临床医学模型数据前处理流程

这是一篇关于临床医学模型数据前处理流程的思维导图，主要内容包括：影像组学资料Imaging histologydata，体格测量指标Physical measurement indicator，生化检测指标Biochemical test index，人口学指标Demographic indicator。

临床医学模型数据前处理流程

• 人口学指标Demographic indicator
• 生化检测指标Biochemical test index
• 体格测量指标Physical measurement indicator
• 数据前处理
  • 样本量计算Sample size calculation
    • 样本量与研究因素的比值大于20:1，如果是队列数据，还应在此基础上考虑失访率 The ratio of sample size to study factors is greater than 20:1, and in the case of cohort data,the lost visit rate should also be considered on this basis
  • 处理缺失值Handle missing values
    • 样本量充足Adequate sample
      • 删除个案数据 Delete data
      • 删除变量(超过50%缺失)Delete variable(more than 50% missing)
    • 样本量较少Small sample size
      • 数据插补法(多重插补法、K-最近邻法、线性回归法等) Data interpolation method (multiple interpolation, K-nearestneighbor method, linear regression, etc.)
  • 处理异常值 Handling outliers
    • 正态分布3σ原则和画箱型图Normal distribution and the 3σ principle of drawing box plots
      • 识别异常值Identify outliers
        • 删除或平均值修正Delete or use average correction
  • 数据转换 Data conversion
    • 例如:通过研究对象的出生日期计算出年龄 Calculating the age of the study participants by their date of birth 通过血压测量结果将研究对象分为高血压和正常两组Dividing the study subjects into two groups: hypertensive and normal,using the results of blood pressure measurements 通过身高和体重计算研究对象的BMI Calculate the BMl of the study subjects by height and weight.....
• 影像组学资料Imaging histology data
  • 通过分割软件分割感兴趣区域(ROI)
    • 提取高通量的特征来定量分析ROI的实质属性Extract high-throughput features to quantify the substantive attributes of ROI
      • 一阶统计特征(最大值、最小值、均值、范围等),反映肿瘤内灰度强度的分布，肿瘤内的异质性。First-order statistical features(maximum, minimum, mean, range, etc.), reflecting the distribution of gray intensity within the tumor, and the heterogeneity within the tumor.
      • 形状大小特征(肿瘤的长径、体积和表面积等)，反映肿瘤的形态、大小和规则度等信息。Shape and size characteristics (length-diameter, volume and surface area of the tumor, etc.), reflecting information on the morphology, size and regularity ofthe tumor
        • 特征选择 Feature Selection
          • 过滤法Filtering method
          • 包装法Packing method
          • 嵌入法Embedding method
      • 纹理特征主要通过灰度相关矩阵、邻域灰度差分矩阵等获得，用于量化肿瘤内部的纹理模式或组织分布等难以被视觉简单感知的信息。Texture features are mainly obtained by gray correlation matrix,neighborhood gray difference matrix, etc., and are used to quantifyinformation that is difficult to be perceived simply by vision, suchas texture patterns or tissue distribution inside the tumor.
      • 小波特征能够得到多频域多尺度的影像信息，对于难以用简单的肿瘤影像视觉特征进行描述的临床问题Wavelet features can obtain multi-frequency domain and multi-scale image information for clinical problems that are difficult to describe with simple visual features of tumor images