Advances in cancer therapy have prolonged patient survival even in the presence of disseminated disease and an increasing number of cancer patients are living with metastatic bone disease (MBD). The proximal femur is the most common long bone involved in MBD and pathologic fractures of the femur are associated with significant morbidity, mortality and loss of quality of life (QoL). Successful prophylactic surgery for an impending fracture of the proximal femur has been shown in multiple cohort studies to result in longer survival, preserved mobility, lower transfusion rates and shorter post-operative hospital stays. However, there is currently no optimal method to predict a pathologic fracture. The most well-known tool is Mirel's criteria, established in 1989 and is limited from guiding clinical practice due to poor specificity and sensitivity. The ideal clinical decision support tool will be of the highest sensitivity and specificity, non-invasive, generalizable to all patients, and not a burden on hospital resources or the patient's time. Our research uses novel machine learning techniques to develop a model to fill this considerable gap in the treatment pathway of MBD of the femur. The goal of our study is to train a convolutional neural network (CNN) to predict fracture risk when metastatic bone disease is present in the proximal femur. Our fracture risk prediction tool was developed by analysis of prospectively collected data of consecutive MBD patients presenting from 2009–2016. Patients with primary bone tumors, pathologic fractures at initial presentation, and hematologic malignancies were excluded. A total of 546 patients comprising 114 pathologic fractures were included. Every patient had at least one Anterior-Posterior X-ray and clinical data including patient demographics, Mirel's criteria, tumor biology, all previous radiation and chemotherapy received, multiple pain and function scores, medications and time to fracture or time to death. We have trained a convolutional neural network (CNN) with AP X-ray images of 546 patients with metastatic bone disease of the proximal femur. The digital X-ray data is converted into a matrix representing the color information at each pixel. Our CNN contains five convolutional layers, a fully connected layers of 512 units and a final output layer. As the information passes through successive levels of the network, higher level features are abstracted from the data. The model converges on two fully connected deep neural network layers that output the risk of fracture. This prediction is compared to the true outcome, and any errors are back-propagated through the network to accordingly adjust the weights between connections, until overall prediction accuracy is optimized. Methods to improve learning included using stochastic gradient descent with a learning rate of 0.01 and a momentum rate of 0.9. We used average classification accuracy and the average F1 score across five test sets to measure model performance. We compute F1 = 2 x (precision x recall)/(precision + recall). F1 is a measure of a model's accuracy in binary classification, in our case, whether a lesion would result in pathologic fracture or not. Our model achieved 88.2% accuracy in predicting fracture risk across five-fold cross validation testing. The F1 statistic is 0.87. This is the first reported application of convolutional neural networks, a machine learning algorithm, to this important Orthopaedic problem. Our neural network model was able to achieve reasonable accuracy in classifying fracture risk of metastatic proximal femur lesions from analysis of X-rays and clinical information. Our future work will aim to externally validate this algorithm on an international cohort

Advances in cancer therapy have prolonged cancer patient survival even in the presence of disseminated disease and an increasing number of cancer patients are living with metastatic bone disease (MBD). The proximal femur is the most common long bone involved in MBD and pathologic fractures of the femur are associated with significant morbidity, mortality and loss of quality of life (QoL). Successful prophylactic surgery for an impending fracture of the proximal femur has been shown in multiple cohort studies to result in patients more likely to walk after surgery, longer survival, lower transfusion rates and shorter post-operative hospital stays. However, there is currently no optimal method to predict a pathologic fracture. The most well-known tool is Mirel's criteria, established in 1989 and is limited from guiding clinical practice due to poor specificity and sensitivity. The goal of our study is to train a convolutional neural network (CNN) to predict fracture risk when metastatic bone disease is present in the proximal femur. Our fracture risk prediction tool was developed by analysis of prospectively collected data for MBD patients (2009–2016) in order to determine which features are most commonly associated with fracture. Patients with primary bone tumors, pathologic fractures at initial presentation, and hematologic malignancies were excluded. A total of 1146 patients comprising 224 pathologic fractures were included. Every patient had at least one Anterior-Posterior X-ray. The clinical data includes patient demographics, tumor biology, all previous radiation and chemotherapy received, multiple pain and function scores, medications and time to fracture or time to death. Each of Mirel's criteria has been further subdivided and recorded for each lesion. We have trained a convolutional neural network (CNN) with X-ray images of 1146 patients with metastatic bone disease of the proximal femur. The digital X-ray data is converted into a matrix representing the color information at each pixel. Our CNN contains five convolutional layers, a fully connected layers of 512 units and a final output layer. As the information passes through successive levels of the network, higher level features are abstracted from the data. This model converges on two fully connected deep neural network layers that output the fracture risk. This prediction is compared to the true outcome, and any errors are back-propagated through the network to accordingly adjust the weights between connections. Methods to improve learning included using stochastic gradient descent with a learning rate of 0.01 and a momentum rate of 0.9. We used average classification accuracy and the average F1 score across test sets to measure model performance. We compute F1 = 2 x (precision x recall)/(precision + recall). F1 is a measure of a test's accuracy in binary classification, in our case, whether a lesion would result in pathologic fracture or not. Five-fold cross validation testing of our fully trained model revealed accurate classification for 88.2% of patients with metastatic bone disease of the proximal femur. The F1 statistic is 0.87. This represents a 24% error reduction from using Mirel's criteria alone to classify the risk of fracture in this cohort. This is the first reported application of convolutional neural networks, a machine learning algorithm, to an important Orthopaedic problem. Our neural network model was able to achieve impressive accuracy in classifying fracture risk of metastatic proximal femur lesions from analysis of X-rays and clinical information. Our future work will aim to validate this algorithm on an external cohort

