Artificial Intelligence for Clinically Meaningful Outcome Prediction in Orthopedic Research: Current Applications and Limitations.

PURPOSE OF REVIEW: Patient-reported outcome measures (PROM) play a critical role in evaluating the success of treatment interventions for musculoskeletal conditions. However, predicting which patients will benefit from treatment interventions is complex and influenced by a multitude of factors. Artificial intelligence (AI) may better anticipate the propensity to achieve clinically meaningful outcomes through leveraging complex predictive analytics that allow for personalized medicine. This article provides a contemporary review of current applications of AI developed to predict clinically significant outcome (CSO) achievement after musculoskeletal treatment interventions. RECENT FINDINGS: The highest volume of literature exists in the subspecialties of total joint arthroplasty, spine, and sports medicine, with only three studies identified in the remaining orthopedic subspecialties combined. Performance is widely variable across models, with most studies only reporting discrimination as a performance metric. Given the complexity inherent in predictive modeling for this task, including data availability, data handling, model architecture, and outcome selection, studies vary widely in their methodology and results. Importantly, the majority of studies have not been externally validated or demonstrate important methodological limitations, precluding their implementation into clinical settings. A substantial body of literature has accumulated demonstrating variable internal validity, limited scope, and low potential for clinical deployment. The majority of studies attempt to predict the MCID-the lowest bar of clinical achievement. Though a small proportion of models demonstrate promise and highlight the utility of AI, important methodological limitations need to be addressed moving forward to leverage AI-based applications for clinical deployment.

Artificial intelligence for automated identification of total shoulder arthroplasty implants.

BACKGROUND: Accurate and rapid identification of implant manufacturer and model is critical in the evaluation and management of patients requiring revision total shoulder arthroplasty (TSA). Failure to correctly identify implant designs in these circumstances may lead to delay in care, unexpected intraoperative challenges, increased morbidity, and excess health care costs. Deep learning (DL) permits automated image processing and holds the potential to mitigate such challenges while improving the value of care rendered. The purpose of this study was to develop an automated DL algorithm to identify shoulder arthroplasty implants from plain radiographs. METHODS: A total of 3060 postoperative images from patients who underwent TSA between 2011 and 2021 performed by 26 fellowship-trained surgeons at 2 independent tertiary academic hospitals in the Pacific Northwest and Mid-Atlantic Northeast were included. A DL algorithm was trained using transfer learning and data augmentation to classify 22 different reverse TSA and anatomic TSA prostheses from 8 implant manufacturers. Images were split into training and testing cohorts (2448 training and 612 testing images). Optimized model performance was assessed using standardized metrics including the multiclass area under the receiver operating characteristic curve (AUROC) and compared with a reference standard of implant data from operative reports. RESULTS: The algorithm classified implants at a mean speed of 0.079 seconds (±0.002 seconds) per image. The optimized model discriminated between 8 manufacturers (22 unique implants) with AUROCs of 0.994-1.000, accuracy of 97.1%, and sensitivities between 0.80 and 1.00 on the independent testing set. In the subset of single-institution implant predictions, a DL model identified 6 specific implants with AUROCs of 0.999-1.000, accuracy of 99.4%, and sensitivity >0.97 for all implants. Saliency maps revealed key differentiating features across implant manufacturers and designs recognized by the algorithm for classification. CONCLUSION: A DL model demonstrated excellent accuracy in identifying 22 unique TSA implants from 8 manufacturers. This algorithm may provide a clinically meaningful adjunct in assisting with preoperative planning for the failed TSA and allows for scalable expansion with additional radiographic data and validation efforts.

An Interpretable Machine Learning Model for Predicting 10-Year Total Hip Arthroplasty Risk.

