extracted from the American College of Surgeons (ACS) National Surgical and Quality Improvement (NSQIP) database. A deep multilayer perceptron model was trained to predict duration of surgery (DOS) based on pre-operative clinical and biochemical patient factors. A two-stage approach, utilizing predicted DOS from a held out “test” dataset, was utilized to inform the daily OR schedule. The objective function of the optimization was the total OR utilization, with a penalty for overtime. The scheduling problem and constraints were simulated based on a high-volume elective arthroplasty centre in Canada. This approach was compared to current patient scheduling based on mean procedure DOS. Approaches were compared by performing 1000 simulated OR schedules.

The predict then optimize approach achieved an 18% increase in OR utilization over the mean regressor. The two-stage approach reduced overtime by 25-minutes per OR day, however it created a 7-minute increase in underutilization. Better objective value was seen in 85.1% of the simulations.

With deep learning prediction and mathematical optimization of patient scheduling it is possible to improve overall OR utilization compared to typical scheduling practices. Maximizing utilization of existing healthcare resources can, in limited resource environments, improve patient's access to arthritis care by increasing patient throughput, reducing surgical wait times and in the immediate future, help clear the backlog associated with the COVID-19 pandemic.

Physiotherapy is a critical element in successful conservative management of low back pain (LBP). The aim of this study was to develop and evaluate a system with wearable inertial sensors to objectively detect sitting postures and performance of unsupervised exercises containing movement in multiple planes (flexion, extension, rotation).

A set of 8 inertial sensors were placed on 19 healthy adult subjects. Data was acquired as they performed 7 McKenzie low-back exercises and 3 sitting posture positions. This data was used to train two models (Random Forest (RF) and XGBoost (XGB)) using engineered time series features. In addition, a convolutional neural network (CNN) was trained directly on the time series data. A feature importance analysis was performed to identify sensor locations and channels that contributed most to the models. Finally, a subset of sensor locations and channels was included in a hyperparameter grid search to identify the optimal sensor configuration and the best performing algorithm(s) for exercise classification. Models were evaluated using F1-score in a 10-fold cross validation approach.

The optimal hardware configuration was identified as a 3-sensor setup using lower back, left thigh, and right ankle sensors with acceleration, gyroscope, and magnetometer channels. The XBG model achieved the highest exercise (F1=0.94±0.03) and posture (F1=0.90±0.11) classification scores. The CNN achieved similar results with the same sensor locations, using only the accelerometer and gyroscope channels for exercise classification (F1=0.94±0.02) and the accelerometer channel alone for posture classification (F1=0.91±0.03).

This study demonstrates the potential of a 3-sensor lower body wearable solution (e.g. smart pants) that can identify proper sitting postures and exercises in multiple planes, suitable for low back pain. This technology has the potential to improve the effectiveness of LBP rehabilitation by facilitating quantitative feedback, early problem diagnosis, and possible remote monitoring.

To detect early signs of infection infrared thermography has been suggested to provide quantitative information. Our vision is to invent a pin site infection thermographic surveillance tool for patients at home. A preliminary step to this goal is the aim of this study, to automate the process of locating the pin and detecting the pin sites in thermal images efficiently, exactly, and reliably for extracting pin site temperatures.

A total of 1708 pin sites was investigated with Thermography and augmented by 9 different methods in to totally 10.409 images. The dataset was divided into a training set (n=8325), a validation set (n=1040), and a test set (n=1044) of images. The Pin Detection Model (PDM) was developed as follows: A You Only Look Once (YOLOv5) based object detection model with a Complete Detection Intersection over Union (CDIoU), it was pre-trained and finetuned by the through transfer learning. The basic performance of the YOLOv5 with CDIoU model was compared with other conventional models (FCOS and YOLOv4) for deep and transition learning to improve performance and precision. Maximum Temperature Extraction (MTE) Based on Region of Interest (ROI) for all pin sites was generated by the model. Inference of MTE using PDM with infected and un-infected datasets was investigated.

An automatic tool that can identify and annotate pin sites on conventional images using bounding boxes was established. The bounding box was transferred to the infrared image. The PMD algorithm was built on YOLOv5 with CDIoU and has a precision of 0.976. The model offers the pin site detection in 1.8 milliseconds. The thermal data from ROI at the pin site was automatically extracted.

