文章背景与核心概要
临床决策支持系统常常随着医疗方案的演变而面临性能退化的问题。更新这些模型的一个主要障碍在于缺乏可审计性：当模型被完全重新训练时，几乎无法准确追踪究竟修改了哪些参数。本文介绍了一种双流架构，将生理特征表征（由LSTM处理）与治疗表征（由MLP处理）进行了分离。
通过冻结生理特征编码器并将更新限制在治疗流和融合头中，作者确保了模型既具备可审计性又保持稳定。这种方法在维持高性能的同时，为模型的演进提供了结构上的保障，有效地平衡了自适应性与治理合规性。
Clinical decision support systems often struggle as medical protocols evolve over time. A major barrier to updating these models is the lack of auditability: when a model is fully retrained, it becomes impossible to trace exactly which parameters were modified.
This paper introduces a two-stream architecture that separates physiological representations (handled by an LSTM) from treatment representations (handled by an MLP). By freezing the physiological encoder and confining updates to the treatment stream and fusion head, the authors ensure that the model remains auditable and stable. This approach maintains high performance while providing a structural guarantee over the model's evolution, effectively balancing adaptability with governance.
漂移自适应的ICU干预预测：冻结生理特征编码器以实现可审计的模型更新
作者： Fatema Ferdous Tamanna, K. M. Merajul Arefin, Md. Abdul Masud
arXiv ID： 2607.19020
学科领域： 机器学习 (cs.LG)
最后修订日期： 2026年8月26日
Drift-Adaptive ICU Intervention Prediction: Freezing the Physiological Encoder for Auditable Model Updating
Authors: Fatema Ferdous Tamanna, K. M. Merajul Arefin, Md. Abdul Masud
arXiv ID: 2607.19020
Subject: Machine Learning (cs.LG)
Date: Last revised August 26, 2026
摘要
临床决策支持系统常常随着医疗方案的演变而面临性能退化的问题。更新这些模型的一个主要障碍在于缺乏可审计性：当模型被完全重新训练时，几乎无法准确追踪究竟修改了哪些参数。
本文介绍了一种双流架构，将生理特征表征（由LSTM处理）与治疗表征（由MLP处理）进行了分离。通过冻结生理特征编码器并将更新限制在治疗流和融合头中，作者确保了模型既具备可审计性又保持稳定。这种方法在维持高性能的同时，为模型的演进提供了结构上的保障，有效地平衡了自适应性与治理合规性。
Summary
Clinical decision support systems often struggle as medical protocols evolve over time. A major barrier to updating these models is the lack of auditability: when a model is fully retrained, it becomes impossible to trace exactly which parameters were modified.
This paper introduces a two-stream architecture that separates physiological representations (handled by an LSTM) from treatment representations (handled by an MLP). By freezing the physiological encoder and confining updates to the treatment stream and fusion head, the authors ensure that the model remains auditable and stable. This approach maintains high performance while providing a structural guarantee over the model's evolution, effectively balancing adaptability with governance.
核心贡献
- 架构解耦： 将生理数据处理与治疗特定逻辑解耦，从而实现针对性的更新。
- 可审计更新： 通过保持生理特征编码器与源模型在位级别（bitwise）完全一致，系统确保了更新的透明度，并将其限制在特定的架构模块中。
- 基于证据的检索： 结合了针对每个样本的 PubMed 查询以及冻结的编码器，以提供临床背景。
- 性能稳定性： 对 84,792 例 MIMIC-IV ICU 住院记录的实证评估表明，这种受约束的方法在性能上与全面重新训练具有竞争力（平均 AUROC 为 0.9316 对比 0.9249），同时在不同的自适应随机种子下具有显著更高的稳定性。
Key Contributions
- Architectural Separation: Decouples physiological data processing from treatment-specific logic, allowing for targeted updates.
- Auditable Updates: By keeping the physiological encoder bitwise identical to the source model, the system ensures that updates are transparent and restricted to specific architectural blocks.
- Evidence-Based Retrieval: Integrates per-instance PubMed queries coupled with the frozen encoder to provide clinical context.
- Performance Stability: Empirical evaluation on 84,792 MIMIC-IV ICU stays demonstrates that this constrained approach performs competitively with full retraining (mean AUROC 0.9316 vs. 0.9249) while being significantly more stable across adaptation seeds.
元数据与访问
| 字段 | 详情 |
|---|---|
| 主要学科 | 机器学习 (cs.LG) |
| 其他学科 | 人工智能 (cs.AI)，信息检索 (cs.IR)，定量方法 (q-bio.QM) |
| DOI | https://doi.org/10.48550/arXiv.2607.19020 |
| 许可证 |  查看许可证 |
Metadata & Access
Field Details Primary Subject Machine Learning (cs.LG) Additional Subjects Artificial Intelligence (cs.AI), Information Retrieval (cs.IR), Quantitative Methods (q-bio.QM) DOI https://doi.org/10.48550/arXiv.2607.19020 License
全文链接
Full-Text Links
提交历史
- v1： 2026年7月21日
- v2： 2026年8月20日
- v3： 2026年8月26日（当前版本）
Submission History
- v1: 21 Jul 2026
- v2: 20 Aug 2026
- v3: 26 Aug 2026 (Current)