Artificial intelligence for the assessment of distal radius fracture instability on wrist radiographs: A Diagnostic Study

Artificial intelligence for the assessment of distal radius fracture instability on wrist radiographs: A Diagnostic Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Artificial intelligence for the assessment of distal radius fracture instability on wrist radiographs: A Diagnostic Study Tharanas Jantharagsarangsee, Thepparat Kanchanathepsak, Tulyapruek Tawonsawatruk, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8952141/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Misinterpretation of radiographs of distal radius fractures (DRFs) can lead to malalignment and hand and wrist disability. The application of artificial intelligence (AI) in detecting and predicting fracture instability still needs to be explored. This study aimed to evaluate the diagnostic accuracy of AI in identifying correlated factors, as well as in predicting unstable DRF, Fernandez types 1 and 3. Methods DRF radiographs (Fernandez 1 and 3) of adults aged > 18 years were retrieved from a university hospital and a provincial hospital between January 2017 and May 2022. Radiographs with any concomitant fracture, inadequate imaging, or radiocarpal pathology were excluded. Unstable fracture indicated operative intervention one the basis of 1) the Lafontaine criteria, and 2) radiographic operative factors (metaphyseal comminution and radial shortening). Two AI distal end radius (AIDER) models, AIDER1, identified each radiographic Lafontaine criterion and operative factor, and AIDER2, which predicts Lafontaine criteria ≥ 3 or unstable fractures, were trained, validated, and tested (70:15:15 ratio). Results Among the 1,548 initial radiographs, 1,469 films were eligible. The patients’ characteristics were as follows: 694 (47%) were elderly, 984 (67%) were female, and 840 (57%) were Fernandez type 3. Diagnostic accuracy of the testing groups: AIDER1 achieved the accuracy threshold in dorsal comminution (74%), dorsal tilt > 20 degrees (76%), and metaphyseal comminution (74%). This model provided low accuracy for ulnar styloid fractures, intra-articular fractures, and radial shortening (67%-71%). For AIDER2, the detection of unstable fractures achieved an accuracy of 82%, with an area under the curve (AUC) of 0.63 (95% CI: 0.53, 0.73), suggesting moderate discriminative ability. Conclusion The AI models performed well for selected radiographic features but showed limited accuracy for several operation-related findings. These results indicate that the system may assist with preliminary assessment, but further refinement, broader training data, and external validation are needed before it can be used to support treatment decisions reliably. Diagnostic accuracy Lafontaine criteria metaphyseal comminution radial shortening unstable fracture pattern Figures Figure 1 Background Distal radius fractures (DRFs) are frequent injuries, constituting up to 18% of all fractures from accidents [ 1 ] and approximately 44% of forearm fractures [ 2 ]. The incidence rate is 20 per 10,000 person-years, with a higher prevalence among individuals aged 50 years and older [ 3 ]. Among our 706 patients aged 20 years and older, 90% had undergone surgery during the past ten years. DRF is commonly classified by the mechanism of injury via the Fernandez classification [ 4 ], which aids in treatment decisions [ 5 ]. This classification system has demonstrated reliable interobserver and intraobserver agreement [ 6 – 9 ]. To identify instability patterns from initial radiographic assessment, the Lafontaine criteria (≥ 3 out of 5), metaphyseal comminution, and radial shortening > 5 mm are pivotal in determining proper treatment [ 5 , 10 – 15 ]. Most patients with DRF, especially in Thailand, initially seek treatment from general practitioners or emergency physicians before referral to orthopedists. However, some cases are not appropriately transferred due to diagnostic errors, particularly in the misinterpretation of radiographic images [ 16 , 17 ]. Other factors contributing to these errors include communication issues, lack of experience in treatment, heavy workload, and fatigue, particularly during off-hours [ 18 ]. Consequently, complications such as malunion of the distal radius are common, leading to chronic wrist pain[ 19 , 20 ] and reduced wrist function [ 20 , 21 ]. In the United States, 0.04% of DRF cases managed by orthopaedic surgeons insured by the Medical Liability Mutual Insurance Company (MLMIC) result in malpractice lawsuits annually [ 22 , 23 ]. To address these challenges, this article proposes the use of artificial intelligence (AI) to assist in predicting operative versus nonoperative treatment of DRF on the basis of initial radiographic images. AI can help recognize key radiographic factors, thereby reducing diagnostic errors and approving therapeutic judgment [ 24 ]. This application aims to enhance physicians' ability to deliver personalized patient care, increase confidence in treatment decisions, and mitigate biases in treatment selection [ 25 ]. The utility of AI in orthopedic imaging has been widely established, demonstrating the ability to accurately identify fractures in major areas (e.g., humerus, hip, and ankle) [ 26 ] and aid in the diagnostic screening of DRF as effectively as expert surgeons do [ 27 – 33 ]. However, a limitation frequently reported is a notable drop in performance when complex features such as intra-articular fracture and fracture displacement are being assessed [ 28 ]. Additionally, the other radiographic parameters from the Lafontaine criteria were not considered. To our knowledge, no previous studies have attempted to use AI to assist in operative decision-making on the basis of initial radiographs. The objectives of this study were to assess the accuracy of AI compared with the consensus of two hand orthopedists in identifying each radiographic component of the Lafontaine criteria (ulnar styloid fracture, dorsal comminution, intra-articular fracture, and dorsal tilt > 20 degrees) and to predict stable or unstable DRF in initial radiographs of Fernandez types 1 and 3. Methods Initial wrist radiographs (PA and lateral views) of DRF, including Fernandez type 1 (extra-articular metaphyseal fracture) and type 3 (intra-articular fracture) [ 4 ], were collected from the Picture Archiving and Communication System (PACS) of Ramathibodi Hospital and Lampang Hospital. Radiographs from patients older than 18 years between January 2007 and May 2022 were included. The radiographs of those with concomitant fractures other than ulnar styloid fractures, pathologic fractures, inadequate image quality, radiocarpal abnormalities, and the presence of foreign bodies or implants were excluded. All radiographs were anonymized by replacing patient-specific data with study numbers to ensure confidentiality and were subsequently converted into the JPEG format. An appropriate window was selected in all the images, and additional annotations, such as handedness labels, were removed to reduce the error in the AI learning process. This study was approved by the Institutional Review Board of both study hospitals (COA. MURA 2022/486, and EC 002/67). Demographic data including age, sex, Fernandez classification[ 4 ], and treatment, were recorded. Two 12-year-experienced hand orthopedists evaluated and consensused each radiographic component of Lafontaine. We use a consensus method to establish the ground truth. This critical step reduces interrater variability and ensures labeling consistency, thereby preventing negative effects on the training accuracy of the AI model. The presence of metaphyseal comminution, radial shortening greater than 5 mm, or at least 3 out of 5 Lafontaine criteria was defined as an unstable fracture. The full dataset was then partitioned into two phases for AI model development and evaluation: Phase I: AI Model Development and Internal Testing. This phase utilized a combined dataset for the creation of the AI distal end radius (AIDER), which was divided into training (70%), validation (15%), and internal testing (15%) groups. Phase II: External Testing. This phase consisted of separate radiographs used to evaluate the final model's performance via a diagnostic study. Two AIDERs were developed to detect radiographic abnormalities and predict unstable fractures. AIDER1 was designed to be an analytical model to detect each radiographic component of the Lafontaine criteria, metaphyseal comminution and radial shortening greater than 5 mm. AIDER2 is a practical application model to predict unstable fractures of DRF, which are defined as Lafontaine criteria of 3 or more, metaphyseal comminution, or radial shortening greater than 5 mm. Criteria for adequate radiographic images were set up. In the wrist posteroanterior (PA) view, adequate images clearly display the ulnar styloid at the edge of the bone in a peripheral orientation. In the lateral view of the wrist, the pisiform should lie between the volar surface of the capitate and the volar tuberosity of the scaphoid. Additional specific radiographic parameters were defined for the evaluation of DRF. On the PA view, radial shortening greater than 5 mm was defined as a radial height measurement of less than 6 mm. In the lateral view, dorsal comminution was defined as the presence of fragmented bone at the dorsal metaphysis of the distal radius. Dorsal tilt was defined as the angle formed between a line perpendicular to the long axis of the radius and a line drawn from the volar apex to the dorsal rim of the distal radius [ 34 , 35 ]. The AI models were developed via deep learning in several steps. First, image augmentation was applied to the radiographs by introducing a ± 10° rotation, ± 0.5 scaling factor, and ± 20 pixel shift to simulate variations that commonly occur in clinical practice and to improve model robustness. The model then learns relevant image features, such as bone edges and fracture patterns, through convolutional neural networks or CNNs (the core architecture uses mathematical filters to automatically and hierarchically extract features such as edges and fracture lines from image data) To improve efficiency, transfer learning was used by adapting pretrained network architectures (ResNet-18 and GoogLeNet) to the DRF dataset. The models were further refined using the study images to optimize the performance for this specific task. Finally, model performance was evaluated via standard diagnostic metrics, including accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC), and the best-performing models were selected for further analysis. The sample size of the external testing phase was calculated via MedCalc Statistical Software version 19.2.6 (MedCalc Software, Ostend, Belgium; https://www.medcalc.org ; 2020) [ 36 ]. The parameters used were set as follows: null hypothesis value = 0.5, AUC or accuracy threshold = 0.725, alpha error = 0.05, beta error = 0.2, ratio of negative/positive radiographic factor = 1, and ratio of stable/unstable fracture = 3:7 in our real-life database. The AIDER1 model accounted for 48 wrist radiographs: 24 for each negative and positive finding. A total of 56 wrist radiographs (17 stable and 39 unstable fractures) were needed for the AIDER2 model. The final dataset was substantially larger than the number required to detect moderate discriminative ability under these assumptions for early-stage model evaluation. The data were recorded in Microsoft Excel and analyzed via STATA 18.0 (StataCorp, College Station, Texas, USA). Categorical variables were compared via the chi-square test, and continuous variables were summarized via means and standard deviations. Diagnostic performance was evaluated by calculating the sensitivity, specificity, accuracy, positive predictive value, negative predictive value, and positive and negative likelihood ratios, along with the AUC. Agreement between the AI models and hand orthopaedists was assessed on the basis of percent agreement and Cohen’s kappa coefficient. Results A total of 1,548 wrist radiographs of DRF (Fernandez types 1 and 3) were collected; however, 79 cases were excluded due to abnormal pathology and low-quality films, leaving 1,469 images divided into 1,384 and 85 cases for the AI development and external testing phases, respectively (Fig. 1 ). The baseline characteristics of the included wrist radiographs verified by two hand orthopedists’ consensuses are presented in Table 1 . The accuracy results of the learning models are presented in Table 2 for the internal testing group of the AI development phase. For AIDER1, the highest accuracy models included ResNet-18 for ulnar styloid fractures, dorsal comminution, intra-articular fractures, and dorsal tilts exceeding 20 degrees, whereas GoogleNet performed best for metaphyseal comminution and radial shortening greater than 5 mm. Similarly, AIDER2 achieved its highest accuracy with ResNet-18 for Lafontaine criteria of 3 or more and unstable fractures. AIDER1 provided 3 independent models that reached the accuracy cutoff point: dorsal comminution, dorsal tilt greater than 20 degrees, and metaphyseal comminution. AIDER2 can predict up to 82% of unstable fractures. The most accurate models were chosen for further evaluation in the external testing group. Table 2 The maximum percentage of accuracy in predicting distal radius fracture (DRF) characteristics from the internal testing group achieved by AIDER models. Model Architecture AIDER1 AIDER2 Ulnar styloid fracture Dorsal comminution Dorsal tilt > 20 Intra-articular fracture Metaphyseal comminution Radial shortening > 5 mm Lafontaine ≥ 3/5 Unstable fracture ResNet-18 66.5 73.5 75.9 70.6 71.8 67.7 60.0 82.4 GoogLeNet 62.4 71.8 63.5 61.1 73.5 69.4 58.2 81.2 *The bold text identifies the optimal model configuration that yielded the best predictive accuracy In the external testing group, radial shortening provided the best accuracy at 80% (95% confidence interval (CI): 69.9, 87.9). The diagnostic ability of radial shortening was a sensitivity of 88.9% (95% CI: 77.4, 95.8), specificity of 64.5% (95% CI: 45.4, 80.8), positive predictive value of 81.4% (95% CI: 69.1, 90.3), negative predictive value of 76.9% (95% CI: 56.4, 91.0), likelihood ratio of positive tests of 2.51 (95% CI: 1.54, 40.6), likelihood ratio of negative tests of 0.17 (95% CI: 0.08, 0.38), and AUC of 0.77 (95% CI: 0.67, 0.86). AIDER1 for radial shortening detection had 35.48% false positives and 11.11% false negatives. For AIDER 2, unstable fracture was superior to the Lafontaine criteria, with a sensitivity of 100% (95% CI: 94.6, 100.0), specificity of 26.3% (95% CI: 9.1, 51.2), positive predictive value of 82.5% (95% CI: 72.4, 90.1), negative predictive value of 100.0% (47.8, 100.0), accuracy of 83.5% (95% CI: 73.9, 90.7), likelihood ratio of positive test results of 1.36 (95% CI: 1.04, 1.78), and AUC of 0.63 (0.53, 0.73) (Table 3 ). Table 3 Diagnostic ability of AIDER models in the external testing group AI, positive Hand orthopaedic consensus Sensitivity (95%CI) Specificity (95%CI) PPV (95%CI) NPV (95%CI) Accuracy (95%CI) LR+ (95%CI) LR- (95%CI) AUC (95%CI) OR (95%CI) Positive n (%) Negative n (%) AIDER1 Ulnar styloid fracture 31/37 (83.9) 14/48 (29.2) 83.8 (68.0,93.8) 70.8 (55.9,83.0) 68.9 (53.4,81.8) 85.0 (70.2,94.3) 76.5 (66.0,85.0) 2.87 (1.81,4.56) 0.23 (0.11,0.49) 0.77 (0.68,0.86) 12.55 (4.36,35.86) Dorsal comminution 58/61 (95.1) 17/24 (70.8) 95.1 (86.3,99.0) 29.2 (12.6,51.1) 77.3 (66.2,86.2) 70.0 (34.8,93.3) 76.4 (66.0,85.0) 1.34 (1.03,1.75) 0.17 (0.05,0.60) 0.62 (0.52,0.72) 7.96 (1.99,31.36) Intra-articular fracture 28/38 (73.7) 20/47 (42.6) 73.7 (56.9,86.6) 57.4 (42.2,71.7) 58.3 (43.2,72.4) 73.0 (55.9,86.2) 64.7 (53.6,74.8) 1.73 (1.18,2.54) 0.46 (0.25,0.82) 0.66 (0.55,0.76) 3.78 (1.51,9.42) Dorsal tilt > 20 degrees 17/31 (54.8) 5/54 (22.7) 58.4 (36.0,72.7) 90.7 (79.7,96.9) 77.3 (54.6,92.2) 77.8 (65.5,87.3) 77.6 (67.3,86.0) 5.92 (2.42,14.48) 0.50 (0.33,0.74) 0.73 (0.63,0.83) 11.90 (3.82,36.74) Metaphyseal comminution 1/24 (4.2) 2/61 (3.3) 4.2 (0.1,21.1) 96.7 (88.7,99.6) 33.3 (0.8,90.6) 72.0 (60.9,81.3) 70.5 (59.7,80.0) 1.27 (0.12,13.37) 0.99 (0.90,1.09) 0.50 (0.46,0.55) 1.28 (0.00,10.40) Radial shortening 48/54 (88.9) 11/31 (35.5) 88.9 (77.4,95.8) 64.5 (45.4,80.8) 81.4 (69.1,90.3) 76.9 (56.4,91.0) 80.0 (69.9,87.9) 2.51 (1.54,40.6) 0.17 (0.08,0.38) 0.77 (0.67,0.86) 14.55 (4.82,43.76) AIDER2 Lafontaine criteria ≥ 3/5 43/46 (93.5) 29/39 (74.4) 93.5 (82.1,98.6) 25.6 (13.0,42.1) 59.7 (47.5,71.1) 76.9 (46.2,95.0) 62.4 (51.2,72.6) 1.26 (1.03,1.53) 0.25 (0.08,0.86) 0.60 (0.52,0.67) 4.93 (1.33, 18.05) Unstable pattern 66/66 (100.0) 14/19 (73/7) 100.0 (94.6,100.0) 26.3 (9.1,51.2) 82.5 (72.4,90.1) 100.0 (47.8,100.0) 83.5 (73.9,90.7) 1.36 (1.04,1.78) 0.00 (-) 0.63 (0.53,0.73) 1 (5.67,-) PPV = positive predictive value, NPV = negative predictive value, LR + = likelihood ratio for a positive test, LR- = likelihood ratio for a negative test, AUC = area under the receiver operating characteristic curve, OR = odds ratio, CI = confidence interval. The Cohen kappa statistic between AIDER1 and hand specialist consensus was the best for radial shortening, at 0.55, with an agreement of 80% and a p value < 0.0001. For AIDER 2, the agreement with the hand specialist consensus for unstable fractures was 83.5%, with a kappa statistics of 0.3568 and a p value < 0.0001 (Table 4 ). Table 4 Agreements between hand specialist consensus and AIDER models in the external testing group Model Agreement (%) Expected agreement (%) Kappa Standard error P value AIDER1 Ulnar styloid fracture 76.47 49.62 0.5330 0.1066 20 degrees 77.65 56.53 0.4858 0.1052 < 0.0001 Metaphyseal comminution 70.59 70.23 0.0121 0.0605 0.4209 Radial shortening 80.00 55.25 0.5530 0.1075 < 0.0001 AIDER2 Lafontaine criteria ≥ 3/5 62.35 52.86 0.2014 0.0825 0.0073 Unstable fracture 83.53 74.39 0.3568 0.0830 < 0.0001 Discussion To our knowledge, no prior studies have used AI to predict treatment decisions of the basis of initial radiographs of DRF. In real-life situations, OpenAI’s ChatGPT may be consulted for recommendations on DRF treatment. However, a notable limitation is that 57% of the recommendations were inconsistent with those provided by orthopedic specialists [ 37 ]. This study addressed this gap by proposing AI models capable of facilitating treatment decision-making in clinical practice. Recent studies have increasingly explored the application of AI in interpreting radiographic images of DRF [ 27 – 33 , 37 – 40 ]. Compared with clinicians, AI has demonstrated comparable accuracy in detecting these fractures [ 29 , 30 , 32 , 33 ]. Ananda et al. evaluated multiple AI models and identified those with high accuracy in detecting abnormalities in distal radius radiographs [ 38 ]. Tobler et al. assessed the ability of AI to detect components of AO/OTA classifications, which demonstrated good accuracy compared with radiologists [ 28 ]. AIDER1 demonstrated poor performance in detecting intra-articular fractures, which is consistent with the findings of Tobler et al. [ 28 ]. This suboptimal outcome was explained by the use of the same model, ResNet-18, for training intra-articular fracture detection. To enhance performance, Min et al. used a double-stage DL algorithm which improved the AUC to 0.82 [ 39 ]. Oka et al. reported high accuracy (91%) when a different pretrained model (VGG16) was used for ulnar styloid fractures (AUC = 0.96) [ 27 ]. In this study, the distribution of operative factors was uneven, which may have influenced model training and performance [ 41 ]. As a result, AIDER1 showed variable accuracy across several parameters. Although AIDER2 achieved relatively high overall accuracy, the modest AUC suggests that its discriminative ability is still limited. These findings highlight the challenges of developing AI for complex fracture assessment and indicate that further refinement and broader datasets are needed before clinical implementation. The strengths of this study include the development of models trained by hand orthopedists using 1,469 