AI
OpenTumorBoard: A Real-World Benchmark of Multidisciplinary Tumor Board Discussion Trajectories
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 25/09/2026, 20:00:00 UTC
Briefing created: 03/10/2026, 08:24:54 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/10/2026, 17:18:24 UTC.
Briefings
OpenTumorBoard introduces a benchmark of 611 patient cases and 19,157 discussion turns across ten specialist roles, transcribed from 12,534 minutes of YouTube recordings. It evaluates two settings: SPECIALIST TURN, where LLMs respond to clinical questions, and BOARD SIMULATION, where they generate full consensus discussions. Evaluation of 14 frontier LLMs shows best models score 3.43/5 in clinical equivalence and 2.78/5 in alignment with recorded conclusions. Supervised finetuning and reinforcement learning improve performance on a held-out test set, indicating real-world trajectories support model adaptation. Three M.D. experts reviewed the subset, noting high information coverage and strong consensus fidelity.
Key points
- OpenTumorBoard contains 611 patient cases and 19,157 discussion turns across ten specialist roles.
- The benchmark evaluates LLM performance in both specialist turn responses and full board simulation scenarios.
Context and limitations
The benchmark relies on publicly available YouTube recordings, which may introduce transcription errors or missing audio details. Evaluation metrics are based on a single snapshot of performance at the time of the study, and no causal claims are made about the effectiveness of supervised or reinforcement learning approaches.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 03/10/2026, 08:24:54 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeidnmqfm36nygymyk5uiotxurmcmf6tu5ils4pvbxnvksb6q4kty44