Gareth Jones’s research while affiliated with Imperial College London and other places

Knee osteoarthritis (OA) affects over 650 million patients worldwide. Total knee replacement is aimed at end-stage OA to relieve symptoms of pain, stiffness and reduced mobility. However, the role of imaging modalities in monitoring symptomatic disease progression remains unclear. This study aimed to compare machine learning (ML) models, with and without imaging features, in predicting the two-year Western Ontario and McMaster Universities Arthritis Index (WOMAC) score for knee OA patients. We included 2408 patients from the Osteoarthritis Initiative (OAI) database, with 629 patients from the Multicenter Osteoarthritis Study (MOST) database. The clinical dataset included 18 clinical features, while the imaging dataset contained an additional 10 imaging features. Minimal Clinically Important Difference (MCID) was set to 24, reflecting meaningful physical impairment. Clinical and imaging dataset models produced similar area under curve (AUC) scores, highlighting low differences in performance AUC < 0.025). For both clinical and imaging datasets, Gradient Boosting Machine (GBM) models performed the best in the external validation, with a clinically acceptable AUC of 0.734 (95% CI 0.687–0.781) and 0.747 (95% CI 0.701–0.792), respectively. The five features identified included educational background, family history of osteoarthritis, co-morbidities, use of osteoporosis medications and previous knee procedures. This is the first study to demonstrate that ML models achieve comparable performance with and without imaging features.

Purpose Femorotibial angle (FTA) is a convenient measure of coronal knee alignment that can be extracted from a short knee radiograph, avoiding the additional radiation exposure and specialist equipment required for full‐leg radiographs. While intra‐ and inter‐reader reproducibility from the same image has been reported, the full scan–rescan reproducibility across images, as calculated in this study, has not. Methods In this study, 4589 FTA measurement pairs from 2586 subjects acquired a year apart were used to estimate FTA scan–rescan reproducibility using data from the Osteoarthritis Initiative. Subjects with radiographic progression of osteoarthritis or other conditions that may cause a change in coronal knee alignment were excluded. Measurement pairs were analysed using paired‐samples tests to detect differences and compared to symptomatic changes in Western Ontario and McMaster Universities Arthritis Index scores for joint pain, stiffness and physical function to detect correlations. Results The 95% limit of agreement and the paired‐samples correlation were calculated with high precision to be [−1.76°, +1.78°] and 0.938, considerably worse than the corresponding figures for intra‐ and inter‐reader reproducibility, without relation to symptomatic or radiographic changes in knee condition. This error will weakly attenuate and values from their true values in correlative studies involving FTA. The realistic maximum value for is 87% and for Pearson's is 93%. Conclusion The scan–rescan reproducibility in FTA is almost double the intra‐ and inter‐reader reliability from a single scan. At almost ±2° accuracy, FTA is inappropriate for surgical use, but it is sufficiently reproducible to produce good correlations in studies predicting disease incidence and progression. Level of Evidence Level II, retrospective study.

Objectives Ultrasound speckle tracking is a safe and non-invasive diagnostic tool to measure soft tissue deformation and strain. In orthopaedics, it could have broad application to measure how injury or surgery affects muscle, tendon or ligament biomechanics. However, its application requires custom tuning of the speckle-tracking algorithm then validation against gold-standard reference data. Implementing an experiment to acquire these data takes months and is expensive, and therefore prohibits use for new applications. Here, we present an alternative optimisation approach that automatically finds suitable machine and algorithmic settings without requiring gold-standard reference data. Methods The optimisation routine consisted of two steps. First, convergence of the displacement field was tested to exclude the settings that would not track the underlying tissue motion (e.g. frame rates that were too low). Second, repeatability was maximised through a surrogate optimisation scheme. All settings that could influence the strain calculation were included, ranging from acquisition settings to post-processing smoothing and filtering settings, totalling >1,000,000 combinations of settings. The optimisation criterion minimised the normalised standard deviation between strain maps of repeat measures. The optimisation approach was validated for the medial collateral ligament (MCL) with quasi-static testing on porcine joints (n=3), and dynamic testing on a cadaveric human knee (n=1, female, aged 49). Porcine joints were fully dissected except for the MCL and loaded in a material-testing machine (0 to 3% strain at 0.2 Hz), which was captured using both ultrasound (>14 repeats per specimen) and optical digital image correlation (DIC). For the human cadaveric knee (undissected), 3 repeat ultrasound acquisitions were taken at 18 different anterior/posterior positions over the MCL while the knee was extended/flexed between 0° and 90° in a knee extension rig. Simultaneous optical tracking recorded the position of the ultrasound transducer, knee kinematics and the MCL attachments (which were digitised under direct visualisation post testing). Half of the data collected was used for optimisation of the speckle tracking algorithms for the porcine and human MCLs separately, with the remaining unseen data used as a validation test set. Results For the porcine MCLs, ultrasound strains closely matched DIC strains (R ² > 0.98, RMSE < 0.59%) (Figure 1A). For the human MCL (Figure 1B), ultrasound strains matched the strains estimated from the optically tracked displacements of the MCL attachments. Furthermore, strains developed during flexion were highly correlated with AP position (R = 0.94) with strains decreasing the further posterior the transducer was on the ligament. This is in line with previously reported length change values for the posterior, intermediate and anterior bundles of the MCL. Conclusions Ultrasound speckle tracking algorithms can be adapted for new applications without ground-truth data by using an optimisation approach that verifies displacement field convergence then minimises variance between repeat measurements. This optimisation routine was insensitive to anatomical variation and loading conditions, working for both porcine and human MCLs, and for quasi-static and dynamic loading. This will facilitate research into changes in musculoskeletal tissue motion due to abnormalities or pathologies. Declaration of Interest (a) fully declare any financial or other potential conflict of interest

