CT成像的放射组学特征在胰腺炎诊断中的评估效能

Evaluation efficacy of radiomic features of CT imaging in the diagnosis of pancreatitis

目的探讨胰腺CT成像的放射组学特征在功能性腹痛(FAP)、复发性急性胰腺炎(RAP)、慢性胰腺炎(CP)患者诊断中的评估效能。方法回顾性分析2017年12月至2020年12月在首都医科大学附属北京天坛医院接受腹部增强CT检查的168例患者的CT影像资料,根据纳排标准,共选取48例患者进行研究,其中男性23例、女性25例,年龄39~84(47.8±10.2)岁;FAP患者16例(FAP组)、RAP患者18例(RAP组)、CP患者14例(CP组)。通过在CT图像上绘制感兴趣区来勾画胰腺轮廓。从每个感兴趣区提取62个放射组学特征,共分为5类,分别是一阶统计量、灰度共生矩阵(GLCM)、灰度行程矩阵(GLRLM)、邻域灰度差矩阵(NGTDM)和形态学特征,并在3组患者间进行比较。根据组别建立3个IsoSVM机器学习模型,对IsoSVM模型进行训练,并在遗漏的交叉验证样本上进行测试。RAP患者、FAP患者、CP患者的放射组学特征的比较采用Wilcoxon秩和检验。使用受试者工作特征曲线及曲线下面积(AUC)评估个体放射组学特征的评估效能。结果在单变量分析中,9个放射组学特征(8个GLCM特征和1个NGTDM特征)在患者组间的差异有统计学意义(Z=3.45~29.76,均P<0.05)。RAP患者与FAP和CP患者分别进行比较,放射组学特征的AUC范围分别为0.76~0.93和0.73~0.91。IsoSVM机器学习模型的总体预测准确率为82.1%。FAP组的灵敏度、特异度分别为78.7%、100%,AUC为0.90。RAP组的灵敏度、特异度分别为95.2%、77.8%,AUC为0.87,而CP组的灵敏度、特异度分别为70.9%、94.8%,AUC为0.89。结论CT成像的部分放射组学特征对胰腺炎的诊断有较好的评估效能,可以区分FAP、RAP和CP患者。

Objective To evaluate the diagnostic efficacy of the radiological features of pancreatic CT imaging in patients with functional abdominal pain(FAP),recurrent acute pancreatitis(RAP),and chronic pancreatitis(CP).Methods Retrospective analysis was performed on the CT image data of 168 patients who received abdominal enhanced CT examination in Beijing Tiantan Hospital,Capital Medical University from December 2017 to December 2020.According to the criteria for admission and emission,48 patients were selected for the study,including 23 males and 25 females,aged 39 to 84(47.8±10.2)years,and 16 cases of FAP(FAP group),18 cases of RAP(RAP group),and 14 cases of CP(CP group).The pancreas outline was obtained by drawing the region of interest on the CT image.Sixty-two radiologic features were extracted from each region of interest,which were divided into five categories,namely,the first-order statistics,the gray-level co-occurrence matrix(GLCM),the gray-level run-length matrix,the neighbouring gray tone difference matrix(NGTDM),and the morphological features,and compared among the three groups.According to the groups,three IsoSVM machine learning models were established,trained,and tested on the missing cross validation samples.The Wilcoxon rank sum test was used to compare the radiation characteristics of the patients with RAP,FAP,and CP.The predictive performance of individual radiological characteristics was evaluated using the receiver operator characteristic curve and the area under the curve(AUC).Results In the univariate analysis,a significant difference was found between the patient groups in nine radiation group characteristics(eight GLCM characteristics and one NGTDM characteristic)(Z=3.45–29.76,all P<0.05).Compared RAP patients with FAP and CP patients,the AUC ranges were 0.76–0.93 and 0.73–0.91.The overall prediction accuracy of the IsoSVM machine learning model was 82.1%.The sensitivity and specificity of the FAP group were 78.7%and 100%,respectively,and the AUC was 0.90.The sensitivity and specificity of the RAP group were 95.2%and 77.8%,respectively,and the AUC was 0.87,while those of the CP group were 70.9%,94.8%,and 0.89,respectively.Conclusion Some of the radiographic features of CT imaging have a good evaluation efficiency in the diagnosis of pancreatitis and can distinguish between patients with FAP,RAP,and CP.