The Large Cancer Assistant (LCA): A Model-Agnostic Orchestration Framework for Scalable Clinical Decision Support in Oncology

Imagine you're trying to build a system that helps doctors make better decisions by using different types of patient data — like images, lab results, and clinical notes. The challenge is that current systems often tie everything together too tightly, making it hard to adapt when new data types or AI models come into play. This rigidity can lead to problems like data bottlenecks or failures in decision-making when the underlying AI models change or when hospital IT systems are unreliable. This is what's called monolithic design failure.

The Large Cancer Assistant (LCA) addresses these issues by creating a flexible framework that separates how data is ingested from how AI models process that data. It uses a structured approach to standardize different types of patient information, allowing for smooth transitions between various AI models without losing the integrity of the data flow. The system includes a Cancer Switching Module that manages this orchestration, ensuring that even if the AI model changes, the routing of data remains consistent and reliable.

What sets the LCA apart from previous work is its focus on maintaining a clear boundary between data handling and AI execution, which enhances adaptability and modularity. In practice, this means that healthcare providers can integrate new AI tools without overhauling their entire system, leading to better decision support and improved patient outcomes.

The framework introduces a novel orchestration approach for multimodal data in oncology, extending existing methods significantly.

The claims are well-supported by a proof of concept and empirical validation across multiple scenarios.