wrist radiographs, which aid in treatment decision-making and address operative factors beyond the Lafontaine criteria. The limitations of this study should be recognized. First, the model's applicability is restricted to Fernandez Type 1 and Type 3 fractures, which commonly present uncertainty in operative treatment decisions. Second, we acknowledge the absence of a human interobserver diagnostic performance comparison; this crucial evaluation is reserved for the next phase of the study, following the planned performance optimization and refinement of the AIDER model. Third, the external dataset was small, but it provided independent data reflecting real-world clinical practice. Finally, the accuracy threshold should be considered preliminary and will require further validation against established clinical outcomes in future studies. Conclusion The AI models showed reasonable performance for some radiographic features of distal radius fractures but remained inconsistent for several key operative indicators. Future work should prioritize improving the data balance, refining the labeling of surgical factors, and validating the models across broader fracture patterns and clinical settings. With these developments, the system has the potential to become a more dependable support tool in fracture care management. Abbreviations AI: Artificial Intelligence AIDER: Artificial Intelligence Distal End Radius AO/OTA: Arbeitsgemeinschaft für Osteosynthesefragen/Orthopaedic Trauma Association AUC: Area under the receiver operating characteristic CI: Confidence Interval CNNs: Convolutional Neural Networks DL: Deep Learning DRF: Distal Radius Fracture MLMIC: Medical Liability Mutual Insurance Company PA: Posteroanterior PACS: Picture Archiving and Communication System Declarations Ethics approval and consent to participate This study was approved by the Institutional Review Board of the Faculty of Medicine Ramathibodi Hospital, Mahidol University and Lampang Hospital (COA. MURA 2022/486, and EC 002/67). Informed consent was waived due to the retrospective nature of the study. Consent for publication Not applicable Availability of data and materials The datasets generated and/or analyzed during the current study are not publicly available owing to patient privacy and institutional regulations but are available from the corresponding author upon reasonable request. Competing interests The authors declare that they have no competing interests. Funding This research received funding from the Faculty of Medicine Ramathibodi Hospital, Mahidol University. Authors’ contributions TJ collected the data, performed the analysis, and drafted the manuscript. TK interpreted the results and conceived and supervised the study. TT interpreted the results. DO contributed to AI model development. SJ contributed radiographs of the study. All the authors read and approved the final manuscript. Acknowledgments The authors thank the staff of Lampang Hospital for their support in radiograph for data collection. ChatGPT (OpenAI) and Curie were used to assist with language editing. The authors verified all the scientific content. References Court-Brown CM, Caesar B. Epidemiology of adult fractures: A review. Injury. 2006;37(8):691-7. Karl JW, Olson PR, Rosenwasser MP. The Epidemiology of Upper Extremity Fractures in the United States, 2009. J Orthop Trauma. 2015;29(8):e242-4. Bentohami A, Bosma J, Akkersdijk GJ, van Dijkman B, Goslings JC, Schep NW. Incidence and characteristics of distal radial fractures in an urban population in The Netherlands. Eur J Trauma Emerg Surg. 2014;40(3):357-61. Fernandez DL. Distal radius fracture: the rationale of a classification. Chir Main. 2001;20(6):411-25. Wu YS, Yang J, Xie LZ, Zhang JY, Yu XB, Hu W, et al. Factors associated with the decision for operative versus conservative treatment of displaced distal radius fractures in the elderly. ANZ J Surg. 2019;89(10):E428-E32. Belloti JC, Tamaoki MJ, Franciozi CE, Santos JB, Balbachevsky D, Chap Chap E, et al. Are distal radius fracture classifications reproducible? Intra and interobserver agreement. Sao Paulo Med J. 2008;126(3):180-5. Hruby LA, Haider T, Laggner R, Gahleitner C, Erhart J, Stoik W, et al. Standard radiographic assessments of distal radius fractures miss involvement of the distal radioulnar joint: a diagnostic study. Arch Orthop Trauma Surg. 2022;142(6):1075-82. Kucuk L, Kumbaraci M, Gunay H, Karapinar L, Ozdemir O. Reliability and reproducibility of classifications for distal radius fractures. Acta Orthop Traumatol Turc. 2013;47(3):153-7. Siripakarn Y, Niempoog S, Boontanapibul K. The comparative study of reliability and reproducibility of distal radius' fracture classification among: AO frykman and Fernandez classification systems. J Med Assoc Thai. 2013;96(1):52-7. Ilyas AM, Jupiter JB. Distal radius fractures--classification of treatment and indications for surgery. Orthop Clin North Am. 2007;38(2):167-73, v. Lafontaine M, Hardy D, Delince P. Stability assessment of distal radius fractures. Injury. 1989;20(4):208-10. Mackenney PJ, McQueen MM, Elton R. Prediction of instability in distal radial fractures. J Bone Joint Surg Am. 2006;88(9):1944-51. Mehta JA, Bain GI, Heptinstall RJ. Anatomical reduction of intra-articular fractures of the distal radius. An arthroscopically-assisted approach. J Bone Joint Surg Br. 2000;82(1):79-86. Nesbitt KS, Failla JM, Les C. Assessment of instability factors in adult distal radius fractures. J Hand Surg Am. 2004;29(6):1128-38. Walenkamp MM, Aydin S, Mulders MA, Goslings JC, Schep NW. Predictors of unstable distal radius fractures: a systematic review and meta-analysis. J Hand Surg Eur Vol. 2016;41(5):501-15. Guly HR. Diagnostic errors in an accident and emergency department. Emerg Med J. 2001;18(4):263-9. Hussain F, Cooper A, Carson-Stevens A, Donaldson L, Hibbert P, Hughes T, et al. Diagnostic error in the emergency department: learning from national patient safety incident report analysis. BMC Emerg Med. 2019;19(1):77. Hallas P, Ellingsen T. Errors in fracture diagnoses in the emergency department--characteristics of patients and diurnal variation. BMC Emerg Med. 2006;6:4. Pogue DJ, Viegas SF, Patterson RM, Peterson PD, Jenkins DK, Sweo TD, et al. Effects of distal radius fracture malunion on wrist joint mechanics. J Hand Surg Am. 1990;15(5):721-7. Prommersberger KJ, Pillukat T, Muhldorfer M, van Schoonhoven J. Malunion of the distal radius. Arch Orthop Trauma Surg. 2012;132(5):693-702. Bronstein AJ, Trumble TE, Tencer AF. The effects of distal radius fracture malalignment on forearm rotation: a cadaveric study. J Hand Surg Am. 1997;22(2):258-62. DeNoble PH, Marshall AC, Barron OA, Catalano LW, 3rd, Glickel SZ. Malpractice in distal radius fracture management: an analysis of closed claims. J Hand Surg Am. 2014;39(8):1480-8. Sandelin H, Waris E, Hirvensalo E, Vasenius J, Huhtala H, Raatikainen T, et al. Patient injury claims involving fractures of the distal radius. Acta Orthop. 2018;89(2):240-5. Panchmatia JR, Visenio MR, Panch T. The role of artificial intelligence in orthopaedic surgery. Br J Hosp Med (Lond). 2018;79(12):676-81. Oosterhoff JHF, Doornberg JN, Machine Learning C. Artificial intelligence in orthopaedics: false hope or not? A narrative review along the line of Gartner's hype cycle. EFORT Open Rev. 2020;5(10):593-603. Langerhuizen DWG, Janssen SJ, Mallee WH, van den Bekerom MPJ, Ring D, Kerkhoffs G, et al. What Are the Applications and Limitations of Artificial Intelligence for Fracture Detection and Classification in Orthopaedic Trauma Imaging? A Systematic Review. Clin Orthop Relat Res. 2019;477(11):2482-91. Oka K, Shiode R, Yoshii Y, Tanaka H, Iwahashi T, Murase T. Artificial intelligence to diagnosis distal radius fracture using biplane plain X-rays. J Orthop Surg Res. 2021;16(1):694. Tobler P, Cyriac J, Kovacs BK, Hofmann V, Sexauer R, Paciolla F, et al. AI-based detection and classification of distal radius fractures using low-effort data labeling: evaluation of applicability and effect of training set size. Eur Radiol. 2021;31(9):6816-24. Bluthgen C, Becker AS, Vittoria de Martini I, Meier A, Martini K, Frauenfelder T. Detection and localization of distal radius fractures: Deep learning system versus radiologists. Eur J Radiol. 2020;126:108925. Gan K, Xu D, Lin Y, Shen Y, Zhang T, Hu K, et al. Artificial intelligence detection of distal radius fractures: a comparison between the convolutional neural network and professional assessments. Acta Orthop. 2019;90(4):394-400. Kim DH, MacKinnon T. Artificial intelligence in fracture detection: transfer learning from deep convolutional neural networks. Clin Radiol. 2018;73(5):439-45. Lindsey R, Daluiski A, Chopra S, Lachapelle A, Mozer M, Sicular S, et al. Deep neural network improves fracture detection by clinicians. Proc Natl Acad Sci U S A. 2018;115(45):11591-6. Thian YL, Li Y, Jagmohan P, Sia D, Chan VEY, Tan RT. Convolutional Neural Networks for Automated Fracture Detection and Localization on Wrist Radiographs. Radiol Artif Intell. 2019;1(1):e180001. Medoff RJ. Essential radiographic evaluation for distal radius fractures. Hand Clin. 2005;21(3):279-88. Hardy DC, Totty WG, Reinus WR, Gilula LA. Posteroanterior wrist radiography: importance of arm positioning. J Hand Surg Am. 1987;12(4):504-8. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143(1):29-36. Knee CJ, Campbell RJ, Graham DJ, Handford C, Symes M, Sivakumar BS. Examining the role of ChatGPT in the management of distal radius fractures: insights into its accuracy and consistency. ANZ J Surg. 2024;94(7-8):1391-6. Ananda A, Ngan KH, Karabag C, Ter-Sarkisov A, Alonso E, Reyes-Aldasoro CC. Classification and Visualisation of Normal and Abnormal Radiographs; A Comparison between Eleven Convolutional Neural Network Architectures. Sensors (Basel). 