Kunze, Kyle N.

AbstractPurposeTo systematically review the literature regarding machine learning in leg length discrepancy (LLD) and to provide insight into the most relevant manuscripts on this topic in order to highlight the importance and future clinical implications of machine learning in the diagnosis and treatment of LLD.MethodsA systematic electronic search was conducted using PubMed, OVID/Medline and Cochrane libraries in accordance with Preferred Reporting Items for Systematic Review and Meta‐Analysis guidelines. Two observers independently screened the abstracts and titles of potential articles.ResultsA total of six studies were identified in the search. All measurements were calculated using standardized anterior‐posterior long‐leg radiographs. Five (83.3%) of the studies used measurements of the femoral length, tibial length and leg length to assess LLD, whereas one (16.6%) study used the iliac crest height difference to quantify LLD. The deep learning models showed excellent reliability in predicting all length measurements with intraclass correlation coefficients ranging from 0.98 to 1.0 and mean absolute error (MAE) values ranging from 0.11 to 0.45 cm. Three studies reported measurements of LLD, and the convolutional neural network model showed the lowest MAE of 0.13 cm in predicting LLD.ConclusionsMachine learning models are effective and efficient in determining LLD. Implementation of these models may reduce cost, improve efficiency and lead to better overall patient outcomes.Clinical RelevanceThis review highlights the potential of deep learning (DL) algorithms for accurate and reliable measurement of lower limb length and leg length discrepancy (LLD) on long‐leg radiographs. The reported mean absolute error and intraclass correlation coefficient values indicate that the performance of the DL models was comparable to that of radiologists, suggesting that DL‐based assessments could potentially be used to automate the measurement of lower limb length and LLD in clinical practice.Level of EvidenceLevel IV.

Purpose To assess the ability of ChatGPT‐4, an automated Chatbot powered by artificial intelligence, to answer common patient questions concerning the Latarjet procedure for patients with anterior shoulder instability and compare this performance with Google Search Engine. Methods Using previously validated methods, a Google search was first performed using the query “Latarjet.” Subsequently, the top 10 frequently asked questions (FAQs) and associated sources were extracted. ChatGPT‐4 was then prompted to provide the top 10 FAQs and answers concerning the procedure. This process was repeated to identify additional FAQs requiring discrete‐numeric answers to allow for a comparison between ChatGPT‐4 and Google. Discrete, numeric answers were subsequently assessed for accuracy on the basis of the clinical judgment of 2 fellowship‐trained sports medicine surgeons who were blinded to search platform. Results Mean (± standard deviation) accuracy to numeric‐based answers was 2.9 ± 0.9 for ChatGPT‐4 versus 2.5 ± 1.4 for Google ( P = .65). ChatGPT‐4 derived information for answers only from academic sources, which was significantly different from Google Search Engine ( P = .003), which used only 30% academic sources and websites from individual surgeons (50%) and larger medical practices (20%). For general FAQs, 40% of FAQs were found to be identical when comparing ChatGPT‐4 and Google Search Engine. In terms of sources used to answer these questions, ChatGPT‐4 again used 100% academic resources, whereas Google Search Engine used 60% academic resources, 20% surgeon personal websites, and 20% medical practices ( P = .087). Conclusions ChatGPT‐4 demonstrated the ability to provide accurate and reliable information about the Latarjet procedure in response to patient queries, using multiple academic sources in all cases. This was in contrast to Google Search Engine, which more frequently used single‐surgeon and large medical practice websites. Despite differences in the resources accessed to perform information retrieval tasks, the clinical relevance and accuracy of information provided did not significantly differ between ChatGPT‐4 and Google Search Engine. Clinical Relevance Commercially available large language models (LLMs), such as ChatGPT‐4, can perform diverse information retrieval tasks on‐demand. An important medical information retrieval application for LLMs consists of the ability to provide comprehensive, relevant, and accurate information for various use cases such as investigation about a recently diagnosed medical condition or procedure. Understanding the performance and abilities of LLMs for use cases has important implications for deployment within health care settings.

