发表机构
King's College London, University of London; Oxford Immune Algorithmics（伦敦大学国王学院; 牛津免疫算法公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究结合疾病特异性生物标志物图、ODE、CNN、LSTM及GPT-4o-mini RAG构建临床决策流程，在103种疾病的鉴别诊断中取得较高准确率，发现证据基础与诊断正确性存在解耦。
AI 中文摘要
诊断错误带来了沉重负担，而无约束的大型语言模型（LLM）仍易产生幻觉，且对定量实验室动态的整合能力较弱。我们开发了一种决策支持流程，结合疾病特异性生物标志物相关图、常微分方程（ODE）、深度序列分类以及检索增强生成（RAG）。针对全血细胞计数（FBC）库中的103种疾病类别，生物标志物网络被用作耦合矩阵，为每种疾病生成30条轨迹（共3090条）。一维卷积神经网络（CNN）和长短期记忆（LSTM）网络对疾病轨迹及6种动态簇进行分类。受限的GPT-4o-mini RAG层使用包含19种模式的BMJ最佳实践/NICE语料库生成鉴别诊断，并对其诊断适用性、证据基础及临床合理性进行评估。在5次随机种子运行中，CNN的疾病水平准确率为0.940±0.006（95%置信区间0.933--0.948），LSTM的准确率为0.852±0.019（95%置信区间0.828--0.875）；CNN的优势为8.87个百分点（95%置信区间6.47--11.27；t(4)=10.26，p=5.1×10^-4；Hedges' g=3.67）。在100个抽样RAG案例中，96个成功解析；97.9%的案例引用了证据，71.9%的案例提及了真实诊断，综合评分为3.82/5，严格通过率为47.9%。核心发现是证据基础与诊断正确性之间存在解耦：分类器正确与分类器错误的输出在诊断适用性上存在差异，但在证据基础上无差异。事后分析证实，诊断评分存在1.02分的差异（曼-惠特尼检验p=0.0024；Hedges' g=0.72），而证据基础仅相差-0.02分（p=0.839；g=-0.04）。
英文摘要
Diagnostic error carries a burden, while unconstrained large language models (LLMs) remain vulnerable to hallucination and weak integration of quantitative laboratory dynamics. We developed a decision-support pipeline combining disease-specific biomarker correlation graphs, ordinary differential equations (ODEs), deep sequence classification, and retrieval-augmented generation (RAG). For 103 disease classes from a full blood count (FBC) repository, biomarker networks were used as coupling matrices to generate 30 trajectories per disease (3,090 total). A one-dimensional convolutional neural network (CNN) and long short-term memory (LSTM) network classified disease trajectories and six dynamical clusters. A constrained GPT-4o-mini RAG layer used a 19-pattern BMJ Best Practice/NICE corpus to generate differential diagnoses evaluated for diagnostic suitability, evidential grounding, and clinical plausibility. Across five random-seed runs, disease-level accuracy was $0.940 \pm 0.006$ for the CNN (95\% CI 0.933--0.948) and $0.852 \pm 0.019$ for the LSTM (95\% CI 0.828--0.875); the CNN advantage was 8.87 percentage points (95\% CI 6.47--11.27; $t(4)=10.26$, $p=5.1\times10^{-4}$; Hedges' $g=3.67$). Among 100 sampled RAG cases, 96 parsed successfully; evidence was cited in 97.9\%, the true diagnosis was mentioned in 71.9\%, and the composite score was 3.82/5 with a 47.9\% strict pass rate. The central finding was a decoupling between grounding and diagnostic correctness: classifier-correct versus classifier-wrong outputs differed in diagnostic suitability but not evidential grounding. Post-hoc analysis confirmed a 1.02-point diagnostic-score difference (Mann--Whitney $p=0.0024$; Hedges' $g=0.72$), whereas grounding differed by only $-0.02$ points ($p=0.839$; $g=-0.04$).
Comments14 pages
