A cadaver study using six pairs of lower limbs was conducted to investigate the accuracy of computer navigation and standard instrumentation for the placement of the Birmingham Hip Resurfacing femoral component. The aim was to place all the femoral components with a stem-shaft angle of 135°. The mean stem-shaft angle obtained in the standard instrumentation group was 127.7° (120° to 132°), compared with 133.3° (131° to 139°) in the computer navigation group (p = 0.03). The scatter obtained with computer-assisted navigation was approximately half that found using the conventional jig. Computer navigation was more accurate and more consistent in its placement of the femoral component than standard instrumentation. We suggest that image-free computer-assisted navigation may have an application in aligning the femoral component during hip resurfacing

In 100 patients the fulcrum axis which is the line connecting the anterior tip of the coracoid and the posterolateral angle of the acromion, was used to position true anteroposterior radiographs of the shoulder. This method was then compared with the conventional radiological technique in a further 100 patients. Three orthopaedic surgeons counted the number of images without overlap between the humeral head and glenoid and calculated the amount of the glenoid surface visible in each radiograph. The analysis was repeated for intraobserver reliability. The learning curves of both techniques were studied. The amount of free visible glenoid space was significantly higher using the fulcrum-axis method (64 vs 31) and the comparable glenoid size increased significantly (8.56 vs 6.47). Thus the accuracy of the anteroposterior radiographs of the shoulder is impaired by using this technique. The intra and interobserver reliability showed a high consistency. No learning curve was observed for either technique

Dual-energy X-ray absorptiometry (DEXA) is increasingly used to measure changes in bone mineral density (BMD) around femoral prostheses after total hip arthroplasty. We have studied the factors which affect the accuracy of these measurements. The coefficient of variation was < 2% using a hydroxyapatite phantom, 2.7% in an anthropomorphic phantom specimen, and < 1% in repeated measurements on implanted cadaver femora. The precision did not vary with different implant materials or designs. In patients we found a mean precision error of 2.7% to 3.4%. The most significant factor affecting reproducibility was rotation of the femur. We conclude that DEXA is a precise method of measurement for small changes in BMD around femoral implants, but that correct and careful positioning of patients is essential to obtain reliable results

Templates are used in the preoperative planning of many orthopaedic procedures. The magnification of the bones on preoperative radiographs can vary despite using standardised radiological techniques. Templates will give misleading measurements unless this magnification is quantified. A coin may be used to calculate the magnification, with significant improvement in the accuracy of templating (p = 0.05). A group of patients undergoing uncemented arthroplasty of the hip was at high risk for intraoperative fracture of the femur because the magnification of the radiograph was larger than that of the template

There is no single standardised method of measuring the orientation of the acetabular component on plain radiographs after total hip arthroplasty. We assessed the reliability and accuracy of three methods of assessing anteversion of the acetabular component for 551 THAs using the PolyWare software and the methods of Liaw et al, and of Woo and Morrey. All measurements of the three methods had excellent intra- and inter-observer reliability. The values of the PolyWare software, which determines version of the acetabular component by edge detection were regarded as the reference standard. Although the PolyWare software and the method of Liaw et al were similarly precise, the method of Woo and Morrey was significantly less accurate (p < 0.001). The method of Liaw et al seemed to be more accurate than that of Woo and Morrey when compared with the measurements using the PolyWare software. If the qualified lateral radiograph was selected, anteversion measured using the method of Woo and Morrey was considered to be relatively reliable.

Cite this article: Bone Joint J 2015; 97-B:611–16.

Using inaccurate quotations can propagate misleading information, which might affect the management of patients. The aim of this study was to determine the predictors of quotation inaccuracy in the peer-reviewed orthopaedic literature related to the scaphoid. We randomly selected 100 papers from ten orthopaedic journals. All references were retrieved in full text when available or otherwise excluded. Two observers independently rated all quotations from the selected papers by comparing the claims made by the authors with the data and expressed opinions of the reference source. A statistical analysis determined which article-related factors were predictors of quotation inaccuracy. The mean total inaccuracy rate of the 3840 verified quotes was 7.6%. There was no correlation between the rate of inaccuracy and the impact factor of the journal. Multivariable analysis identified the journal and the type of study (clinical, biomechanical, methodological, case report or review) as important predictors of the total quotation inaccuracy rate.