2021;21(16). Min H, Rabi Y, Wadhawan A, Bourgeat P, Dowling J, White J, et al. Automatic classification of distal radius fracture using a two-stage ensemble deep learning framework. Phys Eng Sci Med. 2023;46(2):877-86. Raisuddin AM, Vaattovaara E, Nevalainen M, Nikki M, Jarvenpaa E, Makkonen K, et al. Critical evaluation of deep neural networks for wrist fracture detection. Sci Rep. 2021;11(1):6006. Norori N, Hu Q, Aellen FM, Faraci FD, Tzovara A. Addressing bias in big data and AI for health care: A call for open science. Patterns (N Y). 2021;2(10):100347. Tables Table 1 is available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table1.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8952141","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":604197828,"identity":"5b783632-67f9-4c2b-b633-46f90ff17184","order_by":0,"name":"Tharanas Jantharagsarangsee","email":"","orcid":"","institution":"Ministry of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Tharanas","middleName":"","lastName":"Jantharagsarangsee","suffix":""},{"id":604197829,"identity":"da08fec3-0561-450f-9831-dabef4abcaab","order_by":1,"name":"Thepparat Kanchanathepsak","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBElEQVRIiWNgGAWjYDACCTiLDYgrDoCZBx7g0cED08ID1nLmAJAB1JJAtBbGNogWBnxa7KWbjz0uqLknb8/Alvi5cN4dOXuxww+BttjJ6TbgsEXmWLrxjGPFhj0MbIelZ257ZswjnWYA1JJsbHYAl8NyzKR52BIYexjYG6R5tx1O7JFOAGk5kLgNp5b8b9I8/xLsgVqaf/POAWlJ/0BASw6bNG9bQiLQYcekeRtAWnII2HIjzdx4Zl9Ccs9htjRrnmOHjXlu5xQcSDDA7Rf2GcnPHhd8S7Btb28zvs1Tc1iOfXb65g8fKuzkcGkBAjZmMMWMImiAUzmSllEwCkbBKBgFuAAAq+lZpcsX1nYAAAAASUVORK5CYII=","orcid":"","institution":"Mahidol University","correspondingAuthor":true,"prefix":"","firstName":"Thepparat","middleName":"","lastName":"Kanchanathepsak","suffix":""},{"id":604197830,"identity":"61201c5d-82de-4c8f-a915-5fccec2e8ad9","order_by":2,"name":"Tulyapruek Tawonsawatruk","email":"","orcid":"","institution":"Mahidol University","correspondingAuthor":false,"prefix":"","firstName":"Tulyapruek","middleName":"","lastName":"Tawonsawatruk","suffix":""},{"id":604197831,"identity":"1a2ea539-7471-4b2f-8785-37b651afa52b","order_by":3,"name":"Dhammathat Owasirikul","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Dhammathat","middleName":"","lastName":"Owasirikul","suffix":""},{"id":604197832,"identity":"3e0cb86a-8bc6-4d26-b94f-563795ff12df","order_by":4,"name":"Supaneewan Jaovisidha","email":"","orcid":"","institution":"Mahidol University","correspondingAuthor":false,"prefix":"","firstName":"Supaneewan","middleName":"","lastName":"Jaovisidha","suffix":""}],"badges":[],"createdAt":"2026-02-24 03:24:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8952141/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8952141/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104554702,"identity":"0f0a8177-c185-4c64-957c-0b4248e0a7b5","added_by":"auto","created_at":"2026-03-13 08:57:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":283017,"visible":true,"origin":"","legend":"\u003cp\u003eWrist radiographs for AI model development and the external testing group\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8952141/v1/0e1f9bda648ebf705d3f4a01.png"},{"id":106698522,"identity":"02b341aa-73de-4f09-a693-1cdbf0cd2927","added_by":"auto","created_at":"2026-04-12 04:39:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1073682,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8952141/v1/af6cf6e7-bf47-4b1d-9b75-d4bc907dfce2.pdf"},{"id":104554806,"identity":"dfae095d-d8cb-4adc-b07d-e578a340434c","added_by":"auto","created_at":"2026-03-13 08:58:09","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":19671,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8952141/v1/c5d71cecd4354b0c48f25e8d.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Artificial intelligence for the assessment of distal radius fracture instability on wrist radiographs: A Diagnostic Study","fulltext":[{"header":"Background","content":"\u003cp\u003eDistal radius fractures (DRFs) are frequent injuries, constituting up to 18% of all fractures from accidents [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] and approximately 44% of forearm fractures [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The incidence rate is 20 per 10,000 person-years, with a higher prevalence among individuals aged 50 years and older [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Among our 706 patients aged 20 years and older, 90% had undergone surgery during the past ten years.\u003c/p\u003e \u003cp\u003eDRF is commonly classified by the mechanism of injury via the Fernandez classification [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], which aids in treatment decisions [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. This classification system has demonstrated reliable interobserver and intraobserver agreement [\u003cspan additionalcitationids=\"CR7 CR8\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. To identify instability patterns from initial radiographic assessment, the Lafontaine criteria (\u0026ge;\u0026thinsp;3 out of 5), metaphyseal comminution, and radial shortening\u0026thinsp;\u0026gt;\u0026thinsp;5 mm are pivotal in determining proper treatment [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan additionalcitationids=\"CR11 CR12 CR13 CR14\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMost patients with DRF, especially in Thailand, initially seek treatment from general practitioners or emergency physicians before referral to orthopedists. However, some cases are not appropriately transferred due to diagnostic errors, particularly in the misinterpretation of radiographic images [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Other factors contributing to these errors include communication issues, lack of experience in treatment, heavy workload, and fatigue, particularly during off-hours [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Consequently, complications such as malunion of the distal radius are common, leading to chronic wrist pain[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] and reduced wrist function [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. In the United States, 0.04% of DRF cases managed by orthopaedic surgeons insured by the Medical Liability Mutual Insurance Company (MLMIC) result in malpractice lawsuits annually [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTo address these challenges, this article proposes the use of artificial intelligence (AI) to assist in predicting operative versus nonoperative treatment of DRF on the basis of initial radiographic images. AI can help recognize key radiographic factors, thereby reducing diagnostic errors and approving therapeutic judgment [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. This application aims to enhance physicians' ability to deliver personalized patient care, increase confidence in treatment decisions, and mitigate biases in treatment selection [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. The utility of AI in orthopedic imaging has been widely established, demonstrating the ability to accurately identify fractures in major areas (e.g., humerus, hip, and ankle) [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] and aid in the diagnostic screening of DRF as effectively as expert surgeons do [\u003cspan additionalcitationids=\"CR28 CR29 CR30 CR31 CR32\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. However, a limitation frequently reported is a notable drop in performance when complex features such as intra-articular fracture and fracture displacement are being assessed [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Additionally, the other radiographic parameters from the Lafontaine criteria were not considered.\u003c/p\u003e \u003cp\u003eTo our knowledge, no previous studies have attempted to use AI to assist in operative decision-making on the basis of initial radiographs. The objectives of this study were to assess the accuracy of AI compared with the consensus of two hand orthopedists in identifying each radiographic component of the Lafontaine criteria (ulnar styloid fracture, dorsal comminution, intra-articular fracture, and dorsal tilt\u0026thinsp;\u0026gt;\u0026thinsp;20 degrees) and to predict stable or unstable DRF in initial radiographs of Fernandez types 1 and 3.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eInitial wrist radiographs (PA and lateral views) of DRF, including Fernandez type 1 (extra-articular metaphyseal fracture) and type 3 (intra-articular fracture) [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], were collected from the Picture Archiving and Communication System (PACS) of Ramathibodi Hospital and Lampang Hospital. Radiographs from patients older than 18 years between January 2007 and May 2022 were included. The radiographs of those with concomitant fractures other than ulnar styloid fractures, pathologic fractures, inadequate image quality, radiocarpal abnormalities, and the presence of foreign bodies or implants were excluded. All radiographs were anonymized by replacing patient-specific data with study numbers to ensure confidentiality and were subsequently converted into the JPEG format. An appropriate window was selected in all the images, and additional annotations, such as handedness labels, were removed to reduce the error in the AI learning process. This study was approved by the Institutional Review Board of both study hospitals (COA. MURA 2022/486, and EC 002/67).\u003c/p\u003e \u003cp\u003eDemographic data including age, sex, Fernandez classification[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], and treatment, were recorded. Two 12-year-experienced hand orthopedists evaluated and consensused each radiographic component of Lafontaine. We use a consensus method to establish the ground truth. This critical step reduces interrater variability and ensures labeling consistency, thereby preventing negative effects on the training accuracy of the AI model. The presence of metaphyseal comminution, radial shortening greater than 5 mm, or at least 3 out of 5 Lafontaine criteria was defined as an unstable fracture.