INTRODUCTION. Mechanical loosening of total hip replacement (THR) is primarily diagnosed using radiographs, which are diagnostically challenging and require review by experienced radiologists and orthopaedic surgeons. Automated tools that assist less-experienced clinicians and mitigate human error can reduce the risk of missed or delayed diagnosis. Thus the purposes of this study were to: 1) develop an automated tool to detect mechanical loosening of THR by training a deep convolutional neural network (CNN) using THR x-rays, and 2) visualize the CNN training process to interpret how it functions. METHODS. A retrospective study was conducted using previously collected imaging data at a single institution with IRB approval. Twenty-three patients with cementless primary THR who underwent revision surgery due to mechanical loosening (either with a loose stem and/or a loose acetabular component) had their hip x-rays evaluated immediately prior to their revision surgery (32 “loose” x-rays). A comparison group was comprised of 23 patients who underwent primary cementless THR surgery with x-rays immediately after their primary surgery (31 “not loose” x-rays). Fig. 1 shows examples of “not loose” and “loose” THR x-ray. DenseNet201-CNN was utilized by swapping the top layer with a binary classifier using 90:10 split-validation [1]. Pre-trained CNN on ImageNet [2] and not pre-trained CNN (initial zero weights) were implemented to compare the results. Saliency maps were implemented to indicate the importance of each pixel of a given x-ray on the CNN's performance [3]. RESULTS. Fig. 2 shows the saliency maps for an example x-ray and the corresponding accuracy of the CNN on the entire validation dataset at different stages of the training for both pre-trained (Fig. 2a) and not pre-trained (Fig. 2b) CNNs. Colored regions in the saliency maps, where red denotes higher relative influence than blue, indicate the most influential regions on the CNN's performance. Pre-trained CNN achieved higher accuracy (87%) on the validation set x-rays than not pre-trained CNN (62%) after 10 epochs. The pre-trained CNN's saliency map at 10 epochs identified significant influence of bone-implant interaction regions on the CNN's performance. This indicates that the CNN is ‘looking’ at the clinically relevant features in the x-rays. The saliency maps also demonstrated that the pre-trained CNN quickly learned where to ‘look’, while the not pre-trained CNN struggles. DISCUSSION. An automated tool to detect mechanical loosening of THR was developed that can potentially assist clinicians with accurate diagnosis. By visualizing the influential regions of the x-ray on the CNN performance, this study shed light into CNN learning process and demonstrated that CNN is ‘looking’ at the clinically relevant features to classify the x-rays. This visualization is crucial to build trust in the automated system by interpreting how it functions to increase the confidence in the application of artificial intelligence to the field of orthopaedics. This study also demonstrated that pre-training CNN can accelerate the learning process and achieve high accuracy even on a small dataset. For any figures or tables, please contact the authors directly

