CoMMa: Contribution-Aware Medical Multi-Agents From A Game-Theoretic Perspective
🤖 AI Summary
This work proposes a decentralized large language model multi-agent framework to address the challenges of dynamic reasoning and collaboration over multi-source, heterogeneous data in oncology decision support. Specialized agents perform independent reasoning based on partitioned evidence and coordinate their decisions through a game-theoretic objective. The framework introduces a novel contribution-aware credit assignment mechanism grounded in marginal utility, integrated with deterministic embedding projection and explicit evidence attribution to construct mathematically rigorous and interpretable decision pathways. Evaluated on multiple oncology benchmarks—including real-world multidisciplinary tumor board data—the approach significantly outperforms both centralized and role-based multi-agent baselines in terms of accuracy and performance stability.
Application Category
📝 Abstract
Recent multi-agent frameworks have broadened the ability to tackle oncology decision support tasks that require reasoning over dynamic, heterogeneous patient data. We propose Contribution-Aware Medical Multi-Agents (CoMMa), a decentralized LLM-agent framework in which specialists operate on partitioned evidence and coordinate through a game-theoretic objective for robust decision-making. In contrast to most agent architectures relying on stochastic narrative-based reasoning, CoMMa utilizes deterministic embedding projections to approximate contribution-aware credit assignment. This yields explicit evidence attribution by estimating each agent's marginal utility, producing interpretable and mathematically grounded decision pathways with improved stability. Evaluated on diverse oncology benchmarks, including a real-world multidisciplinary tumor board dataset, CoMMa achieves higher accuracy and more stable performance than data-centralized and role-based multi-agents baselines.
Problem
Research questions and friction points this paper is trying to address.
multi-agent systems
oncology decision support
contribution-aware credit assignment
heterogeneous patient data
interpretable decision-making
Innovation
Methods, ideas, or system contributions that make the work stand out.
contribution-aware
game-theoretic
multi-agent LLM
evidence attribution
deterministic embedding
🔎 Similar Papers
No similar papers found.