These results enable automatic pin site annotation on thermography. The model tracks the correlation between temperature and infection from the detected pin sites and demonstrates it is a promising tool for automatic pin site detection and maximum temperature extraction for further infection studies. Our work for automatic pin site annotation on thermography paves the way for future research on infection assessment using thermography.

Abstract. Objectives. Spinal disorders such as back pain incur a substantial societal and economic burden. Unfortunately, there is lack of understanding and treatment of these disorders are further impeded by the inability to assess spinal forces in vivo. The aim of this project is to address this challenge by developing and testing a novel image-driven approach that will assess the forces in an individual's spine in vivo by incorporating information acquired from multimodal imaging (magnetic resonance imaging (MRI) and biplane X-rays) in a subject-specific model. Methods. Magnetic resonance and biplane X-ray imaging are used to capture information about the anatomy, tissues, and motion of an individual's spine as they perform a range of everyday activities. This information is then utilised in a subject-specific computational model based on the finite element method to predict the forces in their spine. The project is also utilising novel machine learning algorithms and in vitro, six-axis mechanical testing on human, porcine and bovine samples to develop and test the modelling methods rigorously. Results & Discussion. MRI sequences have been identified that provide high-quality image data and information on different tissue types which will be used to predict subject-specific disc properties. In-vivo protocols to capture motion analysis, EMG muscle activity, and video X-rays of the spine have been designed with planned data collection of 15 healthy volunteers. Preliminary modelling work has evaluated potential machine learning approaches and quantified the sensitivity of the models developed to material properties. Conclusion. The development and testing of these image-driven subject-specific spine models will provide a new tool for determining forces in the spine. It will also provide new tools for measuring and modelling spine movement and quantifying the properties of the spinal tissues. Acknowledgments. Funding from the EPSRC: EP/V036602/1 (Meakin, Holsgrove & Javadi) and EP/V032275/1 (Holt & Williams). Declaration of Interest. (b) declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported:I declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research project

Introduction. The increased prevalence of osteoporosis in the patient population undergoing reverse shoulder arthroplasty (RSA) results in significantly increased complication rates. Mainly demographic and clinical predictors are currently taken into the preoperative assessment for risk stratification without quantification of preoperative computed tomography (CT) data (e.g. bone density). It was hypothesized that preoperative CT bone density measures would provide objective quantification with subsequent classification of the patients’ humeral bone quality. Methods. Thirteen bone density parameters from 345 preoperative CT scans of a clinical RSA cohort represented the data set in this study. The data set was divided into testing (30%) and training data (70%), latter included an 8-fold cross validation. Variable selection was performed by choosing the variables with the highest descriptive value for each correlation clustered variables. Machine learning models were used to improve the clustering (Hierarchical Ward) and classification (Support Vector Machine (SVM)) of bone densities at risk for complications and were compared to a conventional statistical model (Logistic Regression (LR)). Results. Clustering partitioned this cohort (training data set) into a high bone density subgroup consisting of 96 patients and a low bone density subgroup consisting of 146 patients. The optimal number of clusters (n = 2) was determined based on optimization metrics. Discrimination of the cross validated classification model showed comparable performance for the training (accuracy=91.2%; AUC=0.967) and testing data (accuracy=90.5 %; AUC=0.958) while outperforming the conventional statistical model (Logistic Regression (LR)). Local interpretable model-agnostic explanations (LIME) were created for each patient to explain how the predicted output was achieved. Conclusion. The trained and tested model provides preoperative information for surgeons treating patients with potentially poor bone quality. The use of machine learning and patient-specific calibration showed that multiple 3D bone density scores improved accuracy for objective preoperative bone quality assessment

Introduction. Experimental bone research often generates large amounts of histology and histomorphometry data, and the analysis of these data can be time-consuming and trivial. Machine learning offers a viable alternative to manual analysis for measuring e.g. bone volume versus total volume. The objective was to develop a neural network for image segmentation, and to assess the accuracy of this network when applied to ectopic bone formation samples compared to a ground truth. Method. Thirteen tissue slides totaling 114 megapixels of ectopic bone formation were selected for model building. Slides were split into training, validation, and test data, with the test data reserved and only used for the final model assessment. We developed a neural network resembling U-Net that takes 512×512 pixel tiles. To improve model robustness, images were augmented online during training. The network was trained for 3 days on a NVidia Tesla K80 provided by a free online learning platform against ground truth masks annotated by an experienced researcher. Result. During training, the validation accuracy improved and stabilised at approx. 95%. The test accuracy was 96.1 %. Conclusion. Most experiments using ectopic bone formation will yield an inter-observer or inter-method variance of far more than 5%, so the current approach may be a valid and feasible technique for automated image segmentation for large datasets. More data or a consensus-based ground truth may improve training stability and validation accuracy. The code and data of this project are available upon request and will be available online as part of our publication