INTRODUCTION. While computational models have been used for many years to contribute to pre-clinical, design phase iterations of total knee replacement implants, the analysis time required has limited the real-time use as required for other applications, such as in patient-specific surgical alignment in the operating room. In this environment, the impact of variation in ligament balance and implant alignment on estimated joint mechanics must be available instantaneously. As neural networks (NN) have shown the ability to appropriately represent dynamic systems, the objective of this preliminary study was to evaluate deep learning to represent the joint level kinetic and kinematic results from a validated finite element lower limb model with varied surgical alignment. METHODS. External hip and ankle boundary conditions were created for a previously-developed finite element lower limb model [1] for step down (SD), deep knee bend (DKB) and gait to best reproduce in-vivo loading conditions as measured on patients with the Innex knee (. orthoload.com. ) (Figure1). These boundary conditions were subsequently used as inputs for the model with a current fixed-bearing total knee replacement to estimate implant-specific kinetics and kinematics during activities of daily living. Implant alignments were varied, including variation of the hip-knee-ankle angle-±3°, the frontal plane joint line −7° to +5°, internal-external femoral rotation ±3°, and the tibial posterior slope 5° and 0°. Through varying these parameters a total of 2464 simulations were completed. A NN was created utilizing the NN toolbox in MATLAB. Sequence data inputs were produced from the alignment and the external boundary conditions for each activity cycle. Sequence outputs for the model were the 6 degree of freedom kinetics and kinematics, totaling 12 outputs. All data was normalized across the entire data set. Ten percent of the simulation runs were removed at random from the training set to be used for validation, leaving 2220 simulations for training and 244 for validation. A nine-layer bi-long short-term memory (LSTM) NN was created to take advantage of bi-LSTM layers ability to learn from past and future data. Training on the network was undertaken using an RMSprop solver until the root mean square error (RMSE) stopped reducing. Evaluation of NN quality was determined by the RMSE of the validation set. RESULTS. The trained NN was able to effectively estimate the validation data. Average RMSE over the kinetics of the validation data set was 140.7N/N∗m while the average RMSE over the kinematics of the validation data set was 4.47mm/deg (Figure 2,3–DKB, Gait shown). It is noted the error may be skewed by the larger magnitude kinetics and kinematics in the DKB activity as the average RMSE for just SD and gait was 85.9N/N∗m and 2.8mm/deg for the kinetics and kinematics, respectively. DISCUSSION. The accuracy of the generated NN indicates its potential for use in real-time modeling, and further work will explore additional changes in post-operative soft-tissue balance as well as scaling to patient-specific geometry

Excessive resident duty hours (RDH) are a recognized issue with implications for physician well-being and patient safety. A major component of the RDH concern is on-call duty. While considerable work has been done to reduce resident call workload, there is a paucity of research in optimizing resident call scheduling. Call coverage is scheduled manually rather than demand-based, which generally leads to over-scheduling to prevent a service gap. Machine learning (ML) has been widely applied in other industries to prevent such issues of a supply-demand mismatch. However, the healthcare field has been slow to adopt these innovations. As such, the aim of this study was to use ML models to 1) predict demand on orthopaedic surgery residents at a level I trauma centre and 2) identify variables key to demand prediction. Daily surgical handover emails over an eight year (2012-2019) period at a level I trauma centre were collected. The following data was used to calculate demand: spine call coverage, date, and number of operating rooms (ORs), traumas, admissions and consults completed. Various ML models (linear, tree-based and neural networks) were trained to predict the workload, with their results compared to the current scheduling approach. Quality of models was determined by using the area under the receiver operator curve (AUC) and accuracy of the predictions. The top ten most important variables were extracted from the most successful model. During training, the model with the highest AUC and accuracy was the multivariate adaptive regression splines (MARS) model, with an AUC of 0.78±0.03 and accuracy of 71.7%±3.1%. During testing, the model with the highest AUC and accuracy was the neural network model, with an AUC of 0.81 and accuracy of 73.7%. All models were better than the current approach, which had an AUC of 0.50 and accuracy of 50.1%. Key variables used by the neural network model were (descending order): spine call duty, year, weekday/weekend, month, and day of the week. This was the first study attempting to use ML to predict the service demand on orthopaedic surgery residents at a major level I trauma centre. Multiple ML models were shown to be more appropriate and accurate at predicting the demand on surgical residents as compared to the current scheduling approach. Future work should look to incorporate predictive models with optimization strategies to match scheduling with demand in order to improve resident well being and patient care

Introduction. Gait laboratory measurement of whole-body kinematics and ground reaction forces during a wide range of activities is frequently performed in joint replacement patient diagnosis, monitoring, and rehabilitation programs. These data are commonly processed in musculoskeletal modeling platforms such as OpenSim and Anybody to estimate muscle and joint reaction forces during activity. However, the processing required to obtain musculoskeletal estimates can be time consuming, requires significant expertise, and thus seriously limits the patient populations studied. Accordingly, the purpose of this study was to evaluate the potential of deep learning methods for estimating muscle and joint reaction forces over time given kinematic data, height, weight, and ground reaction forces for total knee replacement (TKR) patients performing activities of daily living (ADLs). Methods. 70 TKR patients were fitted with 32 reflective markers used to define anatomical landmarks for 3D motion capture. Patients were instructed to perform a range of tasks including gait, step-down and sit-to-stand. Gait was performed at a self-selected pace, step down from an 8” step height, and sit-to-stand using a chair height of 17”. Tasks were performed over a force platform while force data was collected at 2000 Hz and a 14 camera motion capture system collected at 100 Hz. The resulting data was processed in OpenSim to estimate joint reaction and muscle forces in the hip and knee using static optimization. The full set of data consisted of 135 instances from 70 patients with 63 sit-to-stands, 15 right-sided step downs, 14 left-sided step downs, and 43 gait sequences. Two classes of neural networks (NNs), a recurrent neural network (RNN) and temporal convolutional neural network (TCN), were trained to predict activity classification from joint angle, ground reaction force, and anthropometrics. The NNs were trained to predict muscle and joint reaction forces over time from the same input metrics. The 135 instances were split into 100 instances for training, 15 for validation, and 20 for testing. Results. The RNN and TCN yielded classification accuracies of 90% and 100% on the test set. Correlation coefficients between ground truth and predictions from the test set ranged from 0.81–0.95 for the RNN, depending on the activity. Predictions from both NNs were qualitatively assessed. Both NNs were able to effectively learn relationships between the input and output variables. Discussion. The objective of the study was to develop and evaluate deep learning methods for predicting patient mechanics from standard gait lab data. The resulting models classified activities with excellent performance, and showed promise for predicting exact values for loading metrics for a range of different activities. These results indicate potential for real-time prediction of musculoskeletal metrics with application in patient diagnostics and rehabilitation. For any figures or tables, please contact authors directly

