内镜下胃溃疡性病变鉴别诊断的数学模型

The mathematical model for the differential diagnosis of endoscopic gastric ulcerative lesions

目的：借助图像处理软件结合统计学分析方法尝试建立内镜下胃溃疡性病变鉴别诊断的数学模型。方法回顾性收集25例胃溃疡和25例溃疡型胃癌患者的普通光镜图像，利用 Photo-shop CS4软件进行图片处理，采集代表溃疡状态的10个观察指标（面积、周长、圆度、高度、宽度、灰度、累积密度、颜色、明度、像素），应用主成分分析、因子分析、聚类分析对图像数据结果进行分析。结果主成分分析将胃溃疡性病变的内镜图像信息分为两类：第1主成分与面积、周长、高度、宽度、累积密度、像素呈现较强的正相关，代表溃疡的形态特征；第2主成分与灰度、颜色、明度呈现较强的正相关，代表溃疡的色彩特征。因子分析结果显示，溃疡型胃癌较胃溃疡在图像和第1、第2主成分之间呈现更强的正相关关系。聚类结果表明，对于大多数患者来说，发病具有规律性，在建立胃溃疡的数学诊断模型时有效的诊断信息应包括面积≤10000、周长≤250、高度≤90、宽度≤60、灰度≥90、累积密度≤190000、颜色≥110、明度≥110（单位均为像素值），在建立溃疡型胃癌的数学诊断模型时有效的诊断信息应包括面积≥38000、周长≥690、高度≥300、宽度≥426、灰度≤100、累积密度≥14000000、颜色≤125、明度≤130（单位均为像素值）。结论内镜图像中面积、周长、圆度、高度、宽度、灰度、累积密度、颜色、明度、像素均为重要的数据信息，基于这10个观察指标基础上建立的数学诊断模型可以简单地对胃溃疡性病变进行早期预测和诊断。

Objective To establish mathematical model for the differential diagnosis of endoscopic gastric ulcerative lesions with the help of image processing software and statistical analysis method. Methods The microscopic images of 25 gastric ulcers and 25 gastric peptic cancers were collected from January 2013 to December 2013.Then the Photoshop CS4 software was used for image processing and representative measurements gathering（area,perimeter,roundness,height,width,gray scale,involving the density,color,lightness,pixels）.Then the principal component analysis,the scatter plot and cluster analysis were performed by SPSS 16.0.Results According to the results of principal component analysis, Z1 showed a strong positive correlation with area,perimeter,height,width,cumulative density,pixel,which represented the shape statistics of ulcer.And Z2 showed a strong positive correlation with gray scale,color, lightness,which represented the color statistics of ulcer.On the Scatter plot,gastric peptic cancer images showed a stronger positive correlation with the principal component Z1 and Z2 .Clustering analysis results showed that for the majority of patients,there was a pattern in its incidence.In the gastric ulcer diagnostic models,the effective information should include area≤10 000;perimeter≤250;height≤90;width≤60;gray scale≥90;cumulative density≤190 000;color≥110;and lightness≥110（unit：pixel）.In the gastric ulcera-tive cancers,the effective information should include area≥38 000,perimeter≥690,height≥300,width ≥ 426,gray scale ≤ 100,cumulative density ≥ 14 000 000,color ≤ 125,and lightness ≤ 130 （unit：pixel）. Conclusion Among the endoscopic image data,area,perimeter,roundness,height,width,gray scale, cumulative density,color,brightness,pixels are important indices,which can contribute to the differential diagnosis of gastric ulcerative lesions.The established mathematical model based on the ten indices can be applied to predict and diagnose gastric ulcerative lesions.