We concluded that inaccurate quotations in the peer-reviewed orthopaedic literature related to the scaphoid were common and slightly more so for certain journals and certain study types. Authors, reviewers and editorial staff play an important role in reducing this inaccuracy.

CT arthrography and arthroscopy were used to assess tears of the rotator cuff in 259 shoulders. Tear size was determined in the frontal and sagittal planes according to the classification of the French Arthroscopy Society.

CT arthrography had a sensitivity of 99% and a specificity of 100% for the diagnosis of tears of supraspinatus. For infraspinatus these figures were 97.44% and 99.52%, respectively and, for subscapularis, 64.71% and 98.17%. For lesions of the long head of the biceps, the sensitivity was 45.76% and the specificity was 99.57%.

Our study showed an excellent correlation between CT arthrography and arthroscopy when assessing the extent of a rotator cuff tear. CT arthrography should, therefore, be an indispensable part of pre-operative assessment. It allows determination of whether a tear is reparable (retraction of the tendon and fatty degeneration of the corresponding muscle) and whether this is possible by arthroscopy (degree of tendon retraction and extension to subscapularis).

In this meta-analysis we included 32 English-language articles published between January 1975 and June 2004 on the diagnostic performance of plain radiography, subtraction arthrography, nuclear arthrography and bone scintigraphy in detecting aseptic loosening of the femoral component, using criteria based on the Cochrane systematic review of screening and diagnostic tests.

The mean sensitivity and specificity were, respectively, 82% (95% confidence interval (CI) 76 to 87) and 81% (95% CI 73 to 87) for plain radiography and 85% (95% CI 75 to 91) and 83% (95% CI 75 to 89) for nuclear arthrography. Pooled sensitivity and specificity were, respectively, 86% (95% CI 74 to 93) and 85% (95% CI 77 to 91) for subtraction arthrography and 85% (95% CI 79 to 89) and 72% (95% CI 64 to 79) for bone scintigraphy. Although the diagnostic performance of the imaging techniques was not significantly different, plain radiography and bone scintigraphy are preferred for the assessment of a femoral component because of their efficacy and lower risk of patient morbidity.

When performing the Scandinavian Total Ankle Replacement (STAR), the positioning of the talar component and the selection of mobile-bearing thickness are critical. A biomechanical experiment was undertaken to establish the effects of these variables on the range of movement (ROM) of the ankle.

Six cadaver ankles containing a specially-modified STAR prosthesis were subjected to ROM determination, under weight-bearing conditions, while monitoring the strain in the peri-ankle ligaments. Each specimen was tested with the talar component positions in neutral, as well as 3 and 6 mm of anterior and posterior displacement. The sequence was repeated with an anatomical bearing thickness, as well as at 2 mm reduced and increased thicknesses. The movement limits were defined as 10% strain in any ligament, bearing lift-off from the talar component or limitations of the hardware.

Both anterior talar component displacement and bearing thickness reduction caused a decrease in plantar flexion, which was associated with bearing lift-off. With increased bearing thickness, posterior displacement of the talar component decreased plantar flexion, whereas anterior displacement decreased dorsiflexion.

Aims. Precise implant positioning, tailored to individual spinopelvic biomechanics and phenotype, is paramount for stability in total hip arthroplasty (THA). Despite a few studies on instability prediction, there is a notable gap in research utilizing artificial intelligence (AI). The objective of our pilot study was to evaluate the feasibility of developing an AI algorithm tailored to individual spinopelvic mechanics and patient phenotype for predicting impingement. Methods. This international, multicentre prospective cohort study across two centres encompassed 157 adults undergoing primary robotic arm-assisted THA. Impingement during specific flexion and extension stances was identified using the virtual range of motion (ROM) tool of the robotic software. The primary AI model, the Light Gradient-Boosting Machine (LGBM), used tabular data to predict impingement presence, direction (flexion or extension), and type. A secondary model integrating tabular data with plain anteroposterior pelvis radiographs was evaluated to assess for any potential enhancement in prediction accuracy. Results. We identified nine predictors from an analysis of baseline spinopelvic characteristics and surgical planning parameters. Using fivefold cross-validation, the LGBM achieved 70.2% impingement prediction accuracy. With impingement data, the LGBM estimated direction with 85% accuracy, while the support vector machine (SVM) determined impingement type with 72.9% accuracy. After integrating imaging data with a multilayer perceptron (tabular) and a convolutional neural network (radiograph), the LGBM’s prediction was 68.1%. Both combined and LGBM-only had similar impingement direction prediction rates (around 84.5%). Conclusion. This study is a pioneering effort in leveraging AI for impingement prediction in THA, utilizing a comprehensive, real-world clinical dataset. Our machine-learning algorithm demonstrated promising accuracy in predicting impingement, its type, and direction. While the addition of imaging data to our deep-learning algorithm did not boost accuracy, the potential for refined annotations, such as landmark markings, offers avenues for future enhancement. Prior to clinical integration, external validation and larger-scale testing of this algorithm are essential. Cite this article: Bone Jt Open 2024;5(8):671–680

