Development of a radiomics-vision transformer fusion model based on chest CT for predicting adverse respiratory events during recovery in elderly hip fracture patients under general anesthesia.

Background: Hip fracture is a common and serious injury in the elderly. With the aging of the global population, the incidence of hip fracture is increasing. Adverse Respiratory Events (AREs) are common in elderly patients with hip fracture during recovery from general anesthesia, which can lead to serious complications. However, current methods for predicting these events are limited. Methods: This retrospective multicohort study analyzed clinical data from 664 patients across two institutions. Radiomic features were extracted from regions of interest (ROIs) in chest CT scans, and deep learning features were extracted using a vision transformer (ViT) model. A radiomics-ViT fusion model was developed by combining these features. The performance of the models was evaluated using metrics such as area under the curve (AUC), sensitivity, specificity, and F1-score. Results: The radiomics-ViT fusion model demonstrated excellent performance, with an AUC of 0.994 in the internal training set and 0.875 in the external test set. This was significantly better than the XGBoost model (AUC 0.553) and the ViT model alone (AUC 0.788) in the external test set. The fusion model accurately identified high-risk patients, enabling timely interventions and improved outcomes. Conclusion: The developed radiomics-ViT fusion model serves as a valuable tool for predicting AREs during recovery in elderly hip fracture patients under general anesthesia, enhancing clinical decision-making and patient care.