\u003c/p\u003e \u003cp\u003eThe full dataset was then partitioned into two phases for AI model development and evaluation:\u003c/p\u003e \u003cp\u003ePhase I: AI Model Development and Internal Testing. This phase utilized a combined dataset for the creation of the AI distal end radius (AIDER), which was divided into training (70%), validation (15%), and internal testing (15%) groups.\u003c/p\u003e \u003cp\u003ePhase II: External Testing. This phase consisted of separate radiographs used to evaluate the final model's performance via a diagnostic study.\u003c/p\u003e \u003cp\u003eTwo AIDERs were developed to detect radiographic abnormalities and predict unstable fractures. AIDER1 was designed to be an analytical model to detect each radiographic component of the Lafontaine criteria, metaphyseal comminution and radial shortening greater than 5 mm. AIDER2 is a practical application model to predict unstable fractures of DRF, which are defined as Lafontaine criteria of 3 or more, metaphyseal comminution, or radial shortening greater than 5 mm.\u003c/p\u003e \u003cp\u003eCriteria for adequate radiographic images were set up. In the wrist posteroanterior (PA) view, adequate images clearly display the ulnar styloid at the edge of the bone in a peripheral orientation. In the lateral view of the wrist, the pisiform should lie between the volar surface of the capitate and the volar tuberosity of the scaphoid. Additional specific radiographic parameters were defined for the evaluation of DRF. On the PA view, radial shortening greater than 5 mm was defined as a radial height measurement of less than 6 mm. In the lateral view, dorsal comminution was defined as the presence of fragmented bone at the dorsal metaphysis of the distal radius. Dorsal tilt was defined as the angle formed between a line perpendicular to the long axis of the radius and a line drawn from the volar apex to the dorsal rim of the distal radius [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe AI models were developed via deep learning in several steps. First, image augmentation was applied to the radiographs by introducing a\u0026thinsp;\u0026plusmn;\u0026thinsp;10\u0026deg; rotation, \u0026plusmn;\u0026thinsp;0.5 scaling factor, and \u0026plusmn;\u0026thinsp;20 pixel shift to simulate variations that commonly occur in clinical practice and to improve model robustness. The model then learns relevant image features, such as bone edges and fracture patterns, through convolutional neural networks or CNNs (the core architecture uses mathematical filters to automatically and hierarchically extract features such as edges and fracture lines from image data)\u003c/p\u003e \u003cp\u003eTo improve efficiency, transfer learning was used by adapting pretrained network architectures (ResNet-18 and GoogLeNet) to the DRF dataset. The models were further refined using the study images to optimize the performance for this specific task. Finally, model performance was evaluated via standard diagnostic metrics, including accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC), and the best-performing models were selected for further analysis.\u003c/p\u003e \u003cp\u003eThe sample size of the external testing phase was calculated via MedCalc Statistical Software version 19.2.6 (MedCalc Software, Ostend, Belgium; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.medcalc.org\u003c/span\u003e\u003cspan address=\"https://www.medcalc.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e; 2020) [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. The parameters used were set as follows: null hypothesis value\u0026thinsp;=\u0026thinsp;0.5, AUC or accuracy threshold\u0026thinsp;=\u0026thinsp;0.725, alpha error\u0026thinsp;=\u0026thinsp;0.05, beta error\u0026thinsp;=\u0026thinsp;0.2, ratio of negative/positive radiographic factor\u0026thinsp;=\u0026thinsp;1, and ratio of stable/unstable fracture\u0026thinsp;=\u0026thinsp;3:7 in our real-life database. The AIDER1 model accounted for 48 wrist radiographs: 24 for each negative and positive finding. A total of 56 wrist radiographs (17 stable and 39 unstable fractures) were needed for the AIDER2 model. The final dataset was substantially larger than the number required to detect moderate discriminative ability under these assumptions for early-stage model evaluation.\u003c/p\u003e \u003cp\u003e The data were recorded in Microsoft Excel and analyzed via STATA 18.0 (StataCorp, College Station, Texas, USA). Categorical variables were compared via the chi-square test, and continuous variables were summarized via means and standard deviations. Diagnostic performance was evaluated by calculating the sensitivity, specificity, accuracy, positive predictive value, negative predictive value, and positive and negative likelihood ratios, along with the AUC. Agreement between the AI models and hand orthopaedists was assessed on the basis of percent agreement and Cohen\u0026rsquo;s kappa coefficient.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eA total of 1,548 wrist radiographs of DRF (Fernandez types 1 and 3) were collected; however, 79 cases were excluded due to abnormal pathology and low-quality films, leaving 1,469 images divided into 1,384 and 85 cases for the AI development and external testing phases, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The baseline characteristics of the included wrist radiographs verified by two hand orthopedists\u0026rsquo; consensuses are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe accuracy results of the learning models are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e for the internal testing group of the AI development phase. For AIDER1, the highest accuracy models included ResNet-18 for ulnar styloid fractures, dorsal comminution, intra-articular fractures, and dorsal tilts exceeding 20 degrees, whereas GoogleNet performed best for metaphyseal comminution and radial shortening greater than 5 mm. Similarly, AIDER2 achieved its highest accuracy with ResNet-18 for Lafontaine criteria of 3 or more and unstable fractures. AIDER1 provided 3 independent models that reached the accuracy cutoff point: dorsal comminution, dorsal tilt greater than 20 degrees, and metaphyseal comminution. AIDER2 can predict up to 82% of unstable fractures. The most accurate models were chosen for further evaluation in the external testing group.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe maximum percentage of accuracy in predicting distal radius fracture (DRF) characteristics from the internal testing group achieved by AIDER models.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003cp\u003eArchitecture\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e \u003cp\u003eAIDER1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003eAIDER2\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUlnar styloid fracture\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDorsal comminution\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDorsal tilt\u0026thinsp;\u0026gt;\u0026thinsp;20\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIntra-articular fracture\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMetaphyseal comminution\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eRadial shortening\u0026thinsp;\u0026gt;\u0026thinsp;5 mm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eLafontaine\u0026thinsp;\u0026ge;\u0026thinsp;3/5\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eUnstable fracture\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResNet-18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e66.5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e73.5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e75.9\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e70.6\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e71.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e67.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e60.0\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e\u003cb\u003e82.4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGoogLeNet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e62.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e63.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e61.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e73.5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e69.4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e58.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e81.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003e*The bold text identifies the optimal model configuration that yielded the best predictive accuracy\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn the external testing group, radial shortening provided the best accuracy at 80% (95% confidence interval (CI): 69.9, 87.9). The diagnostic ability of radial shortening was a sensitivity of 88.9% (95% CI: 77.4, 95.8), specificity of 64.5% (95% CI: 45.4, 80.8), positive predictive value of 81.4% (95% CI: 69.1, 90.3), negative predictive value of 76.9% (95% CI: 56.4, 91.0), likelihood ratio of positive tests of 2.51 (95% CI: 1.54, 40.6), likelihood ratio of negative tests of 0.17 (95% CI: 0.08, 0.38), and AUC of 0.77 (95% CI: 0.67, 0.86). AIDER1 for radial shortening detection had 35.48% false positives and 11.11% false negatives. For AIDER 2, unstable fracture was superior to the Lafontaine criteria, with a sensitivity of 100% (95% CI: 94.6, 100.0), specificity of 26.3% (95% CI: 9.1, 51.2), positive predictive value of 82.5% (95% CI: 72.4, 90.1), negative predictive value of 100.0% (47.8, 100.0), accuracy of 83.5% (95% CI: 73.9, 90.7), likelihood ratio of positive test results of 1.36 (95% CI: 1.04, 1.78), and AUC of 0.63 (0.53, 0.73) (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDiagnostic ability of AIDER models in the external testing group\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"12\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAI, positive\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eHand orthopaedic consensus\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSensitivity (95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSpecificity (95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003cp\u003e(95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003cp\u003e(95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAccuracy (95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLR+\u003c/p\u003e \u003cp\u003e(95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLR-\u003c/p\u003e \u003cp\u003e(95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003cp\u003e(95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eOR\u003c/p\u003e \u003cp\u003e(95%CI)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003cp\u003en (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003cp\u003en (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAIDER1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUlnar styloid fracture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e31/37 (83.