Aims. This study aimed to evaluate the clinical application of the PJI-TNM classification for periprosthetic joint infection (PJI) by determining intraobserver and interobserver reliability. To facilitate its use in clinical practice, an educational app was subsequently developed and evaluated. Methods. A total of ten orthopaedic surgeons classified 20 cases of PJI based on the PJI-TNM classification. Subsequently, the classification was re-evaluated using the PJI-TNM app. Classification accuracy was calculated separately for each subcategory (reinfection, tissue and implant condition, non-human cells, and morbidity of the patient). Fleiss’ kappa and Cohen’s kappa were calculated for interobserver and intraobserver reliability, respectively. Results. Overall, interobserver and intraobserver agreements were substantial across the 20 classified cases. Analyses for the variable ‘reinfection’ revealed an almost perfect interobserver and intraobserver agreement with a classification accuracy of 94.8%. The category 'tissue and implant conditions' showed moderate interobserver and substantial intraobserver reliability, while the classification accuracy was 70.8%. For 'non-human cells,' accuracy was 81.0% and interobserver agreement was moderate with an almost perfect intraobserver reliability. The classification accuracy of the variable 'morbidity of the patient' reached 73.5% with a moderate interobserver agreement, whereas the intraobserver agreement was substantial. The application of the app yielded comparable results across all subgroups. Conclusion. The PJI-TNM classification system captures the heterogeneity of PJI and can be applied with substantial inter- and intraobserver reliability. The PJI-TNM educational app aims to facilitate application in clinical practice. A major limitation was the correct assessment of the implant situation. To eliminate this, a re-evaluation according to intraoperative findings is strongly recommended. Cite this article: Bone Joint Res 2024;13(1):19–27

Aims. The aim of this audit was to assess and improve the completeness and accuracy of the National Joint Registry (NJR) dataset for arthroplasty of the elbow. Methods. It was performed in two phases. In Phase 1, the completeness was assessed by comparing the NJR elbow dataset with the NHS England Hospital Episode Statistics (HES) data between April 2012 and April 2020. In order to assess the accuracy of the data, the components of each arthroplasty recorded in the NJR were compared to the type of arthroplasty which was recorded. In Phase 2, a national collaborative audit was undertaken to evaluate the reasons for unmatched data, add missing arthroplasties, and evaluate the reasons for the recording of inaccurate arthroplasties and correct them. Results. Phase 1 identified 5,539 arthroplasties in HES which did not match an arthroplasty on the NJR, and 448 inaccurate arthroplasties from 254 hospitals. Most mismatched procedures (3,960 procedures; 71%) were radial head arthroplasties (RHAs). In Phase 2, 142 NHS hospitals with 3,640 (66%) mismatched and 314 (69%) inaccurate arthroplasties volunteered to assess their records. A large proportion of the unmatched data (3,000 arthroplasties; 82%) were confirmed as being missing from the NJR. The overall rate of completeness of the NJR elbow dataset improved from 63% to 83% following phase 2, and the completeness of total elbow arthroplasty data improved to 93%. Missing RHAs had the biggest impact on the overall completeness, but through the audit the number of RHAs in the NJR nearly doubled and completeness increased from 35% to 70%. The accuracy of data was 94% and improved to 98% after correcting 212 of the 448 inaccurately recorded arthroplasties. Conclusion. The rate of completeness of the NJR total elbow arthroplasty dataset is currently 93% and the accuracy is 98%. This audit identified challenges of data capture with regard to RHAs. Collaboration with a trauma and orthopaedic trainees through the British Orthopaedic Trainee Association improved the completeness and accuracy of the NJR elbow dataset, which will improve the validity of the reports and of the associated research. Cite this article: Bone Joint J 2024;106-B(12):1461–1468

