Abstract
Acute aortic syndrome (AAS) is a severe cardiovascular disorder with a high mortality rate in the early stage of onset. In this study, the aim was to develop a multi-modal multi-scale fusion (MMMF) model to enhance early identification and classification of the diagnostic efficacy of AAS. Build a new diagnostic model of human-machine collaboration to reduce the early misdiagnosis and missed diagnosis of AAS. A retrospective analysis was conducted on the CTA images and clinical indicators of 493 patients from 2019 to 2024. In this study, a multi-scale image encoder was used to extract morphological features, which were then deeply integrated with clinical indicators through a graph neural network (GNN). The MMMF model outperforms the single-modal model in all four core classification metrics, confirming that multimodal fusion can effectively achieve complementary gains. The overall comprehensive diagnostic performance of the model (AUC > 0.9) demonstrates excellent diagnostic and classification capabilities. The graph structure modeling approach and the resolution of the visual encoder are the two key influencing factors of the model. The MMMF model has significantly enhanced the diagnostic and classification performance of AAS. It can be used as an auxiliary clinical screening tool to help doctors quickly identify typical cases, mark low-confidence cases, and accurately and promptly diagnose AAS.