CT平扫图像纹理特征鉴别肾脏实性肿瘤良恶性的价值探究

Exploration on the value of image texture features of CT plain scan in differentiating benign and malignant solid tumors of kidney

目的:分析CT平扫图像纹理特征鉴别肾脏实性肿瘤良恶性的价值。方法:选取医院收治的126例肾脏实质肿瘤患者,根据患者病理检查将其分为肾恶性肿瘤组(86例)和肾良性肿瘤组(40例),其中肾恶性肿瘤组再分为肾透明细胞癌(ccRCC)组(60例),乳头状肾细胞癌(pRCC)组(15例),肾嫌色细胞癌(chRCC)组(11例);肾良性肿瘤组再分为肾乏脂性血管平滑肌脂肪瘤(fp-AML)组(28例),肾嗜酸性细胞腺瘤(RO)组(12例)。所有患者均行CT平扫及三期动态增强扫描,选取肿瘤直径最大的层面,包括CT平均灰度值、中等强度、标准差、峰值、偏度及熵。比较各组患者CT相关纹理参数水平,采用受试者工作特征(ROC)曲线分析CT平扫图像纹理特征,鉴别肾脏实性肿瘤良恶性的价值。结果:肾恶性肿瘤组患者CT标准差及偏度明显高于肾良性肿瘤组,熵明显低于肾良性肿瘤组,差异有统计学意义(t=4.170,t=1.995,t=2.163;P<0.05)。pRCC组、chRCC组、fp-AML组及RO组患者CT标准差明显低于ccRCC组,差异有统计学意义(t=3.093,t=3.107,t=3.938,t=2.741;P<0.05);chRCC组及fp-AML组患者偏度明显低于pRCC组、ccRCC组,差异有统计学意义(t_(chRCC组)=2.383,t=2.027;t_(fp-AML组)=4.937,t=5.448;P<0.05);RO组患者偏度明显高于chRCC组及fp-AML组,差异有统计学意义(t=44.812,t=31.850;P<0.05)。CT平扫指标诊断肾恶性肿瘤患者的ROC曲线下面积(AUC)为0.758,灵敏度为79.64%,特异度为78.64%;标准差鉴别ccRCC与pRCC患者的AUC为0.781,ccRCC与chRCC患者的AUC为0.836,ccRCC与pRCC患者的AUC为0.825,ccRCC与fp-AML患者的AUC为0.749,chRCC与pRCC患者的AUC为0.746,pRCC与fp-AML患者的AUC为0.891,fp-AML与RO患者的AUC为0.854。结论:CT平扫图像纹理特征可反应肾脏实性肿瘤的表征、病理分型及分级,在预测肾脏实质肿瘤的良恶方面具有重要价值,其操作具有无创性、可重复性、客观提供病灶信息等优点,可在临床广泛应用。

Objective:To analyze the value of image texture features of computed tomography(CT)plain scan in differentiating benign and malignant solid tumors of kidney.Methods:126 patients with solid tumor of kidney were selected and were divided into renal malignant tumor group(86 cases)and renal benign tumor group(40 cases)according to the pathological examination of them.The renal malignant tumor group was further divided into renal clear cell renal cell carcinoma(ccRCC)group(60 cases),papillary renal cell carcinoma(pRCC)group(15 cases)and chromophobe renal cell carcinoma(chRCC)group(11 cases).And the benign renal parenchymal tumor group was further divided into renal lipoid angiomyolipoma(fp-AML)group(28 cases)and renal oncocyltoma(RO)group(12 cases).All patients underwent CT plain scan and three-phase dynamic contrast-enhanced CT scan.The slice with the largest diameter of tumor was selected,which included CT average gray value,medium intensity,standard deviation,peak value,skewness and entropy.The levels of CT related texture parameters of various groups were compared,and the receiver operating characteristic(ROC)curve was used to analyze the texture features of CT plain scan images so as to identify the value of benign and malignant solid tumors of kidney.Results:The standard deviation and skewness of CT of renal malignant tumor group were significantly higher than those of renal benign tumor group,and the entropy of former was significantly lower than that of renal benign tumor group(t=4.170,t=1.995,t=2.163,P<0.05).The CT standard deviations of pRCC group,chRCC group,fp-AML group and RO group were significantly lower than those of ccRCC group(t=3.093,t=3.107,t=3.938,t=2.741,P<0.05).The skewness of chRCC group and fp-AML group were significantly lower than those of pRCC group and ccRCC group(t_(chRCC group)=2.383,t=2.027,t_(fp-AML group)=4.937,t=5.448,P<0.05),respectively.And the skewness of RO group was significantly higher than that of chRCC group and fp-AML group(t=44.812,t=31.850,P<0.05),respectively.The area under curve(AUC)of ROC curve,sensitivity and specificity of CT plain scan were 0.758,79.64%and 78.64%in diagnosing renal malignant tumor,respectively.The AUC of CT standard deviation in differentiating ccRCC and pRCC was 0.781,and the AUC of that in differentiating ccRCC and chRCC was 0.836,and the AUC of that in differentiating ccRCC and pRCC was 0.825,and the AUC of that in differentiating ccRCC and fp-AML was 0.749,and the AUC of that in differentiating chRCC and pRCC was 0.746,and the AUC of that in differentiating pRCC and fp-AML was 0.891,and the AUC of that in differentiating fp-AML and RO was 0.854.Conclusion:The image texture features of CT plain can reflect the characterization,pathological classification and grading of renal solid tumors,and it has important value in predicting the benign and malignant aspects of renal solid tumors.The operation of that has series of advantages such as non-invasion,repeatability and providing objective lesion information and so on.It can be widely applied in clinical work.