Multi modal fusion of medical imaging and biomechanical data using attention based swin-unet and LSTM for sports injury prediction.

Background: Accurately predicting sports injuries remains a significant challenge due to the complexity of factors involved, including anatomical structures and movement mechanics. Traditional approaches often rely on single data sources and fail to provide personalized risk assessments, limiting their effectiveness. Methodology: This study introduces a multimodal approach to predicting sports injuries by combining high resolution computed tomography (CT) scans with biomechanical data from motion capture systems, wearable inertial measurement units (IMUs), and force-sensitive insoles. CT images were denoised and contrast-enhanced before being analyzed with the Swin-UNet architecture, which captures both fine structural details and broader spatial patterns. At the same time, biomechanical signals such as joint movement, ground reaction forces, and loading patterns were processed using orthogonal component decomposition and analyzed with a Long Short-Term Memory (LSTM) network to capture changes over time. The results from both models were combined through a decision level fusion method, producing a single injury-risk score. By integrating anatomical and functional data, the framework provides a more accurate and timely assessment of injury risk, supporting early intervention and improved athlete safety. Results: The proposed model demonstrated strong predictive performance, achieving an accuracy of 94%, precision of 91%, recall of 92%, and an F1 score of 91%. These results highlight the advantage of combining high resolution imaging with biomechanical measurements through an advanced deep learning framework, outperforming traditional methods. Conclusion: By integrating CT imaging and biomechanical data within a Swin Unet based framework, this study offers a precise and personalized approach to sports injury prediction. The inclusion of real-time monitoring further enhances the practical value of the model, supporting early intervention and improving athlete safety and training efficiency.