Artificial Intelligence (AI) is becoming more powerful but is barely used to counter the growth in health care burden. AI applications to increase efficiency in orthopedics are rare. We questioned if (1) we could train machine learning (ML) algorithms, based on answers from digitalized history taking questionnaires, to predict treatment of hip osteoartritis (either conservative or surgical); (2) such an algorithm could streamline clinical consultation. Multiple ML models were trained on 600 annotated (80% training, 20% test) digital history taking questionnaires, acquired before consultation. Best performing models, based on balanced accuracy and optimized automated hyperparameter tuning, were build into our daily clinical orthopedic practice. Fifty patients with hip complaints (>45 years) were prospectively predicted and planned (partly blinded, partly unblinded) for consultation with the physician assistant (conservative) or orthopedic surgeon (operative). Tailored patient information based on the prediction was automatically sent to a smartphone app. Level of evidence: IV. Random Forest and BernoulliNB were the most accurate ML models (0.75 balanced accuracy). Treatment prediction was correct in 45 out of 50 consultations (90%), p<0.0001 (sign and binomial test). Specialized consultations where conservatively predicted patients were seen by the physician assistant and surgical patients by the orthopedic surgeon were highly appreciated and effective. Treatment strategy of hip osteoartritis based on answers from digital history taking questionnaires was accurately predicted before patients entered the hospital. This can make outpatient consultation scheduling more efficient and tailor pre-consultation patient education

Currently implemented accuracy metrics in open-source libraries for segmentation by supervised machine learning are typically one-dimensional scores [1]. While extremely relevant to evaluate applicability in clinics, anatomical location of segmentation errors is often neglected. This study aims to include the three-dimensional (3D) spatial information in the development of a novel framework for segmentation accuracy evaluation and comparison between different methods. Predicted and ground truth (manually segmented) segmentation masks are meshed into 3D surfaces. A template mesh of the same anatomical structure is then registered to all ground truth 3D surfaces. This ensures all surface points on the ground truth meshes to be in the same anatomically homologous order. Next, point-wise surface deviations between the registered ground truth mesh and the meshed segmentation prediction are calculated and allow for color plotting of point-wise descriptive statistics. Statistical parametric mapping includes point-wise false discovery rate (FDR) adjusted p-values (also referred to as q-values). The framework reads volumetric image data containing the segmentation masks of both ground truth and segmentation prediction. 3D color plots containing descriptive statistics (mean absolute value, maximal value,…) on point-wise segmentation errors are rendered. As an example, we compared segmentation results of nnUNet [2], UNet++ [3] and UNETR [4] by visualizing the mean absolute error (surface deviation from ground truth) as a color plot on the 3D model of bone and cartilage of the mean distal femur. A novel framework to evaluate segmentation accuracy is presented. Output includes anatomical information on the segmentation errors, as well as point-wise comparative statistics on different segmentation algorithms. Clearly, this allows for a better informed decision-making process when selecting the best algorithm for a specific clinical application

