气道内超声模型在预测肺外周病变恶性概率中的探索性应用

An exploratory application of the endobronchial ultrasound model to estimate the probability of malignancy in patients with peripheral pulmonary lesions

目的①分析周围型肺癌的气道内超声图像特点；②筛选出周围型肺癌的气道内超声诊断指标，建立肺外周病变恶性概率估算模型，并评价其临床诊断价值。方法2010年9月1日至2011年10月30日，150例患者人选试验。分析135例气道内超声检查清晰显示病灶图象，明确患者病变病理结果，记录、分析超声图像内部结构特征，观察特征包括：①病灶外形；②病灶边缘连续与否；③病灶边界清晰与否；④病灶内部回声强弱；⑤病变内部回声同质或异质；⑥病灶内支气管充气征；⑦同心圆有无；⑧病灶内血管影；⑨病灶内不规则无回声区。结果①单因素分析发现：病灶外形、病灶边界清晰与否、病变内部回声同质或异质、病灶内支气管充气征、同心圆有无、病灶内不规则无回声区6种特征在周围型肺癌良恶性鉴别中差异有统计学意义；②经二分类多因素Logistic回归分析，最终5个超声特征纳入回归方程，恶性概率预测模型P=1／[l＋e-（-1272＋1.271病灶形状＋1.452边界＋4.645异质回声-1.687支气管充气征＋1.899无回声区）]其中，病灶呈圆形或类圆形、边界清晰、内部回声异质、支气管充气征、不规则无回声区存在者记为1，否则为0；绘制ROC曲线，计算曲线下面积，正确度为86．3％。结论①气道内超声在肺周围型病灶的良恶性鉴别诊断方面，具有较高的应用价值，是一种安全、无创、价廉的影像学检查方法。②多变量联合应用，建立二分类Logistic回归模型，能提高肺周围型病灶的恶性概率辨别的准确性。

Objective ①To analyze the endobronchial ultrasound （EBUS） features of peripheral lung cancer. ②To screen out the EBUS characteristics and establish a model to estimate the probability of maligancy in peripheral pulmonary lesions, and evaluate its clinical diagnostic value. Methods Between September 1st 2010 and October 30th 2011, EBUS were performed in 150 patients with peripheral pulmonary lesions. At last,the EBUS images of the 135 cases were enrolled, who had a definite dignosis. nine image patterns were issued：① the lesion figuration,② the continuous hyperechoic margin outside the lesion,③ the distinct borderline, ④ the internal echo intensity, ⑤ the internal echo density, ⑥ air bronchogram,⑦concentric circle,⑧internal vessels prensence or not, ⑨intemal irregular echo free area. Results ① Through the univariate analysis, six distinct image patterns; the lesion figuration, the distinct borderline,the internal echo density, air bronchogram, concentric circle,internal irregular echo free area were demonstrated significant difference by univariate analysis, the others were not and excluded. ② According to the result of binary multivariable logistic regression analysis on the 6 independent predictors and by calculating the area under ROC curve, we concluded five charactenstics which contribute to predicting the presence of malignancy. The equation of malignancy probability for any patient was： P = 1/[ 1＋e-（ 1. 272＋ 1.271]igura,ion＋l.452borderline＋l. 645h ho-1. 687bronchogra＋l. 899nocchoarca）]（ the figuration is 1 if it is round or nearly round （otherwise 0）, the borderline is 1if it is distinct （otherwise 0）, the heterecho is 1 if the internal echo density is different （otherwise 0）, the air bronehogram is the short line hyperecho region exist （otherwise 0）, nonechoarea is 1 if the internal, irregular echo free area exist）. According to the area of the ROC curve, this model gave a accuracy of 86. 3%. Conclusions ① EBUS is a safe and practical examination method. It can provide physicians a chance to observe the lesions from inside and discover more information than that obtained through radiology in discriminating peripheral pulmonary lesions. ②Multivariable logistic regression model discriminated between benign and malignant lesions is better than univariable logistic regression. Combined multiple variables predicticting the malignant probability eould raise their sensitivity, specificity and accuracy.