摘要
目的探究基于多序列MRI影像组学列线图预测脑膜瘤术后脑水肿严重程度的临床应用价值。方法回顾性分析2019年1月至2022年4月在蚌埠医学院第一附属医院经手术病理证实为脑膜瘤患者170例的术前MRI图像、术后CT图像及临床资料,轻度脑水肿患者101例、重度脑水肿患者69例。随机的将病例按照7:3分为训练组(119例)和验证组(51例)。采用医准-达尔文AI平台获得T1WI、T2WI及T1WI+C图像的10个影像组学特征。根据最大绝对值归一化和选择算子回归进行特征降维、最优特征筛选(个数)最后采用逻辑回归建立模型,在此模型基础上绘制列线图。三种不同模型预测脑膜瘤术后水肿程度的诊断效能用特征曲线(ROC)下的面积(AUC)评估。结果基于临床特征构建的临床模型、基于影像特征构建的影像模型及联合临床资料和影像特征构建的临床-影像组学模型在训练组和验证组的AUC均大于0.80,具有一定的预测效能。其中基于临床资料和影像特征构建的临床-影像组学模型在训练组(AUC=0.90)和验证组(AUC=0.91)均高于临床模型和影像组学模型。结论基于临床-影像组学模型列线图能以最大化的准确性预测脑膜瘤术后脑水肿的严重程度。
Objective To explore the clinical value of multi-sequence MRI radiomics nomogram in predicting the severity of brain edema after meningioma surgery.Methods The preoperative MRI images,postoperative CT images and clinical data of 170 patients with meningioma confirmed by surgical pathology in The First Affiliated Hospital of Bengbu Medical College from January 2019 to April 2022 were retrospectively analyzed.There were 101 patients with mild cerebral edema and 69 patients with severe cerebral edema.The patients were randomly divided into training group(119 cases)and validation group(51 cases)according to 7:3.Darwin platform was used to extract 10 radiomics features of T1WI,T2WI and T1WI+C images.According to the maximum absolute value normalization and selection operator regression,feature dimension reduction and optimal feature screening(number)were carried out.Finally,logistic regression was used to establish the model,and the nomogram was drawn on the basis of the model.The diagnostic efficacy of three different models in predicting the degree of postoperative edema in meningiomas was evaluated by the area under the characteristic curve(AU C).Results The AU C of the clinical model based on clinical features,the image model based on image features and the clinical-radiomics model combined with clinical data and image features were all greater than 0.800 in the training group and the validation group,showing good predictive performance.The radiomic model based on clinical data and imaging features was significantly higher in the training group(AU C=0.90)and validation group(AU C=0.91)than in the clinical model and radiomics model.Conclusion Nomogram based on clinical-radiomics model can predict the severity of brain edema after meningioma surgery with maximum accuracy.
作者
张传敏
刘娟娟
王瑞瑞
石士奎
ZHANG Chuan-min;LIU Juan-juan;WANG Rui-rui;SHI Shi-kui(The First Affiliated Hospital of Bengbu Medical College,Department of Radiology,Bengbu 233000,Anhui Province,China)
出处
《中国CT和MRI杂志》
2023年第8期26-29,共4页
Chinese Journal of CT and MRI