Purpose To develop a deep learning model for the detection of Segond fractures on anteroposterior (AP) knee radiographs and to compare model performance to that of trained human experts. Methods AP knee radiographs were retrieved from the Hospital for Special Surgery ACL Registry, which enrolled patients between 2009 and 2013. All images corresponded to patients who underwent anterior cruciate ligament reconstruction by 1 of 23 surgeons included in the registry data. Images were categorized into 1 of 2 classes based on radiographic evidence of a Segond fracture and manually annotated. Seventy percent of the images were used to populate the training set, while 20% and 10% were reserved for the validation and test sets, respectively. Images from the test set were used to compare model performance to that of expert human observers, including an orthopaedic surgery sports medicine fellow and a fellowship‐trained orthopaedic sports medicine surgeon with over 10 years of experience. Results A total of 324 AP knee radiographs were retrieved, of which 34 (10.4%) images demonstrated evidence of a Segond fracture. The overall mean average precision (mAP) was 0.985, and this was maintained on the Segond fracture class (mAP = 0.978, precision = 0.844, recall = 1). The model demonstrated 100% accuracy with perfect sensitivity and specificity when applied to the independent testing set and the ability to meet or exceed human sensitivity and specificity in all cases. Compared to an orthopaedic surgery sports medicine fellow, the model required 0.3% of the total time needed to evaluate and classify images in the independent test set. Conclusions A deep learning model was developed and internally validated for Segond fracture detection on AP radiographs and demonstrated perfect accuracy, sensitivity, and specificity on a small test set of radiographs with and without Segond fractures. The model demonstrated superior performance compared with expert human observers. Clinical Relevance Deep learning can be used for automated Segond fracture identification on radiographs, leading to improved diagnosis of easily missed concomitant injuries, including lateral meniscus tears. Automated identification of Segond fractures can also enable large‐scale studies on the incidence and clinical significance of these fractures, which may lead to improved management and outcomes for patients with knee injuries.

Purpose To (1) review definitions and concepts necessary to interpret applications of deep learning (DL; a domain of artificial intelligence that leverages neural networks to make predictions on media inputs such as images) and (2) identify knowledge and translational gaps in the literature to provide insight into specific areas for improvement as adoption of this technology continues. Methods A comprehensive search of the literature was performed in December 2023 for articles regarding the use of DL in sports medicine. For each study, information regarding the joint of focus, specific anatomic structure/pathology to which DL was applied, imaging modality utilized, source of images used for model training and testing, data set size, model performance, and whether the DL model was externally validated was recorded. A numerical scale was used to rate each DL model’s clinical impact, with 1 corresponding to proof‐of‐concept studies with little to no direct clinical impact and 5 corresponding to practice‐changing clinical impact and ready for clinical deployment. Results Fifty‐five studies were identified, all of which were published within the past 5 years, while 82% were published within the past 3 years. Of the DL models identified, 84% were developed for classification tasks, 9% for automated measurements, and 7% for segmentation. A total of 62% of studies utilized magnetic resonance imaging as the imaging modality, 25% radiographs, and 7% ultrasound, while 1 study each used computed tomography, arthroscopic images, or arthroscopic video. Sixty‐five percent of studies focused on the detection of tears (anterior cruciate ligament [ACL], rotator cuff [RC], and meniscus). The diagnostic performance of ACL tears, as determined by the area under the receiver operator curve (AUROC), ranged from 0.81 to 0.99 for ACL tears (excellent to near perfect), 0.83 to 0.94 for RC tears (excellent), and from 0.75 to 0.96 for meniscus tears (acceptable to excellent). In addition, 3 studies focused on detection of cartilage lesions had AUROC ranging from 0.90 to 0.92 (excellent performance). However, only 4 (7%) studies externally validated their models, suggesting that they may not be generalizable or may not perform well when applied to populations other than that used to develop the model. Finally, the mean clinical impact score was 2 (range, 1‐3) on scale of 1 to 5, corresponding to limited clinical applicability. Conclusions DL models in orthopaedic sports medicine show generally excellent performance (high internal validity) but require external validation to facilitate clinical deployment. In addition, current models have low clinical applicability and fail to advance the field due to a focus on routine tasks and a narrow conceptual framework. Level of Evidence Level IV, scoping review of Level I to IV studies.

