每日医学AI论文简报 v0.3
[日报]日期: 2026-08-07 来源: PubMed + arXiv | 分析: DeepSeek-v4-flash(PMC HTML解析,含表格+图注) 过滤: 近3天滚动窗口 | 期刊门槛(IF≥3/未收录放行) | 双层相关度(硬规则+LLM显式标准) | PMC全文优先(不足摘要级补齐) | 复合排序
💡 今日学习推荐 CS231n 卷积神经网络(Stanford) — 一、卷积神经网络(CNN)架构(进度0%) ⏭ 下一步:CS229 机器学习讲义(Andrew Ng)
1. Reconstructing delivered dose in real time: a beam physics-embedded, language-model-driven approach.
Phys Med Biol (IF: N/A) | 2026 Aug 5 | PubMed ⚠️ | DOI | ⚠️ 中置信 分析来源: 摘要
领域: 医学物理 / 自适应放疗 / 实时三维剂量重建
方法: 两阶段物理嵌入的剂量重建框架。阶段1:利用大语言模型(LLM)从机构射束数据库中提取射束物理特征(射野剖面、百分深度剂量),编码为射束先验;阶段2:在射束视角(BEV)空间中使用多模态神经网络(融合BEV对齐CT、最优通量图、LLM射束先验)进行通量-剂量重建。训练采用复合物理感知损失,评估采用2D/3D伽马分析。
核心发现: 基于180例肺癌患者内部验证和5例前列腺外部测试:3D伽马通过率(3%, 2 mm/10%)均值0.975±0.056,优于3D U-Net(0.931)、Dose-Net(0.939)、CLIP-UNet(0.945),p<0.05;2D伽马通过率0.962±0.073;外部测试0.953±0.088;重建9野计划约6.3秒;误差主要在低剂量区,相对最大剂量75 Gy误差≤±3 Gy。
相关度(医学AI研究者): 高。展示了LLM编码物理知识融入多模态深度学习用于实时剂量重建,性能优于纯学习模型,对自适应放疗和在线剂量监测有直接价值,为物理先验与医学AI结合提供范例。
一句话: 通过LLM提取射束物理特征并嵌入多模态网络,实现快速、高精度的实时三维剂量重建,优于现有深度学习方法。
2. Large Language Model-Driven Throat-Wearable Sensing System for Real-Time Recognition and Evaluation of Swallowing Disorders.
Adv Healthc Mater (IF: N/A) | 2026 Aug 5 | PubMed ⚠️ | DOI | ⚠️ 中置信 分析来源: 摘要
领域: 医学人工智能与可穿戴传感;脑卒中后吞咽障碍的实时监测与定量评估
方法: 提出一种大语言模型(LLM)驱动的喉部可穿戴传感系统(TWSS),包括柔性传感贴片和基于结构化信号序列的LLM框架(S3-LLM)。柔性贴片集成对压力和应变双敏感的拉伸传感器及无线电路模块,贴合喉部采集生理信号;S3-LLM将信号转化为结构化序列,结合时间编码和参数高效微调,在少样本条件下利用LLM的表征与泛化能力进行识别与评估。
核心发现: 在20名参与者的临床验证中,TWSS对正常喉部活动识别的准确率为92.4%,对吞咽功能评估的准确率为87.6%,相较传统模型提升约20%。表明该系统可实现连续、客观、定量的吞咽障碍评估。
相关度(医学AI研究者): 较高。展示了LLM用于生理信号序列建模的新路径,结合可穿戴传感与少样本学习,可推动吞咽障碍等疾病的无创、连续、客观监测,并为个性化医疗提供可行方案。