Single level discectomy (SLD) is one of the most commonly performed spinal surgery procedures. Two key drivers of their cost-of-care are duration of surgery (DOS) and postoperative length of stay (LOS). Therefore, the ability to preoperatively predict SLD DOS and LOS has substantial implications for both hospital and healthcare system finances, scheduling and resource allocation. As such, the goal of this study was to predict DOS and LOS for SLD using machine learning models (MLMs) constructed on preoperative factors using a large North American database. The American College of Surgeons (ACS) National Surgical and Quality Improvement (NSQIP) database was queried for SLD procedures from 2014-2019. The dataset was split in a 60/20/20 ratio of training/validation/testing based on year. Various MLMs (traditional regression models, tree-based models, and multilayer perceptron neural networks) were used and evaluated according to 1) mean squared error (MSE), 2) buffer accuracy (the number of times the predicted target was within a predesignated buffer), and 3) classification accuracy (the number of times the correct class was predicted by the models). To ensure real world applicability, the results of the models were compared to a mean regressor model. A total of 11,525 patients were included in this study. During validation, the neural network model (NNM) had the best MSEs for DOS (0.99) and LOS (0.67). During testing, the NNM had the best MSEs for DOS (0.89) and LOS (0.65). The NNM yielded the best 30-minute buffer accuracy for DOS (70.9%) and ≤120 min, >120 min classification accuracy (86.8%). The NNM had the best 1-day buffer accuracy for LOS (84.5%) and ≤2 days, >2 days classification accuracy (94.6%). All models were more accurate than the mean regressors for both DOS and LOS predictions. We successfully demonstrated that MLMs can be used to accurately predict the DOS and LOS of SLD based on preoperative factors. This big-data application has significant practical implications with respect to surgical scheduling and inpatient bedflow, as well as major implications for both private and publicly funded healthcare systems. Incorporating this artificial intelligence technique in real-time hospital operations would be enhanced by including institution-specific operational factors such as surgical team and operating room workflow

Introduction. Although total knee arthroplasty (TKA) is generally considered successful, 16–30% of patients are dissatisfied. There are multiple reasons for this, but some of the most frequent reasons for revision are instability and joint stiffness. A possible explanation for this is that the implant alignment is not optimized to ensure joint stability in the individual patient. In this work, we used an artificial neural network (ANN) to learn the relation between a given standard cruciate-retaining (CR) implant position and model-predicted post-operative knee kinematics. The final aim was to find a patient-specific implant alignment that will result in the estimated post-operative knee kinematics closest to the native knee. Methods. We developed subject-specific musculoskeletal models (MSM) based on magnetic resonance images (MRI) of four ex vivo left legs. The MSM allowed for the estimation of secondary knee kinematics (e.g. varus-valgus rotation) as a function of contact, ligament, and muscle forces in a native and post-TKA knee. We then used this model to train an ANN with 1800 simulations of knee flexion with random implant position variations in the ±3 mm and ±3° range from mechanical alignment. The trained ANN was used to find the implant alignment that resulted in the smallest mean-square-error (MSE) between native and post-TKA tibiofemoral kinematics, which we term the dynamic alignment. Results. Dynamic alignment average MSE kinematic differences to the native knees were 1.47 mm (± 0.89 mm) for translations and 2.89° (± 2.83°) for rotations. The implant variations required were in the range of ±3 mm and ±3° from the starting mechanical alignment. Discussion. In this study we showed that the developed tool has the potential to find an implant position that will restore native tibiofemoral kinematics in TKA. The proposed method might also be used with other alignment strategies, such as to optimize implant position towards native ligament strains. If native knee kinematics are restored, a more normal gait pattern can be achieved, which might result in improved patient satisfaction. The small changes required to achieve the dynamic alignment do not represent large modifications that might compromise implant survivorship. Conclusion. Patient-specific implant position predicted with MSM and ANN can restore native knee function in a post-TKA knee with a standard CR implant