Aims. Accurate skeletal age and final adult height prediction methods in paediatric orthopaedics are crucial for determining optimal timing of growth-guiding interventions and minimizing complications in treatments of various conditions. This study aimed to evaluate the accuracy of final adult height predictions using the central peak height (CPH) method with long leg X-rays and four different multiplier tables. Methods. This study included 31 patients who underwent temporary hemiepiphysiodesis for varus or valgus deformity of the leg between 2014 and 2020. The skeletal age at surgical intervention was evaluated using the CPH method with long leg radiographs. The true final adult height (FH. TRUE. ) was determined when the growth plates were closed. The final height prediction accuracy of four different multiplier tables (1. Bayley and Pinneau; 2. Paley et al; 3. Sanders – Greulich and Pyle (SGP); and 4. Sanders – peak height velocity (PHV)) was then compared using either skeletal age or chronological age. Results. All final adult height predictions overestimated the FH. TRUE. , with the SGP multiplier table having the lowest overestimation and lowest absolute deviation when using both chronological age and skeletal age. There were no significant differences in final height prediction accuracy between using skeletal age and chronological age with PHV (p = 0.652) or SGP multiplier tables (p = 0.969). Adult height predictions with chronological age and SGP (r = 0.769; p ≤ 0.001), as well as chronological age and PHV (r = 0.822; p ≤ 0.001), showed higher correlations with FH. TRUE. than predictions with skeletal age and SGP (r = 0.657; p ≤ 0.001) or skeletal age and PHV (r = 0.707; p ≤ 0.001). Conclusion. There was no significant improvement in adult height prediction accuracy when using the CPH method compared to chronological age alone. The study concludes that there is no advantage in routinely using the CPH method for skeletal age determination over the simple use of chronological age. The findings highlight the need for more accurate methods to predict final adult height in contemporary patient populations. Cite this article: Bone Jt Open 2023;4(10):750–757

Aims. This study aimed to answer the following questions: do 3D-printed models lead to a more accurate recognition of the pattern of complex fractures of the elbow?; do 3D-printed models lead to a more reliable recognition of the pattern of these injuries?; and do junior surgeons benefit more from 3D-printed models than senior surgeons?. Methods. A total of 15 orthopaedic trauma surgeons (seven juniors, eight seniors) evaluated 20 complex elbow fractures for their overall pattern (i.e. varus posterior medial rotational injury, terrible triad injury, radial head fracture with posterolateral dislocation, anterior (trans-)olecranon fracture-dislocation, posterior (trans-)olecranon fracture-dislocation) and their specific characteristics. First, fractures were assessed based on radiographs and 2D and 3D CT scans; and in a subsequent round, one month later, with additional 3D-printed models. Diagnostic accuracy (acc) and inter-surgeon reliability (κ) were determined for each assessment. Results. Accuracy significantly improved with 3D-printed models for the whole group on pattern recognition (acc. 2D/3D. = 0.62 vs acc. 3Dprint. = 0.69; Δacc = 0.07 (95% confidence interval (CI) 0.00 to 0.14); p = 0.025). A significant improvement was also seen in reliability for pattern recognition with the additional 3D-printed models (κ. 2D/3D. = 0.41 (moderate) vs κ. 3Dprint. = 0.59 (moderate); Δκ = 0.18 (95% CI 0.14 to 0.22); p ≤ 0.001). Accuracy was comparable between junior and senior surgeons with the 3D-printed model (acc. junior. = 0.70 vs acc. senior. = 0.68; Δacc = -0.02 (95% CI -0.17 to 0.13); p = 0.904). Reliability was also comparable between junior and senior surgeons without the 3D-printed model (κ. junior. = 0.39 (fair) vs κ. senior. = 0.43 (moderate); Δκ = 0.03 (95% CI -0.03 to 0.10); p = 0.318). However, junior surgeons showed greater improvement regarding reliability than seniors with 3D-printed models (κ. junior. = 0.65 (substantial) vs κ. senior. = 0.54 (moderate); Δκ = 0.11 (95% CI 0.04 to 0.18); p = 0.002). Conclusion. The use of 3D-printed models significantly improved the accuracy and reliability of recognizing the pattern of complex fractures of the elbow. However, the current long printing time and non-reusable materials could limit the usefulness of 3D-printed models in clinical practice. They could be suitable as a reusable tool for teaching residents. Cite this article: Bone Joint J 2023;105-B(1):56–63