一句话: 该研究将大语言模型与喉部可穿戴传感系统结合,在少样本条件下实现了高精度的吞咽活动识别与吞咽功能定量评估,为吞咽障碍的连续客观监测提供了新方法。
3. Large language models can supplement the assessment of clinical high risk for psychosis.
J Psychopathol Clin Sci (IF: N/A) | 2026 Aug 6 | PubMed ⚠️ | DOI | ⚠️ 中置信
分析来源: PMC全文
领域: 精神病学临床心理学,精神病风险预测,自然语言处理在心理健康评估中的应用
方法: 使用2077份精神病风险综合征结构化访谈(SIPS)的评估者笔记作为输入,训练基于编码器的模型(ModernBERT)并结合零样本/少样本的生成式大语言模型(LLaMA-4、GPT-4.1 mini)生成精神病风险指标;以临床高危综合征(CHR)金标准评级为标签,比较模型分类性能,并分析AI风险指标与传统风险评分(SHARP、NAPLS)的关系及对转化预测的增量效益。
核心发现: 所有模型对CHR存在与否的笔记分类均优于随机水平;训练后的编码器模型表现最佳(准确率82.67%,κ=0.63)。该模型阳性分类可使SHARP和NAPLS风险评分分别增加0.69和0.85个标准差(p<.001);生成式模型风险评分每增加1个标准差,传统风险评分增加0.24至0.47个标准差(均p<.001)。探索性分析中,生成式模型风险评分在纳入SHARP(而非NAPLS)的转化预测模型中具有增量价值。结果表明,即使使用次优但易获取的评估者笔记,AI-NLP也能辅助CHR诊断、风险评估及未来转化预测。
相关度(医学AI研究者): 高——展示了大语言模型和编码器模型在真实临床笔记上用于精神病风险筛查与预测的可行性,为医学AI辅助精神科风险评估提供了直接证据和实用路径。
一句话: 利用评估者笔记训练的AI模型可有效补充临床高危精神病风险的评估与转化预测。
4. Comparative evaluation of large language models and clinicians in real-world glaucoma clinical reasoning.
Graefes Arch Clin Exp Ophthalmol (IF: N/A) | 2026 Aug 6 | PubMed ⚠️ | DOI | ⚠️ 中置信 分析来源: 摘要
领域: 眼科人工智能(青光眼临床推理)
方法: 由青光眼专家使用预定义评分标准,对大型语言模型(LLMs)与临床医生在34例真实青光眼病例中的临床推理表现进行比较评估。评分涵盖三个领域:医学准确性(40%)、关键点回忆(30%)、逻辑完整性(30%),计算加权综合得分作为描述性汇总。
核心发现: AI模型的加权平均分与主治医师重叠,并高于部分低年资受训者;但结果仅视为探索性表现模式,而非等价性证据。人类临床医生(尤其住院医师)个体间差异显著。AI系统常包含安全关键的诊断和管理要素,而表现最佳的人类临床医生获得总体最高个人分数。
相关度(医学AI研究者): 中等偏高。提示LLMs在专科临床推理中有潜力作为监督下的决策支持与教育工具,但在有限样本中未建立临床等效性,需专科监督与进一步验证。
一句话: 在34例青光眼病例中,LLMs的临床推理评分与主治医师重叠但未达等价,提示其可作为受监督的辅助工具,而非独立临床决策系统。
5. AI as Core Infrastructure in Radiology: Moving Beyond Pilots to Operational Excellence.
J Imaging Inform Med (IF: N/A) | 2026 Aug 5 | PubMed ⚠️ | DOI | ⚠️ 中置信 分析来源: 摘要
领域: 放射学人工智能(AI)的运营管理
方法: 评论文章,基于现有实践问题提出六部分管理框架和90天实施路线图。
核心发现: 放射科AI部署多停留在孤立试点阶段,导致工作流碎片化、集成重复、治理不一致、责任不清及价值评估困难。作者主张将AI视为共享治理、集成、测量和持续监督支持下的可管理运营能力,并提出临床意图、工作流编排、集成架构、治理与安全、测量与分析、持续改进六部分框架,以及从试点迈向常态化部署的90天路线图。
相关度(医学AI研究者): 高。关注AI从模型性能转向实际临床运营落地,为研究者提供了部署和管理AI系统时的非技术性关键考量。
一句话: 放射科AI应从孤立产品转向受治理的运营能力,通过六部分框架和90天路线图实现规模化卓越运营。
🔬 新增AI临床试验
近1日新增: 64 项 | 数据源: ClinicalTrials.gov
NCT07512271 Generative AI Patient Education Module for Breast Oncology
状态: RECRUITING | 期: NA | 赞助: | 招募: 35 条件: Breast Cancer; Metastatic Breast Cancer; Carcinoma of the Breast
NCT07219966 A Study of Brenipatide in Participants With Moderate-to-Severe Alcohol Use Disorder
状态: RECRUITING | 期: PHASE3 | 赞助: | 招募: 1100 条件: Alcohol Use Disorder
NCT07749833 Artificial Intelligence-Guided Rehabilitation for Low Back Dysfunction
状态: NOT_YET_RECRUITING | 期: NA | 赞助: | 招募: 68 条件: Low Back Pain; Low Back Dysfunction
NCT07279571 GYN-AIDE: Gynecologic Oncology Navigation Through AI-Driven Education
状态: RECRUITING | 期: N/A | 赞助: | 招募: 48 条件: Gynecologic Disease
NCT07750223 Cognitive Effects of Hesperidin in Overweight/Obese Older Adults
状态: RECRUITING | 期: NA | 赞助: | 招募: 105 条件: Cognitive Dysfunction; Overweight and Obesity
📄 医学AI预印本速览(arXiv 近3天)
1. Teaching Nemotron Greek: Mining a Corpus, Adapting Retrieval, and Grounding Generation for Modern Greek across Specialist Domains Ayoub Kirouane, Christos Petrocheilos | 2026-08-05 | arXiv Modern Greek is absent from NVIDIA’s Nemotron retrieval models and from major multilingual retrieval benchmarks, despite being important for retrieval-augmented generation (RAG) in legal, energy, financial, and medical applications. We present an end-to-end adaptation of the Nemo…