Disorders of human joints manifest during dynamic movement, yet no objective tools are widely available for clinicians to assess or diagnose abnormal joint motion during functional activity. Machine learning tools have supported advances in many applications for image interpretation and understanding and have the potential to enable clinically and economically practical methods for objective assessment of human joint mechanics. We performed a study using convolutional neural networks to autonomously segment radiographic images of knee replacements and to determine the potential for autonomous measurement of knee kinematics. The autonomously segmented images provided superior kinematic measurements for both femur and tibia implant components. We believe this is an encouraging first step towards realization of a completely autonomous capability to accurately quantify dynamic joint motion using a clinically and economically practical methodology

Participation in a physical therapy program is considered one of the greatest predictors for successful conservative management of common shoulder disorders, however, adherence to standard exercise protocols is often poor (around 50%) and typically worse for unsupervised home exercise programs. Currently, there are limited tools available for objective measurement of adherence and performance of shoulder rehabilitation in the home setting. The goal of this study was to develop and evaluate the potential for performing home shoulder physiotherapy monitoring using a commercial smartwatch. We hypothesize that shoulder physiotherapy exercises can be classified by analyzing the temporal sequence of inertial sensor outputs from a smartwatch worn on the extremity performing the exercise. Twenty healthy adult subjects with no prior shoulder disorders performed seven exercises from a standard evidence-based rotator cuff physiotherapy protocol: pendulum, abduction, forward elevation, internal/external rotation and trapezius extension with a resistance band, and a weighted bent-over row. Each participant performed 20 repetitions of each exercise bilaterally under the supervision of an orthopaedic surgeon, while 6-axis inertial sensor data was collected at 50 Hz from an Apple Watch. Using the scikit-learn and keras platforms, four supervised learning algorithms were trained to classify the exercises: k-nearest neighbour (k-NN), random forest (RF), support vector machine classifier (SVC), and a deep convolutional recurrent neural network (CRNN). Algorithm performance was evaluated using 5-fold cross-validation stratified first temporally and then by subject. Categorical classification accuracy was above 94% for all algorithms on the temporally stratified cross validation, with the best performance achieved by the CRNN algorithm (99.4± 0.2%). The subject stratified cross validation, which evaluated classifier performance on unseen subjects, yielded lower accuracies scores again with CRNN performing best (88.9 ± 1.6%). This proof-of concept study demonstrates the feasibility of a smartwatch device and machine learning approach to more easily monitor and assess the at-home adherence of shoulder physiotherapy exercise protocols. Future work will focus on translation of this technology to the clinical setting and evaluating exercise classification in shoulder disorder populations

Introduction. Automated identification of arthroplasty implants could aid in pre-operative planning and is a task which could be facilitated through artificial intelligence (AI) and deep learning. The purpose of this study was to develop and test the performance of a deep learning system (DLS) for automated identification and classification of knee arthroplasty (KA) on radiographs. Methods. We collected 237 AP knee radiographs with equal proportions of native knees, total KA (TKA), and unicompartmental KA (UKA), as well as 274 radiographs with equal proportions of Smith & Nephew Journey and Zimmer NexGen TKAs. Data augmentation was used to increase the number of images available for DLS development. These images were used to train, validate, and test deep convolutional neural networks (DCNN) to 1) detect the presence of TKA; 2) differentiate between TKA and UKA; and 3) differentiate between the 2 TKA models. Receiver operating characteristic (ROC) curves were generated with area under the curve (AUC) calculated to assess test performance. Results. The DCNNs trained to detect KA and to distinguish between TKA and UKA both achieved AUC of 1. In both cases, heatmap analysis demonstrated appropriate emphasis of the KA components in decision-making. The DCNN trained to distinguish between the 2 TKA models also achieved AUC of 1. Heatmap analysis of this DCNN showed emphasis of specific unique features of the TKA model designs for decision making, such as the anterior flange shape of the Zimmer NexGen TKA (Figure 1) and the tibial baseplate/stem shape of the Smith & Nephew Journey TKA (Figure 2). Conclusion. DCNNs can accurately identify presence of TKA and distinguish between specific designs. The proof-of-concept of these DCNNs may set the foundation for DCNNs to identify other prosthesis models and prosthesis-related complications. For any figures or tables, please contact the authors directly

