Katsuragawa S, Doi K, MacMahon H, Sasaki Y, Yanagisawa T
Department of Radiology, Iwate Medical University.
Nihon Igaku Hoshasen Gakkai Zasshi. 1990 Jul 25;50(7):753-66.
We are developing an automated method for determination of quantitative physical measures of lung textures in digital chest radiographs in order to detect and characterize interstitial lung disease. We describe a scheme of our approach for lung texture analysis, an automated classification method for distinction between normal and abnormal lungs with interstitial disease, and the effect of digital parameters on the accuracy of this computerized analysis, as well as applications of this method to the ILO pneumoconioses standard radiographs. The root-mean-square (rms) variation and the first moment of the power spectrum of the lung texture were determined as quantitative texture measures based on a frequently analysis of lung textures, which represent the magnitude and coarseness (or fineness) of the lung textures, respectively. The computerized classification method is based on the analysis of these texture measures and on a data base derived from clinical cases. This classification method includes three independent tests, one for a definitely abnormal focal pattern, one for a relatively localized abnormal pattern, and one for a diffuse abnormal pattern. A comparison of receiver operating characteristic (ROC) curves obtained by radiologists and by means of the computerized classification method indicates that the computerized approach may provide performance similar to human observers in distinguishing lungs with mild interstitial disease from normal lungs. By investigating the effect of digital parameters such as pixel size, ROI size and the number of quantization levels on these texture measures obtained from the lung texture analysis and the performance of this computerized method, we attained a useful guide in the design of this computerized scheme. Texture measures obtained from computer analysis of the ILO pneumoconioses standard radiographs corresponded closely with the ILO classification categories for small opacities, though it was necessary for a qualified radiologist to identify representative areas in each ILO radiograph because of the inhomogeneous distribution of texture patterns in these standard radiographs. Our results suggest strongly that this computerized method can be a valuable aid to radiologists in their assessment of interstitial lung diseases.
我们正在开发一种自动方法，用于确定数字化胸部X光片中肺纹理的定量物理指标，以便检测和表征间质性肺病。我们描述了肺纹理分析方法的方案、用于区分正常肺和患有间质性疾病的异常肺的自动分类方法、数字参数对这种计算机化分析准确性的影响，以及该方法在国际劳工组织尘肺病标准X光片上的应用。基于对肺纹理的频域分析，确定了肺纹理的均方根（rms）变化和功率谱的一阶矩作为定量纹理指标，它们分别代表肺纹理的幅度和粗糙度（或精细度）。该计算机化分类方法基于对这些纹理指标的分析以及来自临床病例的数据库。这种分类方法包括三个独立测试，一个用于明确异常的局灶性模式，一个用于相对局限的异常模式，一个用于弥漫性异常模式。放射科医生和通过计算机化分类方法获得的接收者操作特征（ROC）曲线的比较表明，在区分轻度间质性疾病的肺与正常肺方面，计算机化方法可能提供与人类观察者相似的性能。通过研究像素大小、感兴趣区域（ROI）大小和量化级别数量等数字参数对从肺纹理分析获得的这些纹理指标以及这种计算机化方法性能的影响，我们在该计算机化方案的设计中获得了有用的指导。从国际劳工组织尘肺病标准X光片的计算机分析中获得的纹理指标与国际劳工组织关于小阴影的分类类别密切对应，不过由于这些标准X光片中纹理模式分布不均匀，需要合格的放射科医生在每张国际劳工组织X光片中识别代表性区域。我们的结果强烈表明，这种计算机化方法在放射科医生评估间质性肺病时可以成为有价值的辅助工具。
