TumorBoard: Multi-Agent AI System for Neuro-Oncology Decision Support
A new decision-support tool called TumorBoard has been created to improve neuro-oncology practices. It combines different types of data, like MRI scans, pathology reports, molecular markers, treatment histories, and updated guidelines. You can find more details in a paper published on arXiv (2608.03190). TumorBoard functions with a shared case history and a verifiable ledger of claims and evidence. Experts from various fields generate specific claims, while a critic identifies any discrepancies. Additionally, a safety governor decides if recommendations should be made based on the quality and timeliness of the evidence. In tests involving 360 cases, TumorBoard scored an impressive 0.772 in action F1 and 0.914 in evidence entailment, surpassing the previous baseline by 3.1 percentage points.
Key facts
- TumorBoard is a multi-agent decision-support system for neuro-oncology.
- It integrates serial MRI, pathology, molecular markers, treatment history, performance status, and guidelines.
- Specialist agents cover radiology, neuropathology, molecular diagnosis, guidelines, and therapy planning.
- An adversarial critic and safety governor ensure evidence sufficiency and temporal validity.
- On a 360-case benchmark, TumorBoard achieved action F1 of 0.772 and evidence entailment of 0.914.
- It outperformed the strongest typed-council baseline by 3.1 percentage points (95% CI: 1.6 to 4.7, adjusted p = 0.0012).
- Recommendation-to-evidence coverage reached 0.927.
- The system is described in arXiv paper 2608.03190.
Entities
Institutions
- arXiv