摘要
目的:分析CT平扫的2D与3D影像组学模型诊断卵巢良恶性病变价值。方法:回顾性收集2017年7月-2022年7月本院收治的经病理学证实的卵巢良恶性病变患者100例临床资料,根据病理学结果分为恶性组(n=30)和良性组(n=70),CT平扫图像上勾画2D、3D肿瘤感兴趣区并提取图像特征,按照7:3的比例随机分层分为训练集(n=70)与验证集(n=30),提取CT影像组学特征,多因素logistic回归构建2D与3D影像组学模型,采用受试者工作特征(ROC)曲线评估2D与3D影像组学模型对卵巢良恶性病变的诊断效能并比较。结果:以肿块形态、肿瘤囊实性、腹水作为构建2D影像学特征模型,该模型训练集诊断卵巢良恶性病变的敏感度、特异度、曲线下面积(AUC)为88.9%、77.0%、0.86;验证集诊断卵巢良恶性病变的敏感度、特异度、AUC为81.8%、78.9%、0.82。以肿块形态、肿瘤囊实性、边界、腹水作为构建3D影像学特征模型,该模型训练集诊断卵巢良恶性病变的敏感度、特异度、AUC为90.7%、76.7%、0.86;验证集诊断敏感度、特异度、AUC为81.8%、73.7%、0.86。2D与3D影像组学模型诊断卵巢良恶性病变的敏感度、特异度及AUC未见差异(P>0.05)。结论:基于CT平扫的2D与3D影像组学模型诊断卵巢良恶性病变价值相当且均较高,但考虑到影像组学特征计算成本,更推荐使用2D影像组学模型。
Objective:To analyze the values of CT scan-based 2D and 3D radiomics models for diagnosing the benign and malignant ovarian lesions of patients.Methods:The clinical data of 100 patients with benign or malignant ovarian lesions who had been confirmed by pathology from July 2017 to July 2022 were collected retrospectively.According to the pathological results,the patients were divided into group A(30 patients with malignant)and group B(70 patients with benign ovarian lesions).2D and 3D regions of interested by tumor were delineated on CT scan images,and the image features were extracted.According to the ratio of 7:3,these images were randomly divided into the training set(n=70)and the validation set(n=30).The CT radiomics features were extracted,and 2D and 3D radiomics models were constructed by multivariate logistic regression.The area under the ROC curve(AUC)was used to evaluate the diagnostic efficacy of 2D and 3D radiomics models for the benign and malignant ovarian lesions.The diagnostic efficacy for the benign and malignant ovarian lesions was compared between 2D and 3D radiomics models.Results:The 2D imaging feature model was constructed based on the morphology of the lump,cystic and solidity tumor,and ascites.The sensitivity,the specificity,and the AUC of the training set of 2D imaging feature model for diagnosing the benign and malignant ovarian lesions were 88.9%,77.0% and 0.86.The sensitivity,the specificity,and the AUC of 2D imaging feature model for the validation set were 81.8%,78.9%,and 0.82.The 3D imaging feature model was constructed based on the morphology of the lump,cystic and solidity tumor,boundary and ascites.The sensitivity,the specificity,and the AUC of the training set of 3Dimaging feature model for diagnosing the benign and malignant ovarian lesions were90.7%,76.7%,and 0.86.The sensitivity,the specificity,and the AUC of 2Dimaging feature model for the validation set were 81.8%,73.7%,and 0.86.There were no statistically significant differences in the sensitivity,the specificity,and the AUC
作者
张永华
胡苗苗
金煜芳
丁远辉
ZHANG Yonghua;HU Miaomiao;JIN Yufang;DING Yuanhui(Wuxing District People's Hospital,Huzhou,Zhejiang Province,313008;The First People's Hospital of Huzhou City)
出处
《中国计划生育学杂志》
2023年第11期2733-2737,共5页
Chinese Journal of Family Planning
