Uppaluri R, Hoffman E A, Sonka M, Hunninghake G W, McLennan G
Department of Electrical and Computer Engineering, University of Iowa, Iowa City 52242, USA.
Am J Respir Crit Care Med. 1999 Feb;159(2):519-25. doi: 10.1164/ajrccm.159.2.9707145.
We have previously described an adaptive multiple feature method (AMFM) for the objective assessment of global and regional changes in pulmonary parenchyma to detect emphysema. This computerized method uses a combination of statistical and fractal texture features for characterization of lung tissues based upon high resolution computed tomography (HRCT) scans. This present study was a substantial extension of the AMFM to simultaneously discriminate between multiple pulmonary disease processes. Normal subjects and those with emphysema, idiopathic pulmonary fibrosis (IPF), or sarcoidosis were studied. The AMFM was compared with two currently utilized computer-based methods: mean lung density (MLD) and the histogram analysis (HIST). Globally, when comparing two-subject groups the AMFM overall accuracy was 2 to 18% better than the overall accuracy of MLD and as much as 36% better than the accuracy of the HIST methods. In three-subject group discrimination tasks, the AMFM performed 7 to 27% better than the MLD and 4 to 36% better than the HIST methods. Finally, in discriminating all four subject groups at a time, the AMFM overall accuracy was 81%, which was 21% better than the MLD and 25% better than the HIST method. In most three-subject group comparisons and in the four-subject group comparison, the AMFM was significantly (p < 0.01) better than the MLD and HIST methods. Next, the AMFM was applied to local discrimination between normal and each disease group individually. The normal versus emphysema, normal versus IPF, and normal versus sarcoidosis samples were discriminated with an accuracy of 95, 86, and 77%, respectively. The AMFM is an objective quantitative method that can be adapted for successful discrimination of multiple parenchymal lung diseases.
我们之前描述了一种自适应多特征方法（AMFM），用于客观评估肺实质的整体和区域变化以检测肺气肿。这种计算机化方法基于高分辨率计算机断层扫描（HRCT）扫描，使用统计和分形纹理特征的组合来表征肺组织。本研究是对AMFM的实质性扩展，以同时区分多种肺部疾病进程。研究了正常受试者以及患有肺气肿、特发性肺纤维化（IPF）或结节病的受试者。将AMFM与两种目前使用的基于计算机的方法进行比较：平均肺密度（MLD）和直方图分析（HIST）。总体而言，在比较两个受试者组时，AMFM的总体准确率比MLD的总体准确率高2%至18%，比HIST方法的准确率高多达36%。在三受试者组区分任务中，AMFM的表现比MLD好7%至27%，比HIST方法好4%至36%。最后，在一次区分所有四个受试者组时，AMFM的总体准确率为81%，比MLD高21%，比HIST方法高25%。在大多数三受试者组比较和四受试者组比较中，AMFM明显（p < 0.01）优于MLD和HIST方法。接下来，将AMFM应用于正常组与每个疾病组之间的局部区分。正常组与肺气肿组、正常组与IPF组以及正常组与结节病组样本的区分准确率分别为95%、86%和77%。AMFM是一种客观定量方法，可适用于成功区分多种实质性肺部疾病。
