Background: Deep vein thrombosis (DVT) following total knee arthroplasty (TKA) remains a major postoperative complication. Conventional risk assessment tools such as the Caprini score show limited predictive performance (AUC 0.65–0.72). We developed a multimodal deep learning framework that integrates early postoperative radiographs with clinical data to improve DVT risk stratification.
Methods: This single-center retrospective cohort study analyzed 1200 patients undergoing primary unilateral TKA (DVT incidence: 18.0%) from January 2018 to June 2023. A dual-branch architecture was implemented: ViT-B/16 or ResNet50 processed standardized radiographs (metal artifact suppression; 512 × 512 resolution), while Clinical-BERT encoded structured electronic medical record (EMR) representations. A dynamic attention fusion module generated sample-specific modality weights (α) for feature fusion (α·v_img + (1 − α)·v_txt). Performance was evaluated on a stratified test set (n = 180) against the Caprini score, unimodal models, and a nonattention concatenation fusion baseline. AUCs were compared using the DeLong test, and sensitivity/specificity differences were assessed using the McNemar test.
Results: The multimodal model achieved an AUC of 0.89 (95% CI: 0.85–0.92; DeLong p < 0.001 vs. all comparators), sensitivity 83% (76%–89%), specificity 88% (82%–93%), and F1-score 0.79. It identified 91.7% (11/12) of DVT cases labeled low-risk by Caprini. The learned attention coefficient α suggested modality reliance patterns associated with specific findings (e.g., higher α in cases with prosthesis–bone lucency > 2 mm; higher text contribution in cases with D-dimer > 1.0 mg/L), but these associations do not establish causality. Anatomical heatmaps consistently localized high-risk regions, particularly the popliteal vein projection area and tibial interface.
Conclusion: This attention-based multimodal model integrates biomechanical (radiographic) and biochemical/clinical (EMR-derived) signals to improve early postoperative DVT risk prediction while providing interpretable anatomical cues.
| [1] |
Y. Falck-Ytter, C. W. Francis, N. A. Johanson, et al., “Prevention of VTE in Orthopedic Surgery Patients: Antithrombotic Therapy and Prevention of Thrombosis, 9th Ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines,” Chest 141, no. 2 Suppl (2012): e278S–e325S.
|
| [2] |
V. E. Bergstein, W. L. Taylor, IV, A. I. Weinblatt, L. C. Lesser, and W. J. Long, “The Financial Burden of Aspirin Versus Oral Factor Xa Inhibitors for Thromboprophylaxis Following Total Knee Arthroplasty,” Journal of Arthroplasty 39, no. 4 (2024): 935–940.
|
| [3] |
H. X. Zhou, L. Q. Peng, Y. Yan, et al., “Validation of the Caprini Risk Assessment Model in Chinese Hospitalized Patients With Venous Thromboembolism,” Thrombosis Research 130, no. 5 (2012): 735–740.
|
| [4] |
G. Agarwal, A. Hajra, S. Chakraborty, et al., “Predictors and Mortality Risk of Venous Thromboembolism in Patients With COVID-19: Systematic Review and Meta-Analysis of Observational Studies,” Therapeutic Advances in Cardiovascular Disease 16 (2022): 17539447221105013.
|
| [5] |
P. A. Cavendish, J. S. Everhart, N. J. Peters, M. F. Sommerfeldt, and D. C. Flanigan, “Osteochondral Allograft Transplantation for Knee Cartilage and Osteochondral Defects: A Review of Indications, Technique,” Rehabilitation, and Outcomes 7, no. 6 (2019): e7.
|
| [6] |
F. Guo, T. Shashikiran, X. Chen, L. Yang, X. Liu, and L. Song, “Clinical Features and Risk Factor Analysis for Lower Extremity Deep Venous Thrombosis in Chinese Neurosurgical Patients,” Journal of Neurosciences in Rural Practice 6, no. 4 (2015): 471–476.
|
| [7] |
M. M. Bern, D. Hazel, E. Deeran, et al., “Low Dose Compared to Variable Dose Warfarin and to Fondaparinux as Prophylaxis for Thromboembolism After Elective Hip or Knee Replacement Surgery; a Randomized, Prospective Study,” Thrombosis Journal 13 (2015): 32.
|
| [8] |
D. Senanayake, S. Seneviratne, M. Imani, et al., “Classification of Fracture Risk in Fallers Using Dual-Energy X-Ray Absorptiometry (DXA) Images and Deep Learning-Based Feature Extraction,” JBMR Plus 7, no. 12 (2023): e10828.
|
| [9] |
D. P. Kuo, Y. C. Chen, S. J. Cheng, et al., “A Vision Transformer-Convolutional Neural Network Framework for Decision-Transparent Dual-Energy X-Ray Absorptiometry Recommendations Using Chest Low-Dose CT,” International Journal of Medical Informatics 199 (2025): 105901.
|
| [10] |
Y. Nomura, Q. Xu, H. Shirato, S. Shimizu, and L. Xing, “Projection-Domain Scatter Correction for Cone Beam Computed Tomography Using a Residual Convolutional Neural Network,” Medical Physics 46, no. 7 (2019): 3142–3155.
|
| [11] |
Y. Wang, C. Liu, X. Zhang, and W. Deng, “Synthetic CT Generation Based on T2 Weighted MRI of Nasopharyngeal Carcinoma (NPC) Using a Deep Convolutional Neural Network (DCNN),” Frontiers in Oncology 9 (2019): 1333.
|
| [12] |
F. Gaertner, Z. Ahmad, G. Rosenberger, et al., “Migrating Platelets Are Mechano-Scavengers That Collect and Bundle Bacteria,” Cell 171, no. 6 (2017): 1368–1382.e23.
|
| [13] |
C. A. Labarrere, A. E. Dabiri, and G. S. Kassab, “Thrombogenic and Inflammatory Reactions to Biomaterials in Medical Devices,” Frontiers in Bioengineering and Biotechnology 8 (2020): 123.
|
| [14] |
S. K. Tripathy, T. Goyal, and R. K. Sen, “Nonunions and Malunions of the Pelvis,” European Journal of Trauma and Emergency Surgery 41, no. 4 (2015): 335–342.
|
| [15] |
E. T. Hurley, Z. S. Aman, T. R. Doyle, et al., “Posterior Shoulder Instability, Part I-Diagnosis, Nonoperative Management, and Labral Repair for Posterior Shoulder Instability-An International Expert Delphi Consensus Statement,” Arthroscopy 41, no. 2 (2025): 166–180.e11.
|
| [16] |
S. Keshavarzi, G. Harris, S. Ramchandran, et al., “Role of Spinal MRI for Pre-Operative Work Up in Patients With Adolescent Idiopathic Scoliosis: A Retrospective Case Study and Narrative Review,” Spine Deformity 13, no. 2 (2025): 509–518.
|
Rights & permissions
2026 The Author(s). Orthopaedic Surgery published by Tianjin Hospital and John Wiley & Sons Australia, Ltd.
PDF
(466KB)