Serial section electron microscopy (SSEM) was initially developed to map the neural connections in the brain. SSEM eventually led to the term ‘Connectomics’ to be coined to describe process of following a cell or structure through a volume of tissue. This permits the true three-dimensionality to be appreciated and relationships between cells and structures. The purpose of this study was to utilize this methodology to interrogate S. aureus infected bone. Bone samples were harvested from mice tibia infected with S. aureus and were fixed, decalcified, and osmicated. The samples were paraffin embedded and 5-micron sections were cut to identify regions of bacterial invasion into the osteocyte-lacuna-canalicular-network (OLCN). This area was cut from the paraffin block, deparaffinized, post-fixed and reprocessed into epoxy resin. Serial sections were cut at 60nm and collected onto Kapton tape utilizing the Automated Tape-collecting Ultramicrotome (ATUMtome) system. Samples were mounted onto 4” silicon wafers and post-stained with 2% uranyl acetate followed by 0.3% lead citrate and carbon coated. A ZEISS GeminiSEM 450 scanning electron microscope fitted with an electron backscatter diffusion detector was used to image the sections. The image stack was aligned and segmented using the open-source software, VASTlite. 264 serial sections were imaged, representing approximately 40 × 45 × 15-micron (x, y, z) volume of tissue. 70% of the canaliculi demonstrated infiltration by S. aureus. This study demonstrates that SSEM can be applied to the skeletal system and provide a new solution to investigate the OLCN system. It is feasible that this methodology could be implemented to investigate why some canaliculi are resistant to colonization and potentially opens up a new direction for the prevention of chronic osteomyelitis. In order to make this a realistic target, automated segmentation methodologies utilizing machine learning must be developed and applied to the bone tissue datasets

Gait analysis is an indispensable tool for scientific assessment and treatment of individuals whose ability to walk is impaired. The high cost of installation and operation are a major limitation for wide-spread use in clinical routine. Advances in Artificial Intelligence (AI) could significantly reduce the required instrumentation. A mobile phone could be all equipment necessary for 3D gait analysis. MediaPipe Pose provided by Google Research is such a Machine Learning approach for human body tracking from monocular RGB video frames that is detecting 3D-landmarks of the human body. Aim of this study was to analyze the accuracy of gait phase detection based on the joint landmarks identified by the AI system. Motion data from 10 healthy volunteers walking on a treadmill with a fixed speed of 4.5km/h (Callis, Sprintex, Germany) was sampled with a mobile phone (iPhone SE 2nd Generation, Apple). The video was processed with Mediapipe Pose (Version 0.9.1.0) using custom python software. Gait phases (Initial Contact - IC and Toe Off - TO) were detected from the angular velocities of the lower legs. For the determination of ground truth, the movement was simultaneously recorded with the AS-200 System (LaiTronic GmbH, Innsbruck, Austria). The number of detected strides, the error in IC detection and stance phase duration was calculated. In total, 1692 strides were detected from the reference system during the trials from which the AI-system identified 679 strides. The absolute mean error (AME) in IC detection was 39.3 ± 36.6 ms while the AME for stance duration was 187.6 ± 140 ms. Landmark detection is a challenging task for the AI-system as can clearly be seen be the rate of only 40% detected strides. As mentioned by Fadillioglu et al., error in TO-detection is higher than in IC-detection

Abstract. Objectives. Biomechanics is an essential form of measurement in the understanding of the development and progression of osteoarthritis (OA). However, the number of participants in biomechanical studies are often small and there is limited ways to share or combine data from across institutions or studies. This is essential for applying modern machine learning methods, where large, complex datasets can be used to identify patterns in the data. Using these data-driven approaches, it could be possible to better predict the optimal interventions for patients at an early stage, potentially avoiding pain and inappropriate surgery or rehabilitation. In this project we developed a prototype database platform for combining and sharing biomechanics datasets. The database includes methods for importing and standardising data and associated variables, to create a seamless, searchable combined dataset of both healthy and knee OA biomechanics. Methods. Data was curated through calls to members of the OATech Network+ (. https://www.oatechnetwork.org/. ). The requirements were 3D motion capture data from previous studies that related to analysing the biomechanics of knee OA, including participants with OA at any stage of progression plus healthy controls. As a minimum we required kinematic data of the lower limbs, plus associated kinetic data (i.e. ground reaction forces). Any additional, complementary data such as EMG could also be provided. Relevant ethical approvals had to be in place that allowed re-use of the data for other research purposes. The datasets were uploaded to a University hosted cloud platform. The database platform was developed using Javascript and hosted on a Windows server, located and managed within the department. Results. Three independent datasets were curated following the call to OATech Network+ members. These originated from separate studies collected from biomechanics labs at Cardiff University, Keele University, and Imperial College London. Participants with knee OA were at various stages of progression and all datasets included healthy controls. The total sample size of the three datasets is n=244, split approximately equally between healthy and knee OA participants. Naming conventions and formatting of the exported data varied greatly across datasets. Datasets were therefore formatted into a common format prior to upload, with guidelines developed for future contributions. Uploading data at the marker set level was too complicated for combination at the prototype stage. Therefore, processed variables relating to joint angles and joint moments were used. The resulting prototype database included an import function to align and standardise variables. A a simple query tool was further developed to extract outputs from the database, along with a suitable user interface for basic data exploration. Conclusion. Combining biomechanics dataset presents a wide range of challenges from both a technical and data governance context. Here we have taken the first steps to demonstrate a proof-of-concept that can combine heterogenous data from independent OA-related biomechanics studies into a combined, searchable resource. Expanding this in the future to a fully open access database will create an essential resource that will facilitate the application of data-driven models and analyses for better understanding, stratification and prediction of OA progression. Declaration of Interest. (b) declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported:I declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research project