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14/48 (29.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e83.8 (68.0,93.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e70.8 (55.9,83.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e68.9 (53.4,81.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e85.0 (70.2,94.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e76.5 (66.0,85.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e2.87 (1.81,4.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.23 (0.11,0.49)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.77 (0.68,0.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e12.55 (4.36,35.86)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDorsal comminution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e58/61 (95.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17/24 (70.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e95.1 (86.3,99.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e29.2 (12.6,51.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e77.3 (66.2,86.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e70.0 (34.8,93.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e76.4 (66.0,85.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.34 (1.03,1.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.17 (0.05,0.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.62 (0.52,0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e7.96 (1.99,31.36)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntra-articular fracture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28/38 (73.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20/47 (42.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e73.7 (56.9,86.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e57.4 (42.2,71.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e58.3 (43.2,72.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e73.0 (55.9,86.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e64.7 (53.6,74.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.73 (1.18,2.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.46 (0.25,0.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.66 (0.55,0.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e3.78 (1.51,9.42)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDorsal tilt\u0026thinsp;\u0026gt;\u0026thinsp;20 degrees\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17/31 (54.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5/54 (22.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e58.4 (36.0,72.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e90.7 (79.7,96.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e77.3 (54.6,92.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e77.8 (65.5,87.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e77.6 (67.3,86.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e5.92 (2.42,14.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.50 (0.33,0.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.73 (0.63,0.83)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e11.90 (3.82,36.74)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetaphyseal comminution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1/24\u003c/p\u003e \u003cp\u003e(4.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2/61\u003c/p\u003e \u003cp\u003e(3.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.2 (0.1,21.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e96.7 (88.7,99.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e33.3 (0.8,90.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e72.0 (60.9,81.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e70.5 (59.7,80.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.27 (0.12,13.37)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.99 (0.90,1.09)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.50 (0.46,0.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e1.28 (0.00,10.40)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRadial shortening\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e48/54 (88.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11/31 (35.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e88.9 (77.4,95.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e64.5 (45.4,80.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e81.4 (69.1,90.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e76.9 (56.4,91.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e80.0 (69.9,87.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e2.51 (1.54,40.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.17 (0.08,0.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.77 (0.67,0.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e14.55 (4.82,43.76)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAIDER2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLafontaine criteria\u0026thinsp;\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e\u0026ge;\u003c/span\u003e\u0026thinsp;3/5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e43/46 (93.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29/39 (74.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e93.5 (82.1,98.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e25.6 (13.0,42.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e59.7 (47.5,71.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e76.9 (46.2,95.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e62.4 (51.2,72.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.26 (1.03,1.53)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.25 (0.08,0.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.60 (0.52,0.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e4.93\u003c/p\u003e \u003cp\u003e(1.33, 18.05)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnstable pattern\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66/66 (100.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14/19 (73/7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100.0 (94.6,100.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e26.3 (9.1,51.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e82.5 (72.4,90.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e100.0 (47.8,100.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e83.5 (73.9,90.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.36 (1.04,1.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003cp\u003e(-)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.63 (0.53,0.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e1\u003c/p\u003e \u003cp\u003e(5.67,-)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"12\"\u003ePPV\u0026thinsp;=\u0026thinsp;positive predictive value, NPV\u0026thinsp;=\u0026thinsp;negative predictive value, LR\u0026thinsp;+\u0026thinsp;=\u0026thinsp;likelihood ratio for a positive test, LR- = likelihood ratio for a negative test, AUC\u0026thinsp;=\u0026thinsp;area under the receiver operating characteristic curve, OR\u0026thinsp;=\u0026thinsp;odds ratio, CI\u0026thinsp;=\u0026thinsp;confidence interval.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe Cohen kappa statistic between AIDER1 and hand specialist consensus was the best for radial shortening, at 0.55, with an agreement of 80% and a p value\u0026thinsp;\u0026lt;\u0026thinsp;0.0001. For AIDER 2, the agreement with the hand specialist consensus for unstable fractures was 83.5%, with a kappa statistics of 0.3568 and a p value\u0026thinsp;\u0026lt;\u0026thinsp;0.0001 (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAgreements between hand specialist consensus and AIDER models in the external testing group\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAgreement (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eExpected agreement (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eKappa\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eStandard error\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAIDER1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUlnar styloid fracture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e76.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e49.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.5330\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.1066\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.0001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDorsal comminution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e76.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e66.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.2946\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.0943\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.0009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntra-articular fracture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e64.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e49.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.3037\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.1055\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.0020\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDorsal tilt\u0026thinsp;\u0026gt;\u0026thinsp;20 degrees\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e77.