基于PET图像纹理分析食管癌^(18)F-FDG摄取异质性

^(18)F-FDG uptake heterogeneity of esophageal carcinoma based on PET image textural analysis

目的提取食管癌原发灶PET图像纹理特征,并量化肿瘤18 F-FDG摄取异质性,探讨食管癌18 F-FDG摄取异质性与最大标准摄取值(SUVmax)及病理参数的关系。方法回顾性分析30例术前接受18 F-FDG PET/CT全身扫描的食管癌患者。应用Matlab 7.6软件计算食管癌原发灶PET图像纹理参数(对比度、相关性、熵、能量),分析各纹理参数与SUVmax、肿瘤浸润深度、分化程度及淋巴结转移情况的相关性。结果食管癌原发灶纹理参数均与SUVmax相关,分别与对比度、熵呈正相关(r=0.537,P=0.002;r=0.434,P=0.017),与相关性、能量呈负相关(r=-0.471,P=0.009;r=-0.450,P=0.012);在不同浸润深度和淋巴结转移情况下,纹理参数熵和能量的组间差异有统计学意义(P均<0.05),并分别与熵呈正相关(rs=0.574,P=0.001;rs=0.366,P=0.047),与能量呈负相关(rs=-0.428,P=0.018;rs=-0.436,P=0.016)。各代谢参数与食管癌分化程度无相关性。结论纹理参数可量化18F-FDG摄取异质性,提供能反映肿瘤生物学特征的丰富影像学信息。

Objective To quantify ^（18）F-FDG uptake heterogeneity of esophageal carcinoma using PET image texture features, and to explore the correlation with maximum standardized uptake value （SUVmax） and pathological parameters. Methods Thirty patients with esophageal carcinoma underwent whole body ^（18）F-FDG PET/CT scan before surgical operation were analyzed retrospectively. To quantify TM F-FDG uptake heterogeneity on pre-treatment 18 F-FDG PET images, four texture features （contrast, correlation, entropy and energy） were extracted using Matlab 7.6 software algorithm. The texure parameters were correlated with SUVmax, depth of invasion, differentiation of the primary lesions and lymph metastasis status. Results The primary esophageal tumors with high SUVmax were more heterogeneous on uptake. Correlation was found in contrast （r=0. 537, P=0. 002）, correlation （r=-0. 471, P=0. 009）, entropy （r=0. 434, P=0. 017） and energy （r=- 0. 450, P = 0. 012）, respectively. Depth of invasion was correlated with entropy （r3 = 0.574, P = 0.001 ） and energy （r3 =- 0. 428, P= 0.018）. There were also statistically significant differences in entropy （r3 = 0.366, P = 0. 047） and energy （r3 =- 0. 436, P= 0. 016） between the groups with or without lymph node metastasis. There was no significant correlation between texture features and degree of differentiation. Conclusion Tumor is F-FDG uptake heterogeneity quantified by texture features has potential to provide more functional image information on biological characteristics.