Prediction of Postoperative Diabetes Insipidus in Craniopharyngioma Based on an Improved CMT Model

It has been a major challenge in medicine to accurately predict the occurrence of postoperative diabetes insipidus after craniopharyngioma. In recent years, many researchers have begun to focus on studying multimodal medical data. However, traditional computer vision models often have problems in dealing with multimodal data. Examples include insufficient extraction of local features and insufficient cross-modal information fusion. These issues make it difficult to accurately capture deep correlations in multimodal data and limit their development in complex medical tasks. To solve these issues, a CAK-CMT model based on an improved CMT framework is proposed in this paper. The model combines the advantages of convolutional neural networks and Vision Transformers, while incorporating the Kolmogorov-Arnold Network (KAN) module to enhance feature representation capabilities. In addition, a Cross-modal Attention Mechanism (CA) is introduced to achieve efficient fusion of imaging and clinical data. The experimental results show that the CAK-CMT model outperforms the traditional computer vision models in all performance metrics. It achieves an accuracy of 82.5%. It shows remarkable performance in multimodal fusion and prediction tasks under complex scenarios, and is able to accurately predict the occurrence of postoperative diabetes insipidus after craniopharyngioma. This provides new insights for medical imaging analysis.