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e56.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.4858\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.1052\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.0001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetaphyseal comminution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e70.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.0121\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.0605\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.4209\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRadial shortening\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e80.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e55.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.5530\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.1075\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.0001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAIDER2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLafontaine criteria\u0026thinsp;\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e\u0026ge;\u003c/span\u003e\u0026thinsp;3/5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e62.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e52.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.2014\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.0825\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.0073\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnstable fracture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e83.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e74.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.3568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.0830\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.0001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eTo our knowledge, no prior studies have used AI to predict treatment decisions of the basis of initial radiographs of DRF. In real-life situations, OpenAI\u0026rsquo;s ChatGPT may be consulted for recommendations on DRF treatment. However, a notable limitation is that 57% of the recommendations were inconsistent with those provided by orthopedic specialists [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. This study addressed this gap by proposing AI models capable of facilitating treatment decision-making in clinical practice.\u003c/p\u003e \u003cp\u003eRecent studies have increasingly explored the application of AI in interpreting radiographic images of DRF [\u003cspan additionalcitationids=\"CR28 CR29 CR30 CR31 CR32\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan additionalcitationids=\"CR38 CR39\" citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Compared with clinicians, AI has demonstrated comparable accuracy in detecting these fractures [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Ananda et al. evaluated multiple AI models and identified those with high accuracy in detecting abnormalities in distal radius radiographs [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Tobler et al. assessed the ability of AI to detect components of AO/OTA classifications, which demonstrated good accuracy compared with radiologists [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAIDER1 demonstrated poor performance in detecting intra-articular fractures, which is consistent with the findings of Tobler et al. [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. This suboptimal outcome was explained by the use of the same model, ResNet-18, for training intra-articular fracture detection. To enhance performance, Min et al. used a double-stage DL algorithm which improved the AUC to 0.82 [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Oka et al. reported high accuracy (91%) when a different pretrained model (VGG16) was used for ulnar styloid fractures (AUC\u0026thinsp;=\u0026thinsp;0.96) [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn this study, the distribution of operative factors was uneven, which may have influenced model training and performance [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. As a result, AIDER1 showed variable accuracy across several parameters. Although AIDER2 achieved relatively high overall accuracy, the modest AUC suggests that its discriminative ability is still limited. These findings highlight the challenges of developing AI for complex fracture assessment and indicate that further refinement and broader datasets are needed before clinical implementation.\u003c/p\u003e \u003cp\u003eThe strengths of this study include the development of models trained by hand orthopedists using 1,469 wrist radiographs, which aid in treatment decision-making and address operative factors beyond the Lafontaine criteria. The limitations of this study should be recognized. First, the model's applicability is restricted to Fernandez Type 1 and Type 3 fractures, which commonly present uncertainty in operative treatment decisions. Second, we acknowledge the absence of a human interobserver diagnostic performance comparison; this crucial evaluation is reserved for the next phase of the study, following the planned performance optimization and refinement of the AIDER model. Third, the external dataset was small, but it provided independent data reflecting real-world clinical practice. Finally, the accuracy threshold should be considered preliminary and will require further validation against established clinical outcomes in future studies.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe AI models showed reasonable performance for some radiographic features of distal radius fractures but remained inconsistent for several key operative indicators. Future work should prioritize improving the data balance, refining the labeling of surgical factors, and validating the models across broader fracture patterns and clinical settings. With these developments, the system has the potential to become a more dependable support tool in fracture care management.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eAI: Artificial Intelligence\u003c/p\u003e\n\u003cp\u003eAIDER: Artificial Intelligence Distal End Radius\u003c/p\u003e\n\u003cp\u003eAO/OTA: \u003cstrong\u003eArbeitsgemeinschaft für Osteosynthesefragen/Orthopaedic Trauma Association\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAUC: Area under the receiver operating characteristic\u003c/p\u003e\n\u003cp\u003eCI: Confidence Interval\u003c/p\u003e\n\u003cp\u003eCNNs: Convolutional Neural Networks\u003c/p\u003e\n\u003cp\u003eDL: Deep Learning\u003c/p\u003e\n\u003cp\u003eDRF: Distal Radius Fracture\u003c/p\u003e\n\u003cp\u003eMLMIC: Medical Liability Mutual Insurance Company\u003c/p\u003e\n\u003cp\u003ePA: Posteroanterior\u003c/p\u003e\n\u003cp\u003ePACS: Picture Archiving and Communication System\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Institutional Review Board of the Faculty of Medicine Ramathibodi Hospital, Mahidol University and Lampang Hospital (COA. MURA 2022/486, and EC 002/67). Informed consent was waived due to the retrospective nature of the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are not publicly available owing to patient privacy and institutional regulations but are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research received funding from the Faculty of Medicine Ramathibodi Hospital, Mahidol University.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’ contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTJ collected the data, performed the analysis, and drafted the manuscript. TK interpreted the results and conceived and supervised the study. TT interpreted the results. DO contributed to AI model development. SJ contributed radiographs of the study. All the authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank the staff of Lampang Hospital for their support in radiograph for data collection. ChatGPT (OpenAI) and Curie were used to assist with language editing. The authors verified all the scientific content.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eCourt-Brown CM, Caesar B. Epidemiology of adult fractures: A review. Injury. 2006;37(8):691-7.\u003c/li\u003e\n\u003cli\u003eKarl JW, Olson PR, Rosenwasser MP. The Epidemiology of Upper Extremity Fractures in the United States, 2009. J Orthop Trauma. 2015;29(8):e242-4.\u003c/li\u003e\n\u003cli\u003eBentohami A, Bosma J, Akkersdijk GJ, van Dijkman B, Goslings JC, Schep NW. Incidence and characteristics of distal radial fractures in an urban population in The Netherlands. Eur J Trauma Emerg Surg. 2014;40(3):357-61.\u003c/li\u003e\n\u003cli\u003eFernandez DL. Distal radius fracture: the rationale of a classification. Chir Main. 2001;20(6):411-25.\u003c/li\u003e\n\u003cli\u003eWu YS, Yang J, Xie LZ, Zhang JY, Yu XB, Hu W, et al. Factors associated with the decision for operative versus conservative treatment of displaced distal radius fractures in the elderly. ANZ J Surg. 2019;89(10):E428-E32.\u003c/li\u003e\n\u003cli\u003eBelloti JC, Tamaoki MJ, Franciozi CE, Santos JB, Balbachevsky D, Chap Chap E, et al. Are distal radius fracture classifications reproducible? Intra and interobserver agreement. Sao Paulo Med J. 2008;126(3):180-5.\u003c/li\u003e\n\u003cli\u003eHruby LA, Haider T, Laggner R, Gahleitner C, Erhart J, Stoik W, et al. Standard radiographic assessments of distal radius fractures miss involvement of the distal radioulnar joint: a diagnostic study. Arch Orthop Trauma Surg. 2022;142(6):1075-82.\u003c/li\u003e\n\u003cli\u003eKucuk L, Kumbaraci M, Gunay H, Karapinar L, Ozdemir O. Reliability and reproducibility of classifications for distal radius fractures. Acta Orthop Traumatol Turc. 2013;47(3):153-7.\u003c/li\u003e\n\u003cli\u003eSiripakarn Y, Niempoog S, Boontanapibul K. The comparative study of reliability and reproducibility of distal radius\u0026apos; fracture classification among: AO frykman and Fernandez classification systems. J Med Assoc Thai. 2013;96(1):52-7.