摘要
目的 探讨结合机器学习早期动态增强磁共振成像(DCE-MRI)的影像组学模型在鉴别良恶性乳腺非肿块强化(NME)病变中的价值。方法 选取行乳腺DCE-MRI检查并获得病理结果的NME病变患者242例,分为训练集163例、测试集55例,外部验证集24例。基于早期DCE-MRI序列的特征选择,采用支持向量机(SVM)建立组学预测模型;由2位放射科医师独立评估MRI特征,建立传统诊断模型,预测病灶的良恶性;运用测试集和外部验证集进行测试和外部验证。采用受试者工作特征(ROC)曲线评价组学模型与放射医师的诊断效能。结果 影像组学模型鉴别乳腺NME病变良恶性达到了与高年资放射医师[曲线下面积(AUC)=0.82, 95%CI 0.66, 0.89]相当的诊断水平[(AUC=0.82, 95%CI 0.67, 0.90);P=0.30],均优于低年资放射医师的评估结果(Z=2.63,P=0.01;Z=2.41,P=0.02),同时利用外部验证集进一步验证该模型的预测效能。结论 基于早期DCE-MRI组学模型可以有效地鉴别NME病变的良恶性,与高年资放射医师诊断水平相当,并优于低年资医师诊断水平,可以辅助低年资医师做出更佳诊断。
Objective To explore the value of radiomics model combined with machine learning of early dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)in distinguishing benign and malignant breast non-mass enhancement(NME)lesions.Methods 242 patients with NME lesions,who underwent breast DCE-MRI examination and obtained pathological results,were selected anddivided into training set of 163 lesions,testing set of 55 lesions,and external validation set of 24 lesions.Based on feature selection of early DCE-MRI,radiomics model was to established by support vector machine.Two radiologists independently evaluated MRI features,established a traditional diagnostic model,and predicted the benign and malignant lesions.Test sets and external validation sets were used for testing and external validation.The receiver operating characteristic(ROC)curve was used to evaluate the diagnostic efficiency of the radiomics model and radiologists.Results The differentiation of benign and malignant breast NME lesions using radiomics models reached a diagnostic level comparable to that of senior radiologists[AUC=0.82,95%confidence interval(CI)0.66,0.89][(AUC=0.82,95%CI 0.67,0.90);P=0.30],both of which were superior to the evaluation results of junior radiologists(Z=2.63,P=0.01;Z=2.41,P=0.02),Meanwhile,external validation sets were used to further validate the predictive performance of the model.Conclusion Based on the early DCE-MRI radiomics model,it can effectively distinguish between benign and malignant NME lesions,which is comparable to the diagnostic level of senior radiologists and superior to the diagnostic level of junior radiologists.This can assist junior radiologists in making better diagnoses.
作者
李珍
刘磊
仲海
王翠艳
LI Zhen;LIU Lei;ZHONG Hai;WANG Cuiyan(Shandong University,Jinan 250100,China;Department of Medical Imaging,Shandong Provincial Hospital,Jinan 250021,China;Shandong Zhongyang Health Technology Group Co.Ltd,Jinan 250101,China;Department of Medical maging,The Second Hospital of Shandong University,Jinan 250031,China;Department of Medical Imaging,Shandong Provincial Hospital Affiliated to Shandong First Medical University,Jina 250021,China.)
出处
《医学影像学杂志》
2024年第2期46-51,共6页
Journal of Medical Imaging
基金
山东省医学会乳腺疾病科研基金项目(编号:YXH2020ZX068)。
关键词
影像组学
磁共振成像
乳腺非肿块强化病变
Radiomics
Magnetic resonance imaging
Non-mass enhancement lesions of breast