During OA the homeostasis of healthy articular chondrocytes is dysregulated, which leads to a phenotypical transition of the cells, further influenced by external stimuli. Chondrocytes sense those stimuli, integrate them at the intracellular level and respond by modifying their secretory and molecular state. This process is controlled by a complex interplay of intracellular factors. Each factor is influenced by a myriad of feedback mechanisms, making the prediction of what will happen in case of external perturbation challenging. Hampering the hypertrophic phenotype has emerged as a potential therapeutic strategy to help OA patients (Ripmeester et al. 2018). Therefore, we developed a computational model of the chondrocyte's underlying regulatory network (RN) to identify key regulators as potential drug targets. A mechanistic mathematical model of articular chondrocyte differentiation was implemented with a semi-quantitative formalism. It is composed of a protein RN and a gene RN(GRN) and developed by combining two strategies. First, we established a mechanistic network based on accumulation of decades of biological knowledge. Second, we combined that mechanistic network with data-driven modelling by inferring an OA-GRN using an ensemble of machine learning methods. This required a large gene expression dataset, provided by distinct public microarrays merged through an in-house pipeline for cross-platform integration. We successfully merged various micro-array experiments into one single dataset where the biological variance was predominant over the batch effect from the different technical platforms. The gain of information provided by this merge enabled us to reconstruct an OA-GRN which subsequently served to complete our mechanistic model. With this model, we studied the system's multi-stability, equating the model's stable states to chondrocyte phenotypes. The network structure explained the occurrence of two biologically relevant phenotypes: a hypertrophic-like and a healthy-like phenotype, recognized based on known cell state markers. Second, we tested several hypotheses that could trigger the onset of OA to validate the model with relevant biological phenomena. For instance, forced inflammation pushed the chondrocyte towards hypertrophy but this was partly rescued by higher levels of TGF-β. However, we could annihilate this rescue by concomitantly mimicking an increase in the ALK1/ALK5 balance. Finally, we performed a screening of in-silico (combinatorial) perturbations (inhibitions and/or over-activations) to identify key molecular factors involved in the stability of the chondrocyte state. More precisely, we looked for the most potent conditions for decreasing hypertrophy. Preliminary validation experiments have confirmed that PKA activation could decrease the hypertrophic phenotype in primary chondrocytes. Importantly the in-silico results highlighted that targeting two factors at the same time would greatly help reducing hypertrophic changes. A priori testing of conditions with in-silico models may cut time and cost of experiments via target prioritization and opens new routes for OA combinatorial therapies

Our lab uses computer-aided design to build in silico libraries of surface topographies, which we reproduce on polymeric chips and analyse for cellular responses using high content imaging and machine learning. In addition, we use transcriptomics and mass spectrometry to obtain a holistic view of biomaterial-mediated cellular responses and build gene regulatory networks thereof. This approach enables us to parameterize both the biomaterial properties as well as the cell response and to correlate them using computational tools. We think that this approach can be translated to other biomaterial platforms, such as polymer arrays, and foresee large scale crosstalk between them if we can standardize our methodology to describe the materials and to analyse the cells. To this end, we have started cBIT, the compendium for biomaterial-induced transcriptomics, an open-source database in which scientists can deposit and search material-induced transcriptomics data. The meta-analyses that cBIT enables, could lead to the identification of genes, pathways or expression profiles that can inform the design and development of new biomaterials. As such, by generating new information and simultaneously accumulating it in cBIT, we expect it is possible to one day predict cell responses to biomaterials