基于常规MRI成像及纹理分析技术预测垂体大腺瘤质地的价值

Prediction of pituitary macroadenoma consistency using MRI and texture analysis

目的:探讨基于常规磁共振成像(MRI)及纹理分析技术预测垂体大腺瘤质地的价值。方法:回顾性分析经手术病理证实的103例垂体大腺瘤患者的临床及影像资料。所有患者术前均经常规MRI平扫及增强检查。在MRI图像上测量肿瘤最大径、分析肿瘤信号特点并手动勾画ROI提取纹理特征参数。单因素分析两组间一般资料、肿瘤最大径、MRI信号和纹理参数,并对两组间有统计意义的变量采用多因素Logistic回归分析建立预测垂体腺瘤质地的联合模型,绘制受试者工作特征(ROC)曲线并评估模型诊断效能。结果:2种质地垂体大腺瘤最大径、T2WI信号间差异有统计学意义(P<0.01),曲线下面积(AUC)分别为0.774、0.688;最终筛选获得9个价值较大的纹理参数,鉴别2种质地肿瘤效能AUC为0.659~0.843;多因素回归分析显示,联合模型鉴别2种质地肿瘤的效能最高(AUC=0.995)。结论:垂体大腺瘤质地与肿瘤最大径、T2WI信号和基于常规MRI提取的纹理参数有关;联合模型对于垂体大腺瘤质地的判断具有更高的准确性。

Objective:To evaluate the potential value of magnetic resonance imaging(MRI)and texture analysis in predicting pituitary macroadenoma consistency.Methods:The clinical and imaging data were retrospectively analyzed in 103 patients with pituitary macroadenoma confirmed by postoperative pathology.All patients underwent preoperative plain MRI and enhanced MRI scans.MRI images were used to measure the maximum diameter of tumor,analyze signal characteristics and extract texture feature parameters by manually delineating the regions of interest(ROI).Univariate analysis was used to analyze the clinical data,tumor maximum diameter,MRI signals and texture parameters of the two groups,and then significant variables were assessed by multivariate logistic regression analysis.A joint model was established to predict pituitary macroadenoma consistency and its efficacy was evaluated by receiver operating characteristic(ROC)curve.Results:There was significant difference in maximum diameter and T2WI signal between two groups of pituitary macroadenomas(P<0.01),and the area under AUC was 0.774 and 0.688,respectively.Nine valuable parameters with featured texture were identified.The AUC ranged from 0.659 to 0.843 in differential diagnosis of the two tumors.Multivariate regression analysis showed that the joint model had the highest efficiency in identifying pituitary macroadenomas with different consistency(AUC=0.995).Conclusion:The consistency of pituitary macroadenomas is related to the maximum diameter of tumor,T2WI signal and texture feature parameters extracted from conventional MRI images.The joint model has higher accuracy in identifying the consistency.