Large Cancer Assistant: Orchestration Framework for Oncology Decision Support

Researchers have developed the Large Cancer Assistant (LCA), a novel, model-agnostic orchestration framework aimed at providing scalable clinical decision support in oncology. This framework addresses the limitations of current monolithic deep learning models by rigidly separating data ingestion, clinical routing, and AI inference components. The LCA is mathematically formalized with a 7-tuple architecture, adhering to the principle of Algorithmic Impermeability, which ensures that the orchestration logic remains independent of the underlying black-box AI models. It employs Entry Theory, utilizing Geometric Deep Learning (GDL) to standardize diverse multimodal patient data across structural and medical dimensions. The system dynamically routes data via a Cancer Switching Module and isolates core AI execution from hospital IT infrastructures by generating a Standardized Intermediate Payload (SIP). A proof-of-concept validated its logic, demonstrating negligible overhead, invariant routing during AI model swaps, 100% recall for data anomaly requests, and multi-protocol execution, setting the stage for EMR interoperability.