← 記事一覧に戻る
大規模言語モデルヘルスケアAIAI規制・政策

EndoGov multi-agent system achieves guideline-compliant endometrial cancer risk stratification with 0.943 accuracy on TCGA-UCEC dataset

arXiv cs.MA (Multi-Agent) · 2026年4月28日

AI要約

  • •EndoGov is a two-tier multi-agent expert system that factorizes decision-making as D(x) = G(P(x), R), where specialist agents P extract structured evidence and a governance agent G applies an executable rule set R. Tier 1 comprises pathology, molecular, and clinical agents; Tier 2 queries an evidence-level-weighted Guideline Knowledge Graph using deterministic hard-path rules for high-priority overrides and constrained soft-path reasoning for ambiguous cases.
  • •On TCGA-UCEC (n=541), EndoGov achieved 0.943 accuracy, 0.973 macro AUC, and a conditional logic-violation rate (C-LVR) of 0.93% among trigger-exposed cases. On CPTAC-UCEC (n=95), EndoGov reached 0.842 accuracy compared with < 0.31 for locked-transfer neural baselines.
  • •End-to-end safety decomposition localized residual failures primarily to upstream molecular detection rather than downstream governance. Backend-swap experiments showed that hard-path compliance is invariant to the LLM backend, indicating that explicit clinical-rule governance can provide guideline-compliant, auditable endometrial cancer risk assignment while preserving competitive discrimination.
元記事を読む

関連記事

大規模言語モデル

Nvidia releases Nemotron 3 Nano Omni, a 30-billion-parameter open multimodal model trained on data from competing AI labs including Qwen, OpenAI, and DeepSeek.

THE DECODER·2026年4月29日
大規模言語モデル

Google rolls out Gemini memory feature in Europe and adds tools to import data from other AI assistants

THE DECODER·2026年4月29日
OpenAI shifts cloud partnership from Microsoft to Amazon; China blocks Meta's Manus acquisition; J&J uses AI to halve drug development lead generation time
大規模言語モデル

OpenAI shifts cloud partnership from Microsoft to Amazon; China blocks Meta's Manus acquisition; J&J uses AI to halve drug development lead generation time

Yahoo Finance AI·2026年4月29日
Diplomat-agent static scanner reveals 76% of tool calls in 16 open-source AI agent repos lack security guards
大規模言語モデル

Diplomat-agent static scanner reveals 76% of tool calls in 16 open-source AI agent repos lack security guards

Hacker News·2026年4月29日
PromptPack RFC 0010 adds composition mode—a declarative step-graph orchestration for procedural LLM workflows alongside existing dialogue patterns
大規模言語モデル

PromptPack RFC 0010 adds composition mode—a declarative step-graph orchestration for procedural LLM workflows alongside existing dialogue patterns

Hacker News·2026年4月29日

AIニュースを毎日お届け

200以上のソースから厳選したAIニュースを毎日無料でお届けします。

無料で始める