Research on Varicose Vein Bleeding Risk Prediction Based on Multimodal Machine Learning

Portal hypertension is a common complication of liver cirrhosis, leading to severe conditions such as gastric varices (GV) and esophageal varices (EV) bleeding, which have high mortality rates if not managed promptly. The use of artificial intelligence (AI) and machine learning (ML) in medical fields has shown immense potential, particularly for disease prediction and management. This study leverages multimodal machine learning (MMML) techniques, combining deep learning (DL) algorithms and clinical data, to develop a model for predicting GV and EV bleeding risks. This research, using endoscopic images and clinical data from Shandong Provincial Hospital, addresses the limitations of single-image-source studies. It uniquely integrates both preoperative and postoperative endoscopic images for model training, enhancing feature extraction through multimodal machine learning methods. Results show that the MMML model, which combines computer vision and clinical data, achieves high accuracy and robustness in predicting GV and EV bleeding risks within 12 months. External validation confirms the model's strong generalizability in practical applications, making it a valuable tool for clinical risk prediction.