Alemzadeh Mehrdad, Boylan Colm, Kamath Markad V
Department of Computing and Software, McMaster University.
Department of Radiology, McMaster University and St. Joseph's Health Care Hamilton.
Crit Rev Biomed Eng. 2015;43(2-3):183-200. doi: 10.1615/CritRevBiomedEng.2015011026.
Computer-based identification of abnormal regions and classification of diseases using CT images of the lung has been a goal of many investigators. In this paper, we review research that has used texture analysis along with segmentation and fractal analysis. First, a review of texture methods is performed. Recent research on quantitative analysis of the lung using texture methods is categorized into six groups of computational methods: structural, statistical, model based, transform domain, texture-segmentation, and texture-fractal analysis. Finally, the applications of texture-based methods combined with either segmentation algorithms or fractal analysis is evaluated on lung CT images from patients with diseases such as emphysema, COPD, and cancer. We also discuss applications of artificial neural networks, support vector machine, k-nearest, and Bayesian methods to classify normal and diseased segments of CT images of the lung. A combination of these texture methods followed by classifiers could lead to efficient and accurate diagnosis of pulmonary diseases such as pulmonary fibrosis, emphysema, and cancer.
利用肺部CT图像通过计算机识别异常区域并进行疾病分类一直是众多研究人员的目标。在本文中，我们回顾了使用纹理分析以及分割和分形分析的研究。首先，对纹理方法进行了综述。近期使用纹理方法对肺部进行定量分析的研究被归类为六组计算方法：结构法、统计法、基于模型法、变换域法、纹理分割法和纹理分形分析法。最后，在患有诸如肺气肿、慢性阻塞性肺疾病（COPD）和癌症等疾病患者的肺部CT图像上，评估了基于纹理的方法与分割算法或分形分析相结合的应用。我们还讨论了人工神经网络、支持向量机、k近邻法和贝叶斯方法在对肺部CT图像的正常和病变部分进行分类中的应用。这些纹理方法与分类器相结合可能会导致对诸如肺纤维化、肺气肿和癌症等肺部疾病进行高效且准确的诊断。