Aims. The aim of this study was to create artificial intelligence (AI) software with the purpose of providing a second opinion to physicians to support distal radius fracture (DRF) detection, and to compare the accuracy of fracture detection of physicians with and without software support. Methods. The dataset consisted of 26,121 anonymized anterior-posterior (AP) and lateral standard view radiographs of the wrist, with and without DRF. The convolutional neural network (CNN) model was trained to detect the presence of a DRF by comparing the radiographs containing a fracture to the inconspicuous ones. A total of 11 physicians (six surgeons in training and five hand surgeons) assessed 200 pairs of randomly selected digital radiographs of the wrist (AP and lateral) for the presence of a DRF. The same images were first evaluated without, and then with, the support of the CNN model, and the diagnostic accuracy of the two methods was compared. Results. At the time of the study, the CNN model showed an area under the receiver operating curve of 0.97. AI assistance improved the physician’s sensitivity (correct fracture detection) from 80% to 87%, and the specificity (correct fracture exclusion) from 91% to 95%. The overall error rate (combined false positive and false negative) was reduced from 14% without AI to 9% with AI. Conclusion. The use of a CNN model as a second opinion can improve the diagnostic accuracy of DRF detection in the study setting. Cite this article: Bone Joint Res 2024;13(10):588–595

Aims. The aim of this study was to assess the accuracy of pedicle screw placement, as well as intraoperative factors, radiation exposure, and complication rates in adult patients with degenerative disorders of the thoracic and lumbar spines who have undergone robotic-navigated spinal surgery using a contemporary system. Methods. The authors reviewed the prospectively collected data on 196 adult patients who had pedicle screws implanted with robot-navigated assistance (RNA) using the Mazor X Stealth system between June 2019 and March 2022. Pedicle screws were implanted by one experienced spinal surgeon after completion of a learning period. The accuracy of pedicle screw placement was determined using intraoperative 3D fluoroscopy. Results. A total of 1,123 pedicle screws were implanted: 1,001 screws (89%) were placed robotically, 63 (6%) were converted from robotic placement to a freehand technique, and 59 (5%) were planned to be implanted freehand. Of the robotically placed screws, 942 screws (94%) were determined to be Gertzbein and Robbins grade A with median deviation of 0.8 mm (interquartile range 0.4 to 1.6). Skive events were noted with 20 pedicle screws (1.8%). No adverse clinical sequelae were noted in the 90-day follow-up. The mean fluoroscopic exposure per screw was 4.9 seconds (SD 3.8). Conclusion. RNA is highly accurate and reliable, with a low rate of abandonment once mastered. No adverse clinical sequelae occurred after implanting a large series of pedicle screws using the latest generation of RNA. Understanding of patient-specific anatomical features and the real-time intraoperative identification of risk factors for suboptimal screw placement have the potential to improve accuracy further. Cite this article: Bone Joint J 2023;105-B(5):543–550

Aims. The diagnosis of periprosthetic joint infection (PJI) can be challenging as the symptoms are similar to other conditions, and the markers used for diagnosis have limited sensitivity and specificity. Recent research has suggested using blood cell ratios, such as platelet-to-volume ratio (PVR) and platelet-to-lymphocyte ratio (PLR), to improve diagnostic accuracy. The aim of the study was to further validate the effectiveness of PVR and PLR in diagnosing PJI. Methods. A retrospective review was conducted to assess the accuracy of different marker combinations for diagnosing chronic PJI. A total of 573 patients were included in the study, of which 124 knees and 122 hips had a diagnosis of chronic PJI. Complete blood count and synovial fluid analysis were collected. Recently published blood cell ratio cut-off points were applied to receiver operating characteristic curves for all markers and combinations. The area under the curve (AUC), sensitivity, specificity, and positive and negative predictive values were calculated. Results. The results of the analysis showed that the combination of ESR, CRP, synovial white blood cell count (Syn. WBC), and polymorphonuclear neutrophil percentage (PMN%) with PVR had the highest AUC of 0.99 for knees, with sensitivity of 97.73% and specificity of 100%. Similarly, for hips, this combination had an AUC of 0.98, sensitivity of 96.15%, and specificity of 100.00%. Conclusion. This study supports the use of PVR calculated from readily available complete blood counts, combined with established markers, to improve the accuracy in diagnosing chronic PJI in both total hip and knee arthroplasties. Cite this article: Bone Jt Open 2023;4(11):881–888

