Traj-Evolve: A Self-Evolving Multi-Agent System for Patient Trajectory Modeling in Lung Cancer Early Detection

2026-06-01 · Source: Artificial Intelligence · Field: Science & Research — Artificial Intelligence & Machine Learning, Health & Medical Research, Data Science & Analytics · Depth: Expert, quick

Summary

Traj-Evolve is a self-evolving multi-agent system designed for patient trajectory modeling in lung cancer early detection, addressing challenges like sparse, noisy, and long-context multimodal electronic health records (EHRs). It integrates two evolving mechanisms: an Experience Pool (ExPool) for non-parametric memory, retrieving similar patients as few-shot contexts, and multi-agent reinforcement learning (MARL) for parametric optimization of inter-agent and agent-memory collaboration. A leave-one-out cross-retrieval strategy unifies these mechanisms. On a lung cancer prediction task using up to five years of multimodal EHRs, Traj-Evolve outperformed 9 strong baselines across both overall and never-smoker populations. Analysis revealed ExPool expansion shifts optimal retrieval, MARL's manager agent prediction loss converges quickly, and the two mechanisms complement each other, with ExPool improving specificity and MARL improving sensitivity.

Key takeaway

For AI Scientists developing predictive models for complex medical data, Traj-Evolve demonstrates a powerful hybrid approach to overcome limitations of traditional LLM-based systems. You should explore integrating both non-parametric memory, like the Experience Pool, and parametric learning, such as multi-agent reinforcement learning, to enhance your model's robustness and clinical utility. This is particularly relevant for challenging patient cohorts and when balancing prediction specificity and sensitivity.

Key insights

Traj-Evolve combines non-parametric memory and parametric reinforcement learning for robust patient trajectory modeling in complex EHR data.

Principles

• Hybrid memory systems enhance patient trajectory modeling.
• Retrieval augmentation aligns training and inference behaviors.
• Complementary mechanisms improve both specificity and sensitivity.

Method

Traj-Evolve uses an Experience Pool for few-shot context retrieval and MARL via reward-ranked fine-tuning to optimize inter-agent and agent-memory collaboration, unified by a leave-one-out cross-retrieval strategy.

In practice

• Implement non-parametric memory for few-shot context in EHR analysis.
• Apply MARL to optimize agent collaboration and memory use.
• Combine memory types for balanced specificity and sensitivity.

Topics

• Patient Trajectory Modeling
• Multi-Agent Systems
• Lung Cancer Detection
• Electronic Health Records
• Reinforcement Learning
• Few-Shot Learning
• Retrieval Augmentation

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Reinforcement learning-based design of sequential drug treatment targeting the evolving tumour landscape with SequenTx — SequenTx uses reinforcement learning and a virtual cell model to design effective sequential drug treatments for evolving tumors.
• Agentifying Patient Dynamics within LLMs through Interacting with Clinical World Model — Integrating a Clinical World Model with LLMs via agentic training improves sepsis treatment recommendations and safety.
• Sybil: An AI model that can predict lung cancer risk 6 years in advance — Sybil AI predicts future lung cancer risk from CT scans by analyzing subtle, comprehensive image data.
• UniRG: Scaling medical imaging report generation with multimodal reinforcement learning — Reinforcement learning with clinically meaningful rewards significantly improves medical vision-language model reliability and generality.
• ClinEnv: An Interactive Multi-Stage Long Horizon EHR Environment for Agents — ClinEnv reveals LLMs struggle with sequential, uncertain clinical decisions and information gathering in a multi-stage environment.
• Enhancing Clinical Trial Patient Matching through Knowledge Augmentation and Reasoning with Multi-Agent — Multi-agent systems can augment LLMs with domain knowledge to improve clinical trial patient matching accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.