基于CART决策树的B超影像肝纤维化纹理识别研究

The Research on the Texture Identification of Hepatic Fibrosis of B-mode Ultrasound Image Based on CART Decision Tree

肝纤维化是一种严重影响患者健康的疾病,B超是确诊肝纤维化的必要手段之一。传统上超声科医师通过阅片做出诊断建议,完全凭借主观判断,医生经验、知识水平和疲劳程度往往决定着诊断结果的准确性。拟利用图像分析技术,建立肝脏纤维化自动识别方法。通过对B超增粗病例和对照肝部B超图片感兴趣区进行不同纹理定义方法的特征分析,使用分类和回归决策树CART对上述影像数据进行学习和建树,并通过10倍的交叉校验对这些方法的识别准确率进行比较,发现灰度共生矩阵进行纤维化的识别的准确率更高一些,达到82.51%。因此,利用分类和回归决策树CART结合灰度共生矩阵纹理特征定义方法进行肝脏纤维化B超图像的识别,准确率高,有很好的应用前景。

Hepatic fibrosis is a disease that serious affects the health of the patients, and B-mode ultrasound is one of the necessary methods for the diagnosis of hepatic fibrosis. Traditionally, ultrasound physicians have made diagnostic suggestion through image reading, relying solely on subjective judgment, so the experience, knowledge, and fatigue of physician often determine the accuracy of the diagnosis. This study attempts to build the identification method of hepatic fibrosis with image analysis technique. We carries characteristic analysis with different texture definition methods onto B-mode ultrasound image region interested and selected of case and control on the liver, the classification and regression decision tree （CART） are used to learn and build the decision tree with above image data, the accuracy of these methods was compared by 10-fold cross validation, and it is found that the accuracy of gray co- occurrence matrix is higher on hepatic fibrosis identification, with accuracy rate of eighty-two point five one percent. So, it will obtain high accuracy and provide a hopeful prospect for application, when identifying the image of hepatic fibrosis in B-mode ultrasound with gray co-occurrence matrix texture feature definition method and CART.