Background: Treatment of ulnar collateral ligament (UCL) tears with suture tape augmentation has gained interest given preliminary reports of favorable biomechanical characteristics. No study to date has quantitatively assessed the biomechanical effects of multiple augmentation techniques relative to the native UCL. Purpose: To perform a systematic review and meta-analysis of controlled laboratory studies to assess and comparatively rank biomechanical effects of UCL repair or reconstruction with or without augmentation. Study Design: Systematic review and meta-analysis; Level of evidence, 4. Methods: PubMed, OVID/Medline, and Cochrane databases were queried in January 2023. A frequentist network meta-analytic approach was used to perform mixed-treatment comparisons of UCL repair and reconstruction techniques with and without augmentation, with the native UCL as the reference condition. Pooled treatment estimates were quantified under the random-effects assumption. Competing treatments were ranked in the network meta-analysis by using point estimates and standard errors to calculate P scores (greater P score indicates superiority of treatment for given outcome). Results: Ten studies involving 206 elbow specimens in which a distal UCL tear was simulated were included. UCL reconstruction with suture tape augmentation (AugRecon) restored load to failure to a statistically noninferior magnitude (mean difference [MD], –1.99 N·m; 95% CI, –10.2 to 6.2 N·m; P = .63) compared with the native UCL. UCL reconstruction (Recon) (MD, –12.7 N·m; P < .001) and UCL repair with suture tape augmentation (AugRepair) (MD, –14.8 N·m; P < .001) were both statistically inferior to the native UCL. The AugRecon condition conferred greater load to failure compared with Recon ( P < .001) and AugRepair ( P = .002) conditions. AugRecon conferred greater torsional stiffness relative to all other conditions and was not statistically different from the native UCL (MD, 0.32 N·m/deg; 95% CI, –0.30 to 0.95 N·m/deg; P = .31). Medial ulnohumeral gapping was not statistically different for the AugRepair (MD, 0.30 mm; 95% CI, –1.22 to 1.82 mm; P = .70), AugRecon (MD, 0.57 mm; 95% CI, –0.70 to 1.84 mm; P = .38), or Recon (MD, 1.02 mm; 95% CI, –0.02 to 2.05 mm; P = .055) conditions compared with the native UCL. P-score analysis indicated that AugRecon was the most effective treatment for increasing ultimate load to failure and torsional stiffness, whereas AugRepair was the most effective for minimizing medial gapping. Conclusion: AugRecon restored load to failure and torsional stiffness most similar to the parameters of the native UCL, whereas Recon and AugRepair did not restore the same advantageous properties at time zero. Medial ulnohumeral gapping during a valgus load was minimized by all 3 treatments. Based on network interactions, AugRecon was the superior treatment approach for restoring important biomechanical features of the UCL at time zero that are jeopardized during a complete distal tear.

AbstractMean, median, and mode are among the most basic and consistently used measures of central tendency in statistical analysis and are crucial for simplifying data sets to a single value. However, there is a lack of understanding of when to use each metric and how various factors can impact these values. The aim of this article is to clarify some of the confusion related to each measure and explain how to select the appropriate metric for a given data set. The authors present this work as an educational resource, ensuring that these common statistical concepts are better understood throughout the Orthopedic research community.

AbstractSupervised learning is the most common form of machine learning utilized in medical research. It is used to predict outcomes of interest or classify positive and/or negative cases with a known ground truth. Supervised learning describes a spectrum of techniques, ranging from traditional regression modeling to more complex tree boosting, which are becoming increasingly prevalent as the focus on “big data” develops. While these tools are becoming increasingly popular and powerful, there is a paucity of literature available that describe the strengths and limitations of these different modeling techniques. Typically, there is no formal training for health care professionals in the use of machine learning models. As machine learning applications throughout medicine increase, it is important that physicians and other health care professionals better understand the processes underlying application of these techniques. The purpose of this study is to provide an overview of commonly used supervised learning techniques with recent case examples within the orthopedic literature. An additional goal is to address disparities in the understanding of these methods to improve communication within and between research teams.

Purpose To develop a machine learning algorithm to predict total charges after ambulatory hip arthroscopy and create a risk‐adjusted payment model based on patient comorbidities. Methods A retrospective review of the New York State Ambulatory Surgery and Services database was performed to identify patients who underwent elective hip arthroscopy between 2015 and 2016. Features included in initial models consisted of patient characteristics, medical comorbidities, and procedure‐specific variables. Models were generated to predict total charges using 5 algorithms. Model performance was assessed by the root‐mean‐square error, root‐mean‐square logarithmic error, and coefficient of determination. Global variable importance and partial dependence curves were constructed to show the impact of each input feature on total charges. For performance benchmarking, the best candidate model was compared with a multivariate linear regression using the same input features. Results A total of 5,121 patients were included. The median cost after hip arthroscopy was $19,720 (interquartile range, $12,399‐$26,439). The gradient‐boosted ensemble model showed the best performance (root‐mean‐square error, $3,800 [95% confidence interval, $3,700‐$3,900]; logarithmic root‐mean‐square error, 0.249 [95% confidence interval, 0.24‐0.26]; R 2 = 0.73). Major cost drivers included total hours in facility less than 12 or more than 15, longer procedure time, performance of a labral repair, age younger than 30 years, Elixhauser Comorbidity Index (ECI) of 1 or greater, African American race, residence in extreme urban and rural areas, and higher household and neighborhood income. Conclusions The gradient‐boosted ensemble model effectively predicted total charges after hip arthroscopy. Few modifiable variables were identified other than anesthesia type; nonmodifiable drivers of total charges included duration of care less than 12 hours or more than 15 hours, operating room time more than 100 minutes, age younger than 30 years, performance of a labral repair, and ECI greater than 0. Stratification of patients based on the ECI highlighted the increased financial risk borne by physicians via flat reimbursement schedules given variable degrees of comorbidities. Level of Evidence Level III, retrospective cohort study.

