RadOnc-Agent
An intelligent agent framework for radiation therapy that uses a large language model as its orchestration core. The framework formalizes the full radiotherapy workflow into four clinical phases and exposes 26 callable functions. It connects dispersed radiotherapy AI capabilities through intent mapping, context maintenance, and request routing.
Timeline (last 90 days)
RadOnc-Agent is an intelligent agent framework that uses a large language model as its scheduling core, with the goal of connecting disparate radiotherapy AI capabilities into a collaborative pipeline.
RadOnc-Agent formalizes the radiotherapy process into four clinical stages and provides 26 callable functions through a conversational interface.
RadOnc-Agent's LLM controller is responsible for three tasks: intent mapping, context maintenance, and request routing.
RadOnc-Agent's LLM is positioned as an orchestrator rather than an executor, with specialized computations delegated to domain-specific modules.
The research team tested the system using 2,600 single-function requests (7,800 total executions), and RadOnc-Agent selected the expected function in 98.79% of single-function executions.
RadOnc-Agent completed 96.50% of scripted cross-stage workflows across 200 predefined cross-stage synthetic scenarios (600 total executions).
RadOnc-Agent completed 96.67% of real patient case workflow executions across 120 workflow instances (360 executions) constructed from 60 de-identified patient records.
After removing longitudinal state, RadOnc-Agent's cross-stage workflow completion rate dropped from 96.50% to 84.00%.
After disabling schema constraints and authentication, the erroneous backend dispatch rate spiked from 0% to 95.28% in replay/test evaluations.
The researchers note that these findings establish the technical feasibility of LLM orchestration architecture in coordinating heterogeneous radiotherapy capabilities, but do not demonstrate clinical correctness, clinical utility, or prospective benefit.
2 more timeline events
All Facts (12)
RadOnc-Agent is an intelligent agent framework that uses a large language model as its scheduling core, with the goal of connecting disparate radiotherapy AI capabilities into a collaborative pipeline.
50%UnverifiedRadOnc-Agent formalizes the radiotherapy process into four clinical stages and provides 26 callable functions through a conversational interface.
50%UnverifiedRadOnc-Agent's LLM controller is responsible for three tasks: intent mapping, context maintenance, and request routing.
50%UnverifiedRadOnc-Agent's LLM is positioned as an orchestrator rather than an executor, with specialized computations delegated to domain-specific modules.
50%UnverifiedThe research team tested the system using 2,600 single-function requests (7,800 total executions), and RadOnc-Agent selected the expected function in 98.79% of single-function executions.
50%UnverifiedRadOnc-Agent completed 96.50% of scripted cross-stage workflows across 200 predefined cross-stage synthetic scenarios (600 total executions).
50%UnverifiedRadOnc-Agent completed 96.67% of real patient case workflow executions across 120 workflow instances (360 executions) constructed from 60 de-identified patient records.
50%UnverifiedAfter removing longitudinal state, RadOnc-Agent's cross-stage workflow completion rate dropped from 96.50% to 84.00%.
50%UnverifiedAfter disabling schema constraints and authentication, the erroneous backend dispatch rate spiked from 0% to 95.28% in replay/test evaluations.
50%UnverifiedThe researchers note that these findings establish the technical feasibility of LLM orchestration architecture in coordinating heterogeneous radiotherapy capabilities, but do not demonstrate clinical correctness, clinical utility, or prospective benefit.
50%UnverifiedRadOnc-Agent's real patient case workflows cover two typical processes: decision-to-planning and planning-to-adaptive.
50%UnverifiedRadOnc-Agent was proposed in a recent arXiv paper.
50%