BACKGROUND: As the demand for total hip arthroplasty (THA) rises, a predictive model for THA risk may aid patients and clinicians in augmenting shared decision-making. We aimed to develop and validate a model predicting THA within 10 years in patients using demographic, clinical, and deep learning (DL)-automated radiographic measurements. METHODS: Patients enrolled in the osteoarthritis initiative were included. DL algorithms measuring osteoarthritis- and dysplasia-relevant parameters on baseline pelvis radiographs were developed. Demographic, clinical, and radiographic measurement variables were then used to train generalized additive models to predict THA within 10 years from baseline. A total of 4,796 patients were included [9,592 hips; 58% female; 230 THAs (2.4%)]. Model performance using 1) baseline demographic and clinical variables 2) radiographic variables, and 3) all variables was compared. RESULTS: Using 110 demographic and clinical variables, the model had a baseline area under the receiver operating curve (AUROC) of 0.68 and area under the precision recall curve (AUPRC) of 0.08. Using 26 DL-automated hip measurements, the AUROC was 0.77 and AUPRC was 0.22. Combining all variables, the model improved to an AUROC of 0.81 and AUPRC of 0.28. Three of the top five predictive features in the combined model were radiographic variables, including minimum joint space, along with hip pain and analgesic use. Partial dependency plots revealed predictive discontinuities for radiographic measurements consistent with literature thresholds of osteoarthritis progression and hip dysplasia. CONCLUSION: A machine learning model predicting 10-year THA performed more accurately with DL radiographic measurements. The model weighted predictive variables in concordance with clinical THA pathology assessments.

Deep Learning Phenotype Automation and Cohort Analyses of 1,946 Knees Using the Coronal Plane Alignment of the Knee Classification.

BACKGROUND: The Coronal Plane Alignment of the Knee (CPAK) classification allows for knee phenotyping which can be used in preoperative planning prior to total knee arthroplasty. We used deep learning (DL) to automate knee phenotyping and analyzed CPAK distributions in a large patient cohort. METHODS: Patients who had full-limb radiographs from a large arthritis database were retrospectively included. A DL algorithm was developed to automate CPAK knee alignment parameters including the lateral distal femoral, medial proximal tibia, hip-knee-ankle, and joint line obliquity angles. The algorithm was validated against a fellowship-trained arthroplasty surgeon. After applying the algorithm in a large patient cohort (n = 1,946 knees), the distribution of CPAK was compared across patient sex and baseline Kellgren-Lawrence (KL) scores. RESULTS: There was no significant difference in the CPAK angles (n = 140, P = .66-.98, inter-class correlation coefficient = 0.89-0.91) or phenotype classifications made by the algorithm and surgeon (P = .96). The deep learning algorithm measured the entire cohort (n = 1,946 knees, mean age 61 years [range, 46 to 80 years], 51% women) in < 5 hours. Women had more valgus CPAK phenotypes than men (P < .05). Patients who had higher KL grades at baseline (2 to 4) were more varus using the CPAK classification compared to lower KL grades (0 to 1) (P < .05). CONCLUSION: We applied an accurate, automated DL algorithm on a large patient cohort to determine knee phenotypes, helping to validate and strengthen the CPAK classification system. Analyses revealed that sex-specific and major bone loss adjustments may need to be accounted for when using this system.

Leg-Length Discrepancy Variability on Standard Anteroposterior Pelvis Radiographs: An Analysis Using Deep Learning Measurements.

BACKGROUND: Leg-length discrepancy (LLD) is a critical factor in component selection and placement for total hip arthroplasty. However, LLD radiographic measurements are subject to variation based on the femoral/pelvic landmarks chosen. This study leveraged deep learning (DL) to automate LLD measurements on pelvis radiographs and compared LLD based on several anatomically distinct landmarks. METHODS: Patients who had baseline anteroposterior pelvis radiographs from the Osteoarthritis Initiative were included. A DL algorithm was created to identify LLD-relevant landmarks (ie, teardrop (TD), obturator foramen, ischial tuberosity, greater and lesser trochanters) and measure LLD accurately using six landmark combinations. The algorithm was then applied to automate LLD measurements in the entire cohort of patients. Interclass correlation coefficients (ICC) were calculated to assess agreement between different LLD methods. RESULTS: The DL algorithm measurements were first validated in an independent cohort for all six LLD methods (ICC = 0.73-0.98). Images from 3,689 patients (22,134 LLD measurements) were measured in 133 minutes. When using the TD and lesser trochanter landmarks as the standard LLD method, only measuring LLD using the TD and greater trochanter conferred acceptable agreement (ICC = 0.72). When comparing all six LLD methods for agreement, no combination had an ICC>0.90. Only two (13%) combinations had an ICC>0.75 and eight (53%) combinations had a poor ICC (<0.50). CONCLUSION: We leveraged DL to automate LLD measurements in a large patient cohort and found considerable variation in LLD based on the pelvic/femoral landmark selection. This emphasizes the need for the standardization of landmarks for both research and surgical planning.