Objective. Emergence of low-cost wearable systems has permitted extended data collection for unsupervised subject monitoring. Recognizing individual activities performed during these sessions gives context to recorded data and is an important first step towards automated motion analysis. Convolutional neural networks (CNNs) have been used with great success to detect patterns of pixels in images for object detection and recognition in many different applications. This work proposes a novel image encoding scheme to create images from time-series activity data and uses CNNs to accurately classify 13 daily activities performed by instrumented subjects. Methods. Twenty healthy subjects were instrumented with a previously developed wearable sensor system consisting of four inertial sensors mounted above and below each knee. Each subject performed eight static and five dynamic activities: standing, sitting in a chair/cross-legged, kneeling on left/right/both knees, squatting, laying, walking/running, biking and ascending/descending stairs. Data from each sensor were synchronized, windowed, and encoded as images using a novel encoding scheme. Two CNNs were designed and trained to classify the encoded images of both static and dynamic activities separately. Network performance was evaluated using twenty iterations of a leave-one-out validation process where a single subject was left out for test data to estimate performance on future unseen subjects. Results. Using 19 subjects for training and a single subject left out for testing per iteration, the average accuracy observed when classifying the eight static activities was 98.0% ±2.9%. Accuracy dropped to 89.3% ±10.6% when classifying all dynamic activities using a separate model with the same evaluation process. Ascending/descending stairs, walking/running, and sitting on a chair/squatting were most commonly misclassified. Conclusions. Previous related work on activity recognition using accelerometer and/or gyroscope raw signals fails to provide sufficient data to distinguish static activities. The proposed method operating on lower limb orientations has classified eight static activities with exceptional accuracy when tested on unseen subject data. High accuracy was also observed when classifying dynamic activities despite the similarity of the activities performed and the expected variance of individuals’ gait. Accuracy reported in existing literature classifying comparable activities from other wearable sensor systems ranges between 27.84% to 84.52% when tested using a similar leave-one-subject-out validation strategy[1]. It is expected that incorporating these trained models into the previously developed wearable system will permit activity classification on unscripted instrumented activity data for more contextual motion analysis

Introduction. Real-time tracking of surgical tools has applications for assessment of surgical skill and OR workflow. Accordingly, efforts have been devoted to the development of low-cost systems that track the location of surgical tools in real-time without significant augmentation to the tools themselves. Deep learning methodologies have recently shown success in a multitude of computer vision tasks, including object detection, and thus show potential for the application of surgical tool tracking. The objective of the current study was to develop and evaluate a deep learning-based computer vision system using a single camera for the detection and pose estimation of multiple surgical tools routinely used in both knee and hip arthroplasty. Methods. A computer vision system was developed for the detection and 6-DoF pose estimation of two surgical tools (mallet and broach handle) using only RGB camera frames. The deep learning approach consisted of a single convolutional neural network (CNN) for object detection and semantic key point prediction, as well as an optimization step to place prior known geometries into the local camera coordinate system. Inference on a camera frame with size of 256-by-352 took 0.3 seconds. The object detection component of the system was evaluated on a manually-annotated stream of video frames. The accuracy of the system was evaluated by comparing pose (position and orientation) estimation of a tool with the ground truth pose as determined using three retroreflective markers placed on each tool and a 14 camera motion capture system (Vicon, Centennial CO). Markers placed on the tool were transformed into the local camera coordinate system and compared to estimated location. Results. Detection accuracy determined from frame-wise confusion matrices was 82% and 95% for the mallet and broach handle, respectively. Object detection and key point predictions were qualitatively assessed. Marker error resulting from pose estimation was as little as 1.3 cm for the evaluation scenes. Pose estimation of the tools from each evaluation scene was also qualitatively assessed. Discussion. The proposed computer vision system combined CNNs with optimization to estimate the 6-DoF pose of surgical tools from only RGB camera frames. The system's object detection component performed on par with state-of-the-art object detection literature and the pose estimation error was efficiently computed from CNN predictions. The current system has implications for surgical skill assessment and operations based research to improve operating room efficiency. However, future development is needed to make improvements to the object detection and key point prediction components of the system, in order to minimize potential pose error. Nominal marker errors of 1.3 cm demonstrate the potential of this system to yield accurate pose estimates of surgical tools. For any figures or tables, please contact authors directly

