The deep integration of artificial intelligence and the medical field makes the integration of cross-domain medical knowledge the key to improving the level of intelligent diagnosis and treatment. However, the semantic differences and expression barriers between different medical systems limit the effective integration and collaborative application of knowledge. This paper first constructs a three-layer framework covering resource aggregation, knowledge integration and intelligent services to support the whole process of medical decision-making from data fusion to knowledge services. Aiming at the core problem of cross-domain knowledge semantic alignment, the paper proposes a deep alignment method that integrates general and domain pre-training models, attention mechanisms and contrastive learning, and realizes accurate modeling and efficient alignment of complex semantic associations between heterogeneous medical texts through multi-level semantic extraction, dynamic weight focusing and contrastive semantic space optimization, effectively improves the accuracy and robustness of cross-domain knowledge alignment, and provides a technical foundation and practical path for the development of an intelligent decision support system that can deeply integrate multiple medical knowledge.