Purpose To develop and internally validate a machine‐learning algorithm to reliably predict cost after anterior cruciate ligament reconstruction (ACLR). Methods A retrospective review of the New York State Ambulatory Surgery and Services database was performed to identify patients who underwent elective ACLR from 2015 to 2016. Features included in initial models consisted of patient characteristics (age, sex, insurance status, income, medical comorbidities as classified by the Clinical Classifications Software diagnosis code) as well as intraoperative variables (type of anesthesia and procedure‐specific factors). Models were generated to predict total charges using 4 algorithms: random forest, extreme gradient boost, elastic net penalized regression, and support vector machines with radial kernels. Training was performed with 10‐fold cross‐validation followed by internal validation via 0.632 bootstrapping. Model discriminative performance was assessed by area under the receiver operating characteristic curve, calibration, and the Brier score. Decision curve analysis was performed to demonstrate the net benefit of using the final model in practice. Results In total, 7,311 patients undergoing ambulatory ACLR were included. The random forest model demonstrated the best performance assessed via internal validation (area under the curve = 0.85), calibration, and the Brier score (0.208). Cost incurred was influenced by anesthesia type, operating room time, and number of chronic comorbidities. Decision curve analysis revealed a net benefit for use of the random forest model and the model was integrated into a web‐based open‐access application. Conclusions The random forest model predicted cost after ambulatory ACLR using a large, statewide database with good performance. The top variables found to predict increased charges were general anesthesia, operating room time, meniscal repair, self‐pay insurance, patient neighborhood characteristics, and number of chronic conditions. Level of Evidence III, retrospective cohort study.

Purpose (1) To determine the area of posterior cruciate ligament (PCL) insertion sites on the lateral wall of the medial femoral condyle (LWMFC) that demonstrates the least amount of length change through full range of motion (ROM) and (2) to identify a range of flexion that would be favorable for graft tensioning for single‐bundle (SB) and double‐bundle (DB) PCL reconstruction. Methods Six fresh‐frozen cadaveric knees were obtained. Three‐dimensional computed tomography point‐cloud models were obtained from 0° to 135°. A point grid was placed on the LWMFC and the tibial PCL facet. Intra‐articular length was calculated for each point on the femur to the tibia at all flexion angles and grouped to represent areas for bone tunnels of SB and DB PCLR. Normalized length changes were evaluated. Results Femoral tunnel location and angle of graft fixation were significant contributors to mean, minimum, and maximum normalized length of the PCL (all p < .001). Tibial tunnel location was not significant in any case (all p < .22). A femoral tunnel in the location of the posteromedial bundle of the PCL resulted in the least length change at all tibial positions (maximum change 13%). Fixation of the anterolateral bundle in extension or at 30° flexion resulted in significant overconstraint of the PCL graft. The femoral tunnel location for a SB PCLR resulted in significant laxity at lower ranges of flexion. Conclusion PCL length was significantly dependent on femoral tunnel position and angle of fixation, whereas tibial tunnel position did not significantly contribute to observed differences. All PCL grafts demonstrated anisometry, with the anterolateral bundle being more anisometric than the posteromedial bundle. For DB PCLR, the posteromedial bundle demonstrated the highest degree of isometry throughout ROM, although no area of the LWMFC was truly isometric. The anterolateral bundle should be fixed at 90° to avoid overconstraint, and SB PCLR demonstrated significant laxity at lower ranges of flexion. Clinical Relevance Surgeons can apply the results of this investigation to surgical planning in PCLR to optimize isometry, which may ultimately reduce graft strain and the risk of graft failure. Additionally, DB PCLR demonstrated superiority compared with SB PCLR regarding graft isometry, as significant laxity was encountered at lower ranges of flexion in SB PCLRs. Fixation of the ALB at 90° flexion should be performed to avoid overconstraint in knee extension.