Radiographic findings involved in knee osteoarthritis progression are associated with pain symptom frequency and baseline disease severity: a population-level analysis using deep learning.

PURPOSE: To (1) develop a deep-learning (DL) algorithm capable of producing limb-length and knee-alignment measurements, and (2) determine the association between limb-length discrepancy (LLD), coronal-plane alignment, osteoarthritis (OA) severity, and patient-reported knee pain. METHODS: A multicenter, prospective patient cohort from the Osteoarthritis Initiative between 2004 and 2015 with full-limb standing radiographs at 12 month follow-up was included. A convolutional neural network was developed to automate measurements of the hip-knee-ankle (HKA) angle, femur, and tibia lengths, and LLD. At 12 month follow-up, patients reported their frequency of knee pain since enrollment and current level of knee pain. RESULTS: A total of 1011 patients (2022 knees, 52.3% female) with an average age of 61.2 ± 9.0 years were included. The algorithm performed 12,312 measurements in 5.4 h. ICC values of HKA and LLD ranged between 0.87 and 1.00 when compared against trained radiologist measurements. Knees producing pain most days of the month were significantly more varus (mean HKA:- 3.9° ± 2.8°) or valgus (mean HKA:2.8° ± 2.3°) compared to knees that did not produce any pain (p < 0.05). In varus knees, those producing pain on most days were part of the shorter limb compared to nonpainful knees (p < 0.05). Baseline Kellgren-Lawrence grade was significantly associated with HKA magnitude, LLD, and pain frequency at 12 month follow-up (p < 0.05 all). CONCLUSION: A higher frequency of knee pain was associated with more severe coronal plane deformity, with valgus deviation being one degree less than varus on average, suggesting that the knee tolerates less valgus deformation before symptoms become more consistent. Knee pain frequency was also associated with greater LLD and baseline KL grade, suggesting an association between radiographically apparent joint degeneration and pain frequency. LEVEL OF EVIDENCE: IV case series.

John Charnley Award: Deep Learning Prediction of Hip Joint Center on Standard Pelvis Radiographs.

BACKGROUND: Accurate hip joint center (HJC) determination is critical for preoperative planning, intraoperative execution, clinical outcomes after total hip arthroplasty, and commonly used classification systems in primary and revision hip replacement. However, current methods of preoperative HJC estimation are prone to subjectivity and human error. The purpose of the study was to leverage deep learning (DL) to develop a rapid and objective HJC estimation tool on anteroposterior (AP) pelvis radiographs. METHODS: Radiographs from 3,965 patients (7,930 hips) were included. A DL model workflow was created to detect bony landmarks and estimate HJC based on a pelvic height ratio method. The workflow was utilized to conduct a grid-search for optimal nonspecific, sex-specific, and patient-specific (using contralateral hip) pelvic height ratios on the training/validation cohort (6,344 hips). Algorithm performance was assessed on an independent testing cohort for HJC estimation comparison. RESULTS: The algorithm estimated HJC for the testing cohort at a rate of 0.65 seconds/hip based on features in AP radiographs alone. The model predicted HJC within 5 mm of error for 80% of hips using nonspecific ratios, which increased to 83% with sex-specific and 91% with patient-specific pelvic height ratio models. Mean error decreased utilizing the patient-specific model (3.09 ± 1.69 mm, P < .001). CONCLUSION: Using DL, we developed nonspecific, sex-specific, and patient-specific models capable of estimating native HJC on AP pelvis radiographs. This tool may provide clinical value when considering preoperative component position in patients planned to undergo THA and in reducing the subjective variability in HJC estimation. LEVEL OF EVIDENCE: Diagnostic, level IV.