摘要
目的探讨特定双影像组学模型鉴别CT平扫不同大小肺结节良恶性及浸润性的能力。资料与方法回顾性分析4892个CT平扫肺结节病灶,由高年资医师勾画病灶并标注病灶的良恶性及浸润性(其中2014个结节病理确诊为恶性),依据所有病灶整体体积分布情况(上、下四分位数为界值)分为大、中、小结节组,使用达尔文科研平台的双影像组学模型对所有病灶进行特征提取并分析,判断病灶的良恶性及浸润程度。以医师组标注结果为正确结果进行统计分析,判断在不同大小肺结节上的模型准确率。结果组学模型对病灶体积>434.75 mm3(平均直径9.4 mm)的大结节病灶判定准确率为92%,尤其对恶性浸润性腺癌判断准确率为100%;对病灶体积<51.63 mm3(平均直径4.6 mm)的小病灶判断准确率为100%;对于介于两者中间的中等大小病灶,在浸润程度由低到高的预测准确度分别为0.33、0.61及0.42。结论基于影像组学的模型对较大肺结节病灶良恶性判断准确性较高,尤其是恶性程度较高的浸润性腺癌,可用于辅助医师对恶性病变的进一步判定及诊断。
Purpose To evaluate a dual-function CT-based radiomics model in differentiating the malignancy and invasiveness of pulmonary nodules(PN)in different sizes.Materials and Methods A total of 4892 PN on CT images were retrospectively analyzed.All nodules(including 2014 pathologically diagnosed nodules)were delineated by a qualified senior radiologist.The malignancy and invasiveness of these PN were evaluated and classified into three groups(small,medium and large group)based on the total volume distribution(upper and lower quartile).Radiomic features of all lesions were extracted and analyzed to evaluate the malignancy and invasiveness of these nodules via Darwin scientific research platform.The accuracy performances of the model were evaluated and statistically analyzed according to the radiologists’diagnosis as the correct standard.Results The accuracy for predicting PN malignancy was 92%in large PN group(lesion volume>434.75 mm3,mean diameter>9.4 mm),particularly,100%accuracy both in distinguishing malignant invasive adenocarcinoma and diagnosing in the small PN group(lesion volume<51.63 mm3,diameter<4.6 mm).However,the relatively poor accuracy was showed in distinguishing the invasiveness degree of medium group,with 0.33,0.61 and 0.42,respectively.Conclusion The dual-function CT-based radiomics model presented good accuracy performance in large PN group,especially in invasive adenocarcinoma group,which can be used to assist radiologists to further diagnose malignancy PN.
作者
张力
肖丹丹
ZHANG Li;XIAO Dandan(Department of Medical Engineering,Beijing Hospital Traditional Chinese Medicine,Beijing 101300,China;不详)
出处
《中国医学影像学杂志》
CSCD
北大核心
2021年第5期514-518,共5页
Chinese Journal of Medical Imaging
关键词
肺结节
体层摄影术
螺旋计算机
影像组学
达尔文智能科研平台
包裹式特征筛选方法
逻辑回归
支持向量机
诊断
鉴别
Pulmonary nodule
Tomography,spiral computed
Radiomics
Darwin scientific research platform
Recursive feature elimination
Logistic regression
Support vector machine
Diagnosis,differential