Background: Knee alignment affects the development and surgical treatment of knee osteoarthritis. Automating femorotibial angle (FTA) and hip-knee-ankle angle (HKA) measurement from radiographs could improve reliability and save time. Further, if HKA could be predicted from knee-only radiographs then radiation exposure could be reduced and the need for specialist equipment and personnel avoided. The aim of this research was to assess if deep learning methods could predict FTA and HKA angle from posteroanterior (PA) knee radiographs. Methods: Convolutional neural networks with densely connected final layers were trained to analyse PA knee radiographs from the Osteoarthritis Initiative (OAI) database. The FTA dataset with 6149 radiographs and HKA dataset with 2351 radiographs were split into training, validation, and test datasets in a 70:15:15 ratio. Separate models were developed for the prediction of FTA and HKA and their accuracy was quantified using mean squared error as loss function. Heat maps were used to identify the anatomical features within each image that most contributed to the predicted angles. Results: High accuracy was achieved for both FTA (mean absolute error 0.8°) and HKA (mean absolute error 1.7°). Heat maps for both models were concentrated on the knee anatomy and could prove a valuable tool for assessing prediction reliability in clinical application. Conclusion: Deep learning techniques enable fast, reliable and accurate predictions of both FTA and HKA from plain knee radiographs and could lead to cost savings for healthcare providers and reduced radiation exposure for patients.

Aims The metabolic equivalent of task (MET) score examines patient performance in relation to energy expenditure before and after knee arthroplasty. This study assesses its use in a knee arthroplasty population in comparison with the widely used Oxford Knee Score (OKS) and EuroQol five-dimension index (EQ-5D), which are reported to be limited by ceiling effects. Methods A total of 116 patients with OKS, EQ-5D, and MET scores before, and at least six months following, unilateral primary knee arthroplasty were identified from a database. Procedures were performed by a single surgeon between 2014 and 2019 consecutively. Scores were analyzed for normality, skewness, kurtosis, and the presence of ceiling/floor effects. Concurrent validity between the MET score, OKS, and EQ-5D was assessed using Spearman’s rank. Results Postoperatively the OKS and EQ-5D demonstrated negative skews in distribution, with high kurtosis at six months and one year. The OKS demonstrated a ceiling effect at one year (15.7%) postoperatively. The EQ-5D demonstrated a ceiling effect at six months (30.2%) and one year (39.8%) postoperatively. The MET score did not demonstrate a skewed distribution or ceiling effect either at six months or one year postoperatively. Weak-moderate correlations were noted between the MET score and conventional scores at six months and one year postoperatively. Conclusion In contrast to the OKS and EQ-5D, the MET score was normally distributed postoperatively with no ceiling effect. It is worth consideration as an arthroplasty outcome measure, particularly for patients with high expectations. Cite this article: Bone Jt Open 2023;4(3):129–137.

Objectives Knee osteoarthritis is a major cause of physical disability and reduced quality of life, with end-stage disease often treated by total knee replacement (TKR). We set out to develop and externally validate a machine learning model capable of predicting the need for a TKR in 2 and 5 years time using routinely collected health data. Design A prospective study using datasets Osteoarthritis Initiative (OAI) and the Multicentre Osteoarthritis Study (MOST). OAI data were used to train the models while MOST data formed the external test set. The data were preprocessed using feature selection to curate 45 candidate features including demographics, medical history, imaging assessments, history of intervention and outcome. Setting The study was conducted using two multicentre USA-based datasets of participants with or at high risk of knee OA. Participants The study excluded participants with at least one existing TKR. OAI dataset included participants aged 45–79 years of which 3234 were used for training and 809 for internal testing, while MOST involved participants aged 50–79 and 2248 were used for external testing. Main outcome measures The primary outcome of this study was prediction of TKR onset at 2 and 5 years. Performance was evaluated using area under the curve (AUC) and F1-score and key predictors identified. Results For the best performing model (gradient boosting machine), the AUC at 2 years was 0.913 (95% CI 0.876 to 0.951), and at 5 years 0.873 (95% CI 0.839 to 0.907). Radiographic-derived features, questionnaire-based assessments alongside the patient’s educational attainment were key predictors for these models. Conclusions Our approach suggests that routinely collected patient data are sufficient to drive a predictive model with a clinically acceptable level of accuracy (AUC>0.7) and is the first such tool to be externally validated. This level of accuracy is higher than previously published models utilising MRI data, which is not routinely collected.