\u003c/li\u003e\n\u003cli\u003eIlyas AM, Jupiter JB. Distal radius fractures--classification of treatment and indications for surgery. Orthop Clin North Am. 2007;38(2):167-73, v.\u003c/li\u003e\n\u003cli\u003eLafontaine M, Hardy D, Delince P. Stability assessment of distal radius fractures. Injury. 1989;20(4):208-10.\u003c/li\u003e\n\u003cli\u003eMackenney PJ, McQueen MM, Elton R. Prediction of instability in distal radial fractures. J Bone Joint Surg Am. 2006;88(9):1944-51.\u003c/li\u003e\n\u003cli\u003eMehta JA, Bain GI, Heptinstall RJ. Anatomical reduction of intra-articular fractures of the distal radius. An arthroscopically-assisted approach. J Bone Joint Surg Br. 2000;82(1):79-86.\u003c/li\u003e\n\u003cli\u003eNesbitt KS, Failla JM, Les C. Assessment of instability factors in adult distal radius fractures. J Hand Surg Am. 2004;29(6):1128-38.\u003c/li\u003e\n\u003cli\u003eWalenkamp MM, Aydin S, Mulders MA, Goslings JC, Schep NW. Predictors of unstable distal radius fractures: a systematic review and meta-analysis. J Hand Surg Eur Vol. 2016;41(5):501-15.\u003c/li\u003e\n\u003cli\u003eGuly HR. Diagnostic errors in an accident and emergency department. Emerg Med J. 2001;18(4):263-9.\u003c/li\u003e\n\u003cli\u003eHussain F, Cooper A, Carson-Stevens A, Donaldson L, Hibbert P, Hughes T, et al. Diagnostic error in the emergency department: learning from national patient safety incident report analysis. BMC Emerg Med. 2019;19(1):77.\u003c/li\u003e\n\u003cli\u003eHallas P, Ellingsen T. Errors in fracture diagnoses in the emergency department--characteristics of patients and diurnal variation. BMC Emerg Med. 2006;6:4.\u003c/li\u003e\n\u003cli\u003ePogue DJ, Viegas SF, Patterson RM, Peterson PD, Jenkins DK, Sweo TD, et al. Effects of distal radius fracture malunion on wrist joint mechanics. J Hand Surg Am. 1990;15(5):721-7.\u003c/li\u003e\n\u003cli\u003ePrommersberger KJ, Pillukat T, Muhldorfer M, van Schoonhoven J. Malunion of the distal radius. Arch Orthop Trauma Surg. 2012;132(5):693-702.\u003c/li\u003e\n\u003cli\u003eBronstein AJ, Trumble TE, Tencer AF. The effects of distal radius fracture malalignment on forearm rotation: a cadaveric study. J Hand Surg Am. 1997;22(2):258-62.\u003c/li\u003e\n\u003cli\u003eDeNoble PH, Marshall AC, Barron OA, Catalano LW, 3rd, Glickel SZ. Malpractice in distal radius fracture management: an analysis of closed claims. J Hand Surg Am. 2014;39(8):1480-8.\u003c/li\u003e\n\u003cli\u003eSandelin H, Waris E, Hirvensalo E, Vasenius J, Huhtala H, Raatikainen T, et al. Patient injury claims involving fractures of the distal radius. Acta Orthop. 2018;89(2):240-5.\u003c/li\u003e\n\u003cli\u003ePanchmatia JR, Visenio MR, Panch T. The role of artificial intelligence in orthopaedic surgery. Br J Hosp Med (Lond). 2018;79(12):676-81.\u003c/li\u003e\n\u003cli\u003eOosterhoff JHF, Doornberg JN, Machine Learning C. Artificial intelligence in orthopaedics: false hope or not? A narrative review along the line of Gartner\u0026apos;s hype cycle. EFORT Open Rev. 2020;5(10):593-603.\u003c/li\u003e\n\u003cli\u003eLangerhuizen DWG, Janssen SJ, Mallee WH, van den Bekerom MPJ, Ring D, Kerkhoffs G, et al. What Are the Applications and Limitations of Artificial Intelligence for Fracture Detection and Classification in Orthopaedic Trauma Imaging? A Systematic Review. Clin Orthop Relat Res. 2019;477(11):2482-91.\u003c/li\u003e\n\u003cli\u003eOka K, Shiode R, Yoshii Y, Tanaka H, Iwahashi T, Murase T. Artificial intelligence to diagnosis distal radius fracture using biplane plain X-rays. J Orthop Surg Res. 2021;16(1):694.\u003c/li\u003e\n\u003cli\u003eTobler P, Cyriac J, Kovacs BK, Hofmann V, Sexauer R, Paciolla F, et al. AI-based detection and classification of distal radius fractures using low-effort data labeling: evaluation of applicability and effect of training set size. Eur Radiol. 2021;31(9):6816-24.\u003c/li\u003e\n\u003cli\u003eBluthgen C, Becker AS, Vittoria de Martini I, Meier A, Martini K, Frauenfelder T. Detection and localization of distal radius fractures: Deep learning system versus radiologists. Eur J Radiol. 2020;126:108925.\u003c/li\u003e\n\u003cli\u003eGan K, Xu D, Lin Y, Shen Y, Zhang T, Hu K, et al. Artificial intelligence detection of distal radius fractures: a comparison between the convolutional neural network and professional assessments. Acta Orthop. 2019;90(4):394-400.\u003c/li\u003e\n\u003cli\u003eKim DH, MacKinnon T. Artificial intelligence in fracture detection: transfer learning from deep convolutional neural networks. Clin Radiol. 2018;73(5):439-45.\u003c/li\u003e\n\u003cli\u003eLindsey R, Daluiski A, Chopra S, Lachapelle A, Mozer M, Sicular S, et al. Deep neural network improves fracture detection by clinicians. Proc Natl Acad Sci U S A. 2018;115(45):11591-6.\u003c/li\u003e\n\u003cli\u003eThian YL, Li Y, Jagmohan P, Sia D, Chan VEY, Tan RT. Convolutional Neural Networks for Automated Fracture Detection and Localization on Wrist Radiographs. Radiol Artif Intell. 2019;1(1):e180001.\u003c/li\u003e\n\u003cli\u003eMedoff RJ. Essential radiographic evaluation for distal radius fractures. Hand Clin. 2005;21(3):279-88.\u003c/li\u003e\n\u003cli\u003eHardy DC, Totty WG, Reinus WR, Gilula LA. Posteroanterior wrist radiography: importance of arm positioning. J Hand Surg Am. 1987;12(4):504-8.\u003c/li\u003e\n\u003cli\u003eHanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143(1):29-36.\u003c/li\u003e\n\u003cli\u003eKnee CJ, Campbell RJ, Graham DJ, Handford C, Symes M, Sivakumar BS. Examining the role of ChatGPT in the management of distal radius fractures: insights into its accuracy and consistency. ANZ J Surg. 2024;94(7-8):1391-6.\u003c/li\u003e\n\u003cli\u003eAnanda A, Ngan KH, Karabag C, Ter-Sarkisov A, Alonso E, Reyes-Aldasoro CC. Classification and Visualisation of Normal and Abnormal Radiographs; A Comparison between Eleven Convolutional Neural Network Architectures. Sensors (Basel). 2021;21(16).\u003c/li\u003e\n\u003cli\u003eMin H, Rabi Y, Wadhawan A, Bourgeat P, Dowling J, White J, et al. Automatic classification of distal radius fracture using a two-stage ensemble deep learning framework. Phys Eng Sci Med. 2023;46(2):877-86.\u003c/li\u003e\n\u003cli\u003eRaisuddin AM, Vaattovaara E, Nevalainen M, Nikki M, Jarvenpaa E, Makkonen K, et al. Critical evaluation of deep neural networks for wrist fracture detection. Sci Rep. 2021;11(1):6006.\u003c/li\u003e\n\u003cli\u003eNorori N, Hu Q, Aellen FM, Faraci FD, Tzovara A. Addressing bias in big data and AI for health care: A call for open science. Patterns (N Y). 2021;2(10):100347.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 is available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Diagnostic accuracy, Lafontaine criteria, metaphyseal comminution, radial shortening, unstable fracture pattern","lastPublishedDoi":"10.21203/rs.3.rs-8952141/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8952141/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eMisinterpretation of radiographs of distal radius fractures (DRFs) can lead to malalignment and hand and wrist disability. The application of artificial intelligence (AI) in detecting and predicting fracture instability still needs to be explored. This study aimed to evaluate the diagnostic accuracy of AI in identifying correlated factors, as well as in predicting unstable DRF, Fernandez types 1 and 3.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eDRF radiographs (Fernandez 1 and 3) of adults aged\u0026thinsp;\u0026gt;\u0026thinsp;18 years were retrieved from a university hospital and a provincial hospital between January 2017 and May 2022. Radiographs with any concomitant fracture, inadequate imaging, or radiocarpal pathology were excluded. Unstable fracture indicated operative intervention one the basis of 1) the Lafontaine criteria, and 2) radiographic operative factors (metaphyseal comminution and radial shortening). Two AI distal end radius (AIDER) models, AIDER1, identified each radiographic Lafontaine criterion and operative factor, and AIDER2, which predicts Lafontaine criteria\u0026thinsp;\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e\u0026ge;\u003c/span\u003e\u0026thinsp;3 or unstable fractures, were trained, validated, and tested (70:15:15 ratio).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eAmong the 1,548 initial radiographs, 1,469 films were eligible. The patients\u0026rsquo; characteristics were as follows: 694 (47%) were elderly, 984 (67%) were female, and 840 (57%) were Fernandez type 3. Diagnostic accuracy of the testing groups: AIDER1 achieved the accuracy threshold in dorsal comminution (74%), dorsal tilt\u0026thinsp;\u0026gt;\u0026thinsp;20 degrees (76%), and metaphyseal comminution (74%). This model provided low accuracy for ulnar styloid fractures, intra-articular fractures, and radial shortening (67%-71%). For AIDER2, the detection of unstable fractures achieved an accuracy of 82%, with an area under the curve (AUC) of 0.63 (95% CI: 0.53, 0.73), suggesting moderate discriminative ability.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe AI models performed well for selected radiographic features but showed limited accuracy for several operation-related findings. These results indicate that the system may assist with preliminary assessment, but further refinement, broader training data, and external validation are needed before it can be used to support treatment decisions reliably.\u003c/p\u003e","manuscriptTitle":"Artificial intelligence for the assessment of distal radius fracture instability on wrist radiographs: A Diagnostic Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-13 08:56:42","doi":"10.21203/rs.3.rs-8952141/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3408ba81-8129-4ddc-aa35-f2b2002194e5","owner":[],"postedDate":"March 13th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-12T04:38:59+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-13 08:56:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8952141","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8952141","identity":"rs-8952141","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}