Aims. The aim of this study was to determine the rate of indocyanine green (ICG) staining of bone and soft-tissue tumours, as well as the stability and accuracy of ICG fluorescence imaging in detecting tumour residuals during surgery for bone and soft-tissue tumours. Methods. ICG fluorescence imaging was performed during surgery in 34 patients with bone and soft-tissue tumours. ICG was administered intravenously at a dose of 2 mg/kg over a period of 60 minutes on the day prior to surgery. The tumour stain rate and signal-to-background ratio of each tumour were post hoc analyzed. After tumour resection, the tumour bed was scanned to locate sites with fluorescence residuals, which were subsequently inspected and biopsied. Results. The overall tumour stain rate was 88% (30/34 patients), and specific stain rates included 90% for osteosarcomas and 92% for giant cell tumours. For malignant tumours, the overall stain rate was 94%, while it was 82% for benign tumours. The ICG tumour stain was not influenced by different pathologies, such as malignant versus benign pathology, the reception (or lack thereof) of neoadjuvant chemotherapies, the length of time between drug administration and surgery, the number of doses of denosumab for patients with giant cell tumours, or the tumour response to neoadjuvant chemotherapy. The overall accuracy rate of successfully predicting tumour residuals using fluorescence was 49% (23/47 pieces of tissue). The accuracy rate after en bloc resection was significantly lower than that after piecemeal resection (16% vs 71%; p < 0.001). Conclusion. A high percentage of bone and soft-tissue tumours can be stained by ICG and the tumour staining with ICG was stable. This approach can be used in both benign and malignant tumours, regardless of whether neoadjuvant chemotherapy is adopted. The technique is also useful to detect tumour residuals in the wound, especially in patients undergoing piecemeal resection. Cite this article: Bone Joint J 2023;105-B(5):551–558

Aims. The purpose of this study was to develop a convolutional neural network (CNN) for fracture detection, classification, and identification of greater tuberosity displacement ≥ 1 cm, neck-shaft angle (NSA) ≤ 100°, shaft translation, and articular fracture involvement, on plain radiographs. Methods. The CNN was trained and tested on radiographs sourced from 11 hospitals in Australia and externally validated on radiographs from the Netherlands. Each radiograph was paired with corresponding CT scans to serve as the reference standard based on dual independent evaluation by trained researchers and attending orthopaedic surgeons. Presence of a fracture, classification (non- to minimally displaced; two-part, multipart, and glenohumeral dislocation), and four characteristics were determined on 2D and 3D CT scans and subsequently allocated to each series of radiographs. Fracture characteristics included greater tuberosity displacement ≥ 1 cm, NSA ≤ 100°, shaft translation (0% to < 75%, 75% to 95%, > 95%), and the extent of articular involvement (0% to < 15%, 15% to 35%, or > 35%). Results. For detection and classification, the algorithm was trained on 1,709 radiographs (n = 803), tested on 567 radiographs (n = 244), and subsequently externally validated on 535 radiographs (n = 227). For characterization, healthy shoulders and glenohumeral dislocation were excluded. The overall accuracy for fracture detection was 94% (area under the receiver operating characteristic curve (AUC) = 0.98) and for classification 78% (AUC 0.68 to 0.93). Accuracy to detect greater tuberosity fracture displacement ≥ 1 cm was 35.0% (AUC 0.57). The CNN did not recognize NSAs ≤ 100° (AUC 0.42), nor fractures with ≥ 75% shaft translation (AUC 0.51 to 0.53), or with ≥ 15% articular involvement (AUC 0.48 to 0.49). For all objectives, the model’s performance on the external dataset showed similar accuracy levels. Conclusion. CNNs proficiently rule out proximal humerus fractures on plain radiographs. Despite rigorous training methodology based on CT imaging with multi-rater consensus to serve as the reference standard, artificial intelligence-driven classification is insufficient for clinical implementation. The CNN exhibited poor diagnostic ability to detect greater tuberosity displacement ≥ 1 cm and failed to identify NSAs ≤ 100°, shaft translations, or articular fractures. Cite this article: Bone Joint J 2024;106-B(11):1348–1360