胡小丽,顾潜彪,张堃,李磊,李平,沈宏荣.基于CT平扫影像组学鉴别≤2 cm甲状腺良恶性结节[J].中国介入影像与治疗学,2021,18(2):105-108
基于CT平扫影像组学鉴别≤2 cm甲状腺良恶性结节
Radiomics based on plain CT for differential diagnosis of ≤ 2 cm benign and malignant thyroid nodules
投稿时间:2020-07-22  修订日期:2020-12-10
DOI:10.13929/j.issn.1672-8475.2021.02.010
中文关键词:  甲状腺肿瘤  诊断,鉴别  体层摄影术,X线计算机  影像组学
英文关键词:thyroid neoplasms  diagnosis, differential  tomography, X-ray computed  radiomics
基金项目:国家自然科学基金青年科学基金(81603482)、中国博士后科学基金面上项目(2017M622586)、湖南省自然科学基金(2016JJ6115)、湖南中医药大学重点学科建设项目(4901-020000200806)。
作者单位E-mail
胡小丽 湖南中医药大学第一附属医院放射科, 湖南 长沙 410007  
顾潜彪 湖南省人民医院放射科, 湖南 长沙 410005  
张堃 湖南中医药大学第一附属医院放射科, 湖南 长沙 410007 kunzhang0102@163.com 
李磊 永州市中心医院放射科, 湖南 永州 425000  
李平 湖南中医药大学第一附属医院放射科, 湖南 长沙 410007  
沈宏荣 湖南中医药大学第一附属医院放射科, 湖南 长沙 410007  
摘要点击次数: 118
全文下载次数: 45
中文摘要:
      目的 观察基于CT平扫影像组学模型鉴别直径≤2 cm甲状腺良恶性结节的价值。方法 回顾性分析97例经手术病理证实直径≤2 cm甲状腺结节患者,按7∶3比例将其随机分为训练组(n=67)及验证组(n=30)。提取训练组病灶的影像组学特征并进行预处理,采用最小绝对收缩和选择算子(LASSO)方法筛选最优影像组学特征;以二元Logistic回归方法建立鉴别甲状腺结节良恶性的影像组学模型,计算训练组中良恶性结节的影像组学评分,于验证组中进行验证。以ROC曲线法评估影像组学模型在训练组及验证组中的诊断效能。结果 训练组25例良性病变和42例恶性病变;验证组10例良性、20例恶性病变。共于训练组病灶中提取848个影像组学特征,经预处理及筛选获得8个最优影像组学特征,建立鉴别直径≤2 cm甲状腺良恶性结节的影像组学模型。2组良性结节影像组学评分[-0.08(-1.96,0.78)、0.11(-0.96,0.65)]均低于恶性结节[1.20(0.80,2.56)、1.03(0.80,2.47),t=-3.29、-3.12,P均<0.01]。影像组学模型鉴别训练组及验证组甲状腺良恶性病变的敏感度分别为0.77、0.74,特异度分别为0.79、0.91,AUC分别为0.84、0.88(D=0.35,P=0.73)。结论 基于CT平扫影像组学鉴别直径≤2 cm甲状腺良恶性结节具有较好的应用价值。
英文摘要:
      Objective To investigate the value of radiomics model based on plain CT for differentiating benign and malignant thyroid nodules with diameter ≤2 cm. Methods CT and clinical data of 97 patients with thyroid nodules diameter ≤2 cm confirmed by surgical pathology were retrospectively analyzed. The patients were randomly divided into training group (n=67) and validation group (n=30) according to the ratio of 7∶3. The radiomics features of every lesion were extracted and preprocessed in training group, then the optimal radiomics features were screened with the method of least absolute shrinkage and selection operator (LASSO). In training group, binary Logistic regression analysis was used to establish radiomics model for distinguishing benign and malignant thyroid nodules, and the radiomics score of nodules were calculated. Subsequently, the efficiency of the model was validated in validation group. ROC curve analysis was performed to evaluate the diagnostic performance of this radiomics model in 2 groups. Results In training group, there were 25 patients with benign nodules and 42 with malignant nodules, while in validation group, there were 10 cases with benign and 20 with malignant nodules. A total of 848 radiomic features were extracted from every lesion in training group. After feature preprocessing and screening, 8 optimal radiomics features were obtained to establish model for distinguish benign and malignant thyroid nodules with diameter ≤2 cm. The radiomics scores of benign nodules in both training group and validation group (-0.08 [-1.96, 0.78], 0.11 [-0.96, 0.65]) were all lower than that of malignant nodules (1.20[0.80, 2.56], 1.03 [0.80, 2.47], t=-3.29, -3.12, both P<0.01). The sensitivity of this model in training group and validation group was 0.77 and 0.74, specificity was 0.79 and 0.91, respectively, and the corresponding AUC was 0.84 and 0.88 (D=0.35, P=0.73). Conclusion Radiomics based on plain CT had good diagnostic value for differential diagnosis of benign and malignant thyroid nodules with diameter ≤2 cm.
查看全文  查看/发表评论  下载PDF阅读器
关闭