Accurately predicting the risk of hemorrhagic complications is crucial for doctors to make treatment decisions. In order to quickly assist clinicians in evaluating patients’ conditions, we propose a new method to integrate imaging and clinical data to improve the accuracy of prediction. Unsupervised comparative learning is used for pre training to alleviate the limited number of data sets. Stacking model is used to combine multiple classifiers to reduce bias and improve prediction accuracy. The results show that the performance of the model is significantly improved by contrast learning pre training, and the prediction accuracy is improved by using multimodal data. The superposition model further improves the stability of prediction. Finally, the AUC and ACC of our prediction model for bleeding complications in the validation set were 0.9427 and 0.8824, respectively. Compared with previous work, the AUC and ACC of this model are increased by 3% and 4%, respectively. The model can be used as a diagnostic aid to help doctors better assess the risk of patients and improve the treatment effect.