Background: This systematic review aims to ascertain how accurately 3D models can be predicted from two-dimensional (2D) imaging utilising statistical shape modelling. Methods: A systematic search of published literature was conducted in September 2022. All papers which assessed the accuracy of 3D models predicted from 2D imaging utilising statistical shape models and which validated the models against the ground truth were eligible. Results: 2127 papers were screened and a total of 34 studies were included for final data extraction. The best overall achievable accuracy was 0.45 mm (root mean square error) and 0.16 mm (average error). Conclusion: Statistical shape modelling can predict detailed 3D anatomical models from minimal 2D imaging. Future studies should report the intended application domain of the model, the level of accuracy required, the underlying demographics of subjects, and the method in which accuracy was calculated, with root mean square error recommended if appropriate. This article is protected by copyright. All rights reserved.

Purpose: The aim of this study was to outline a fully automatic tool capable of reliably predicting the most suitable total knee replacement implant sizes for patients, using bi-planar X-ray images. By eliminating the need for manual templating or guiding software tools via the adoption of convolutional neural networks, time and resource requirements for pre-operative assessment and surgery could be reduced, the risk of human error minimized, and patients could see improved outcomes. Methods: The tool utilizes a machine learning-based 2D—3D pipeline to generate accurate predictions of subjects’ distal femur and proximal tibia bones from X-ray images. It then virtually fits different implant models and sizes to the 3D predictions, calculates the implant to bone root-mean-squared error and maximum over/under hang for each, and advises the best option for the patient. The tool was tested on 78, predominantly White subjects (45 female/33 male), using generic femur component and tibia plate designs scaled to sizes obtained for five commercially available products. The predictions were then compared to the ground truth best options, determined using subjects’ MRI data. Results: The tool achieved average femur component size prediction accuracies across the five implant models of 77.95% in terms of global fit (root-mean-squared error), and 71.79% for minimizing over/underhang. These increased to 99.74% and 99.49% with ±1 size permitted. For tibia plates, the average prediction accuracies were 80.51% and 72.82% respectively. These increased to 99.74% and 98.98% for ±1 size. Better prediction accuracies were obtained for implant models with fewer size options, however such models more frequently resulted in a poor fit. Conclusion: A fully automatic tool was developed and found to enable higher prediction accuracies than generally reported for manual templating techniques, as well as similar computational methods.

Background Knee osteoarthritis (OA) is the most common form of OA and a leading cause of disability worldwide. Chronic pain and functional loss secondary to knee OA put patients at risk of developing depression, which can also impair their treatment response. However, no tools exist to assist clinicians in identifying patients at risk. Machine learning (ML) predictive models may offer a solution. We investigated whether ML models could predict the development of depression in patients with knee OA and examined which features are the most predictive. Objective The primary aim of this study was to develop and test an ML model to predict depression in patients with knee OA at 2 years and to validate the models using an external data set. The secondary aim was to identify the most important predictive features used by the ML algorithms. Methods Osteoarthritis Initiative Study (OAI) data were used for model development and external validation was performed using Multicenter Osteoarthritis Study (MOST) data. Forty-two features were selected, which denoted routinely collected demographic and clinical data such as patient demographics, past medical history, knee OA history, baseline examination findings, and patient-reported outcome measures. Six different ML classification models were trained (logistic regression, least absolute shrinkage and selection operator [LASSO], ridge regression, decision tree, random forest, and gradient boosting machine). The primary outcome was to predict depression at 2 years following study enrollment. The presence of depression was defined using the Center for Epidemiological Studies Depression Scale. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and F1 score. The most important features were extracted from the best-performing model on external validation. Results A total of 5947 patients were included in this study, with 2969 in the training set, 742 in the test set, and 2236 in the external validation set. For the test set, the AUC ranged from 0.673 (95% CI 0.604-0.742) to 0.869 (95% CI 0.824-0.913), with an F1 score of 0.435 to 0.490. On external validation, the AUC varied from 0.720 (95% CI 0.685-0.755) to 0.876 (95% CI 0.853-0.899), with an F1 score of 0.456 to 0.563. LASSO modeling offered the highest predictive performance. Blood pressure, baseline depression score, knee pain and stiffness, and quality of life were the most predictive features. Conclusions To our knowledge, this is the first study to apply ML classification models to predict depression in patients with knee OA. Our study showed that ML models can deliver a clinically acceptable level of performance (AUC>0.7) in predicting the development of depression using routinely available demographic and clinical data. Further work is required to address the class imbalance in the training data and to evaluate the clinical utility of the models in facilitating early intervention and improved outcomes.