CT灰度直方图对实性肺结节的鉴别诊断价值

Density histogram analysis of CT scan in the differential diagnosis of solid pulmonary nodule

目的:探讨CT灰度直方图对实性肺结节良恶性的鉴别诊断价值。方法:回顾性分析经组织病理学或临床随访证实的110例实性肺结节患者的CT图像,其中恶性55例,良性55例。选取肺结节最大CT平扫轴面图像勾画感兴趣区并采用Matlab软件生成灰度直方图。采用Mann-Whitney U检验比较良恶性结节灰度直方图参数的差异。建立ROC曲线并以组织病理学或临床随访结果为依据计算曲线下面积(AUC)。使用最佳临界值评价灰度直方图诊断良恶性肺结节的敏感度、特异度、准确度、阳性预测值及阴性预测值。结果:恶性结节的偏度(0.281±0.181)大于良性结节(-0.592±0.194),差异有统计学意义(P<0.001);恶性结节的峰度(2.786±0.252)小于良性结节(3.553±0.331),差异有统计学意义(P<0.05),两者的ROC曲线下最大面积分别为0.894和0.996。使用最佳临界值标准,峰度和偏度鉴别诊断肺结节良恶性的敏感度分别为0.982、0.946,特异度分别为1.000、0.764,准确度分别为0.990、0.845,阳性预测值分别为1.000、0.797,阴性预测值分别为0.982、0.913。结论:CT灰度直方图可作为肺结节良恶性鉴别诊断的重要辅助手段。

Objective：To assess the value of density histogram of CT scan in the differential diagnosis of malignant from benign solid pulmonary nodules. Methods：CT images of 110 cases including 55 malignant and 55 benign solid pulmonary nodules proven by pathology or clinical following up were studied retrospectively. Density histogram was created using a software ＂Matlab＂ by drawing a region of interest on the axial unenhanced CT slice covering the [argest area of the nodule. Differences of density histogram between benign and malignant solid pulmonary nodules were evaluated using Mann-Whitney U tests. Receiver operating characteristic （ROC） curves for each were constructed and the area under the curve （AUC） was calculated with histopathology or clinical following up as reference standard. Optimal threshold criterion was used to estimate the sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of density histogram in the differential diagnosis of solid pulmonary nodules. Results：The skewness of malignant （0. 281±0. 181） was higher than that of benign nodules （-0. 592±0. 194） ,with statistic difference （P〈0.05）. The kurtosis of malignant （2. 786±0. 252） was lower than that of benign nodules （3. 553±0. 331） ,with statistic difference （P〈0.05）. The largest AUC of the two was 0. 894 and 0. 996 respectively. Using optimal threshold criteria,the kurtosis and skewness had the sensitivity as 0. 982 and 0. 946, specificity as 1 and 0. 764, accuracy as 0. 990 and 0. 845, positive predictive value as 1 and 0. 797, and negative predictive value as 0. 982 and 0. 913, respectively. Conclusions： CT density histogram has the potential to accurately differentiate malignant from Benign solid nodules in patients with suspected lung cancer.