2. Predicting Brain Morphometry with MT-GNN: Mesh Evolution in Continuous Time with Graph-Based Metric Tensor Embeddings Hao Ding, Daniel Semchin, Paul M. Thompson et al. | 2026-08-05 | arXiv Predicting how a subcortical structure’s shape will evolve from a few prior scans could support prognosis and clinical-trial enrichment. Existing longitudinal mesh predictors either extrapolate shape trajectories via high-dimensional embeddings or regress vertex deformations dire…
3. OPD-V: Visual On-Policy Self-Distillation with Modality Balance Aniri, Jinhe Bi, Peng Liao et al. | 2026-08-05 | arXiv On-Policy Self-Distillation (OPSD) has become a standard post-training approach for improving visual reasoning in multimodal large language models (MLLMs). Existing methods draw privileged information from diverse input sources to guide self-distillation. Yet these designs overlo…
4. Spoken Function Calling: A New Perspective on Spoken Language Understanding for Large Audio Language Models Yuezhang Peng, Yuxin Liu, Changfeng Gao et al. | 2026-08-05 | arXiv Spoken Language Understanding (SLU) is the core component of task-oriented dialogue systems and a pivotal link in achieving seamless human-agent interaction. While traditional SLU can effectively extract user semantics for closed-set tasks after in-domain supervised fine-tuning, …
5. Chained Recursive Language Models for Multi-Iteration Reasoning Purbesh Mitra, Sennur Ulukus | 2026-08-05 | arXiv Long context reasoning in large language models (LLMs) is usually constrained by the fact that a single inference trajectory has to simultaneously explore the context, store intermediate state, verify evidence, and produce the final answer. This becomes particularly difficult in …
6. DASyR-LLM: Domain-Aware Symbolic Regression with LLMs for Kinetic Model Discovery Roberto Aliaga Medina, Paulina Quintanilla, Antonio del Rio Chanona | 2026-08-05 | arXiv Kinetic model discovery is a central challenge in chemical engineering, as accurate rate expressions are essential for understanding and controlling chemical and biological processes. Symbolic regression (SR) has emerged as a powerful data-driven approach for identifying interpre…
7. CoPlan: A Trustworthy Co-Intelligence Interface for Care Planning through Role-Based Contestable Argument Graphs Hung Truong Thanh Nguyen, Hélène Fournier, Piper Jackson et al. | 2026-08-05 | arXiv AI-supported care planning can help clinicians, patients, caregivers, and care teams coordinate complex decisions across clinical, functional, psychosocial, and environmental needs. However, many AI systems present recommendations as fixed outputs, limiting stakeholders’ ability …
8. Optimizing What Policies Learn From: Recoverability-aware Rollout Intervention Learning Zheyuan Zhang, Manqing Mao, Hong Wang et al. | 2026-08-05 | arXiv Critic-free group-based reinforcement learning has become a scalable approach for post-training large language models. However, most existing methods allocate the same number of rollouts to every task and trajectory state, even though some rollouts provide much more useful learni…