横纹肌溶解症数据集

该知识库以横纹肌溶解症患者的结构化病历数据为基础，集成出入院信息、电子病历、历史诊断、用药、每日化验数据及影像检查等模块，采用规则提取、NLP分类与时间序列建模等多类算法构建。
通过结构化病历表中每日化验记录提取患者关键生化指标，构建时间序列特征矩阵。数据预处理包括缺失值插补（线性或多重插补）、异常值检测（3σ规则）与归一化（Z-score或MinMax）。对时间序列采用滑动窗口特征提取（窗口大小=1天，步长=1天），并使用LSTM模型进行趋势预测和转归判别。模型训练使用Adam优化器，loss函数为binary cross-entropy。
利用基于BERT-BiLSTM-CRF架构的实体识别模型，从电子病历文本中提取横纹肌溶解症相关诊断实体（如肌肉损伤、诱因、并发症等）。训练数据通过手工标注与规则辅助生成，BIO格式。诊断名通过ICD-10词典匹配进行标准化。实体向知识图谱映射时，通过余弦相似度结合医学词向量模型（PubMed预训练词向量）完成模糊匹配。
整个知识库构建过程中，所有模型均支持参数同步与动态更新机制，标注数据与模型输出结果可联动反馈，形成闭环更新体系。

This knowledge base is developed on the basis of structured medical record data of patients with rhabdomyolysis, integrating modules including admission and discharge information, electronic medical records, historical diagnoses, medication records, daily laboratory test data and imaging examinations. It is constructed using multiple algorithms such as rule-based extraction, NLP classification and time series modeling.

Key biochemical indicators of patients are extracted from daily laboratory records in the structured medical record tables to build a time-series feature matrix. Data preprocessing includes missing value imputation (linear or multiple imputation), outlier detection (3σ criterion) and normalization (Z-score or Min-Max normalization). Sliding window feature extraction is applied to the time series with a window size of 1 day and a stride of 1 day, and an LSTM model is used for trend prediction and prognosis judgment. The model is trained with the Adam optimizer, and the loss function is binary cross-entropy.

A named entity recognition (NER) model based on the BERT-BiLSTM-CRF architecture is employed to extract rhabdomyolysis-related diagnostic entities (such as muscle injury, precipitating factors, complications, etc.) from electronic medical record texts. The training data is generated through manual annotation and rule assistance, in the BIO tagging format. Diagnostic names are standardized via matching with the ICD-10 dictionary. When mapping entities to the knowledge graph, fuzzy matching is completed by combining cosine similarity with the medical word embedding model (PubMed pre-trained word embeddings).

During the entire construction process of this knowledge base, all models support parameter synchronization and dynamic update mechanisms. Annotated data and model output results can be linked and fed back to form a closed-loop update system.

搜集汇总

背景与挑战

背景概述

该数据集包含60万条横纹肌溶解症患者的结构化医疗记录，涵盖诊断、用药和化验等多维度信息，适用于医疗研究和流行病学分析。数据通过NLP和时间序列算法处理，支持诊疗模式调研和疗效评估，每年更新一次。