消化粘膜下超声图像斑点的模型分析

Model Analysis of Ultrasound Image Speckles in the Digestive Submucosa

摘要: 由于超声斑点的形成依赖于底层组织结构，利用斑点建模对超声图像中的组织器官进行分割和分类已经成为一个研究热点。本文将探究最能表征消化粘膜下正常组织和肿瘤组织中斑点的概率分布模型。本文对消化粘膜下正常组织和肿瘤组织斑点的一阶灰度值进行建模。模型是通过分析从上海市第六人民医院消化内镜中心提供的197张消化道粘膜下肿瘤超声图像取而得到的。在消化粘膜下超声图像中手动截取了正常组织和肿瘤组织图像，将其灰度直方图作为斑点分布的概率密度函数，使用极大似然估计来拟合斑点分布，并利用K-S检验和R2检验来评估超声斑点与五种概率分布之间的拟合程度。在分析消化黏膜正常组织的直方图时，伽玛分布显示出比其他概率分布模型更好的适用性，具有最小的K-S统计值为0.0065和最高的R方值为0.9967。而在分析消化黏膜肿瘤组织的直方图时，威布尔分布相对于其他模型更合适，显示出最小的K-S统计值为0.0096和最大的R方值为0.9956。因此伽马分布更适合描述消化道粘膜下正常组织的斑点特征，而威布尔分布更适合描述消化道粘膜下肿瘤组织的斑点特征。

Abstract: Since the formation of ultrasound speckles depends on the underlying tissue structure, segmentation and classification of tissues and organs in ultrasound images using speckle modeling has become a research hotspot. In this paper, we will explore the probability distribution model that best characterizes speckle in digestive submucosal normal and tumor tissues. In this paper, the first-order gray values of normal and tumor tissue speckles under the digestive mucosa are modeled. The model was obtained by analyzing 197 ultrasound images of digestive submucosal tumors taken from the Gastrointestinal Endoscopy Center of Shanghai Sixth People’s Hospital. Normal and tumor tissue images were manually intercepted in the digestive submucosal ultra-sound images, and their grayscale histograms were used as the probability density function of the speckle distributions, and great likelihood estimation were used to fit the speckle distribu-tions, and the K-S test and the test were used to assess the degree of fit between the ultrasound speckles and the five probability distributions. In analyzing histograms of digestive mucosal normal tissue, the gamma distribution showed better fit than the other probability distribution models, having the smallest K-S statistic value of 0.0065 and the highest R-squared value of 0.9967. whereas, in analyzing histograms of digestive mucosal tumor tissue, the Weibull distribution was a better fit relative to the other models, showing the smallest K-S statistic value of 0.0096 and the largest R-squared value of 0.9956. Thus Gamma distribution is more suitable to describe the speckle characteristics of normal tissues under the digestive mucosa whereas Weibull distribution is more suitable to describe the speckle characteristics of tumor tissues under the digestive mucosa.