Multimodal Data and Deep Learning-Driven Diagnostic and Therapeutic Assistance Framework for Patellar Dislocation

This study aims to combine multimodal data and artificial intelligence algorithms to construct a model for identifying anatomical risk factors, predicting recurrence risk, and assessing postoperative outcomes. The proposed framework integrates preoperative, intraoperative, and postoperative information through a three-stage modeling framework. In the recurrence risk prediction module, we specifically incorporated lipid metabolism profiles and clinical variables, achieving an area under the curve (AUC) of 0.83, an F1-score of 0.68, and sensitivity of 0.82. Notably, younger patient age and elevated low-density lipoprotein (LDL) levels were found to be significantly associated with a higher likelihood of recurrence. Future work will incorporate real-world imaging data to enhance the performance of the imaging analysis component, ultimately aiming to construct a comprehensive multimodal fusion model. This model is intended to deliver more accurate and individualized clinical decision support for the diagnosis, prognosis, and treatment of patellar dislocation.Clinical Relevance - By integrating deep learning with multimodal data, this study advances the development of an intelligent risk stratification and treatment guidance tool for patellar dislocation. The proposed system holds potential for improving diagnostic precision, minimizing recurrence, and optimizing patient-specific treatment strategies.