Introduction. Gait analysis systems have enjoyed increasing usage and have been validated to provide highly accurate assessments for range of motion. Size, cost, need for marker placement and need for complex data processing have remained limiting factors in uptake outside of what remains predominantly large research institutions. Progress and advances in deep neural networks, trained on millions of clinically labelled datasets, have allowed the development of a computer vision system which enables assessment using a handheld smartphone with no markers and accurate range of motion for knee during flexion and extension. This allows clinicians and therapists to objectively track progress without the need for complex and expensive equipment or time-consuming analysis, which was concluded to be lacking during a recent systematic review of existing applications. Method. A smartphone based computer vision system was assessed for accuracy with a gold standard comparison using a validated ‘traditional’ infra-red motion capture system which had a defined calibrated accuracy of 0.1degrees. A total of 22 subjects were assessed simultaneously using both the computer vision smartphone application and the standard motion capture system. Assessment of the handheld system was made by comparison to the motion capture system for knee flexion and extension angles through a range of motion with a simulated fixed-flexion deformity which prevented full extension to assess the accuracy of the system, repeating movements ten times. The peak extension angles and also numerous discrete angle measurements were compared between the two systems. Repeatability was assessed by comparing several sequential cycles of flexion/extension and comparison of the maximum range of motion in normal knees and in those with a simulated fixed-flexion deformity. In addition, discrete angles were also measured on both legs of three cadavers with both skin and then bone implanted fiducial markers for ground truth reliability accounting for skin movement. Data was processed quickly through an automated secure cloud system. Results. The smartphone application was found to be accurate to 1.47±1.05 degrees through a full range of motion and 1.75±1.56 degrees when only peak extension angles were compared, demonstrating excellent reliability and repeatability. The cadaveric studies despite limitations which will be discussed still showed excellent accuracy with average errors as low as 0.29 degrees for individual angles and 4.09 degrees for an average error in several measurement. Conclusion. This novel solution offers for the first time a way to objectively measure knee range of motion using a markerless handheld device and enables tracking through a range of assessments with proven accuracy and reliability even accounting for traditional issues with the previous marker based systems. Repeatability for both computer vision and motion capture have greater extrinsic than intrinsic error, particularly with marker placement - another benefit of a markerless system. Clinical applications include pre-operative assessment and post-operative follow-up, paired with surgical planning (including with robots) and remote monitoring after knee surgery, with outcomes guiding treatment and rehabilitation and leading to reduced need for manipulation under anaesthesia and greater satisfaction

Pedicle screw fixation is a technically demanding procedure with potential difficulties and reoperation rates are currently on the order of 11%. The most common intraoperative practice for position assessment of pedicle screws is biplanar fluoroscopic imaging that is limited to two- dimensions and is associated to low accuracies. We have previously introduced a full-dimensional position assessment framework based on registering intraoperative X-rays to preoperative volumetric images with sufficient accuracies. However, the framework requires a semi-manual process of pedicle screw segmentation and the intraoperative X-rays have to be taken from defined positions in space in order to avoid pedicle screws' head occlusion. This motivated us to develop advancements to the system to achieve higher levels of automation in the hope of higher clinical feasibility. In this study, we developed an automatic segmentation and X-ray adequacy assessment protocol. An artificial neural network was trained on a dataset that included a number of digitally reconstructed radiographs representing pedicle screw projections from different points of view. This model was able to segment the projection of any pedicle screw given an X-ray as its input with accuracy of 93% of the pixels. Once the pedicle screw was segmented, a number of descriptive geometric features were extracted from the isolated blob. These segmented images were manually labels as ‘adequate’ or ‘not adequate’ depending on the visibility of the screw axis. The extracted features along with their corresponding labels were used to train a decision tree model that could classify each X-ray based on its adequacy with accuracies on the order of 95%. In conclusion, we presented here a robust, fast and automated pedicle screw segmentation process, combined with an accurate and automatic algorithm for classifying views of pedicle screws as adequate or not. These tools represent a useful step towards full automation of our pedicle screw positioning assessment system

Introduction. Most of patients with unilateral hip disease shows muscle volume atrophy of pelvis and thigh in the affected side because of pain and disuse, resulting in reduced muscle weakness and limping. However, it is unclear how the muscle atrophy correlated with muscle strength in the patient with hip disorders. A previous study have demonstrated that the volume of the gluteus medius correlated with the muscle strength by volumetric measurement using 3 dimensional computed tomography (3D-CT) data, however, muscles influence each other during motions and there is no reports focusing on the relationship between some major muscles of pelvis and thigh including gluteus maximus, gluteus medius, iliopsoas and quadriceps and muscle strength in several hip and knee motions. Therefore, the purpose of the present study is to evaluate the relationship between muscle volumetric atrophy of major muscles of pelvis and thigh and muscle strength in flexion, extension and abduction of hip joints and extension of knee joint before surgery in patients with unilateral hip disease. Material and Methods. The subjects were 38 patients with unilateral hip osteoarthritis, who underwent hip joint surgery. They all underwent preoperative computed tomography (CT) for preoperative planning. There were 6 males and 32 females with average age 59.5 years old. Before surgery, isometric muscle strength in hip flexion, hip extension, hip abduction and knee extension were measured using a hand held dynamometer (µTas F-1, ANIMA Japan). Major muscles including gluteus maximus, gluteus medius, iliopsoas and quadriceps were automatically extracted from the preoperative CT using convolutional neural networks (CNN) and were corrected manually by the experienced surgeon. The muscle volumetric atrophy ratio was defined as the ratio of muscle volume of the affected side to that of the unaffected side. The muscle weakness ratio was defined as the ratio of muscle strength of the affected side to that of the unaffected side. The correlation coefficient between the muscle atrophy ratio and the muscle weakness ratio of each muscle were calculated. Results. The average muscle atrophy ratio was 84.5% (63.5%–108.2%) in gluteus maximus, 86.6% (65.5%–112.1%) in gluteus medius, 81.0% (22.1%–130.8%) in psoas major, and 91.0% (63.8%–127.0%) in quadriceps. The average muscle strength ratio was 71.5% (0%–137.5%) in hip flexion, 88.1% (18.8%–169.6%) in hip abduction, 78.6% (21.9%–130.1%) in hip extension and 84.3% (13.1%–122.8%) in knee extension. The correlation coefficient between the muscle atrophy and the ratio of each muscle strength between the affected and unaffected side were shown in Table 1. Conclusion. In conclusion, the muscle atrophy of gluteus medius muscle, psoas major muscle and quadriceps muscle significantly correlated with the muscle weakness in hip flexion. The muscle atrophy of psoas major muscle and quadriceps muscle also significantly correlated with the muscle weakness in knee extension. There were no significant correlation between the muscle atrophy and the muscle weakness in hip extension and abduction

The success of a cementless Total Hip Arthroplasty (THA) depends not only on initial micromotion, but also on long-term failure mechanisms, e.g., implant-bone interface stresses and stress shielding. Any preclinical investigation aimed at designing femoral implant needs to account for temporal evolution of interfacial condition, while dealing with these failure mechanisms. The goal of the present multi-criteria optimization study was to search for optimum implant geometry by implementing a novel machine learning framework comprised of a neural network (NN), genetic algorithm (GA) and finite element (FE) analysis. The optimum implant model was subsequently evaluated based on evolutionary interface conditions. The optimization scheme of our earlier study [1] has been used here with an additional inclusion of an NN to predict the initial fixation of an implant model. The entire CAD based parameterization technique for the implant was described previously [1]. Three objective functions, the first two based on proximal resorbed Bone Mass Fraction (BMF) [1] and implant-bone interface failure index [1], respectively, and the other based on initial micromotion, were formulated to model the multi-criteria optimization problem. The first two objective functions, e.g., objectives f1 and f2, were calculated from the FE analysis (Ansys), whereas the third objective (f3) involved an NN developed for the purpose of predicting the post-operative micromotion based on the stem design parameters. Bonded interfacial condition was used to account for the effects of stress shielding and interface stresses, whereas a set of contact models were used to develop the NN for faster prediction of post-operative micromotion. A multi-criteria GA was executed up to a desired number of generations for optimization (Fig. 1). The final trade-off model was further evaluated using a combined remodelling and bone ingrowth simulation based on an evolutionary interface condition [2], and subsequently compared with a generic TriLock implant. The non-dominated solutions obtained from the GA execution were interpolated to determine the 3D nature of the Pareto-optimal surface (Fig. 2). The effects of all failure mechanisms were found to be minimized in these optimized solutions (Fig. 2). However, the most compromised solution, i.e., the trade-off stem geometry (TSG), was chosen for further assessment based on evolutionary interfacial condition. The simulation-based combined remodelling and bone ingrowth study predicted a faster ingrowth for TSG as compared to the generic design. The surface area with post-operative (i.e., iteration 1) ingrowth was found to be ∼50% for the TSG, while that for the TriLock model was ∼38% (Fig. 3). However, both designs predicted similar long-term ingrowth (∼89% surface area). The long-term proximal bone resorption (upto lesser trochanter) was found to be ∼30% for the TSG, as compared to ∼37% for the TriLock model. The TSG was found to be bone-preserving with prominent frontal wedge and rectangular proximal section for better rotational stability; features present in some recent designs. The optimization scheme, therefore, appears to be a quick and robust preclinical assessment tool for cementless femoral implant design. To view tables/figures, please contact authors directly

Aims

The principles of evidence-based medicine (EBM) are the foundation of modern medical practice. Surgeons are familiar with the commonly used statistical techniques to test hypotheses, summarize findings, and provide answers within a specified range of probability. Based on this knowledge, they are able to critically evaluate research before deciding whether or not to adopt the findings into practice. Recently, there has been an increased use of artificial intelligence (AI) to analyze information and derive findings in orthopaedic research. These techniques use a set of statistical tools that are increasingly complex and may be unfamiliar to the orthopaedic surgeon. It is unclear if this shift towards less familiar techniques is widely accepted in the orthopaedic community. This study aimed to provide an exploration of understanding and acceptance of AI use in research among orthopaedic surgeons.

Aims

The use of technology to assess balance and alignment during total knee surgery can provide an overload of numerical data to the surgeon. Meanwhile, this quantification holds the potential to clarify and guide the surgeon through the surgical decision process when selecting the appropriate bone recut or soft tissue adjustment when balancing a total knee. Therefore, this paper evaluates the potential of deploying supervised machine learning (ML) models to select a surgical correction based on patient-specific intra-operative assessments.