摘要
目的探讨基于多参数磁共振成像(magnetic resonance imaging,MRI)的影像组学模型对非肿块型乳腺癌和非哺乳期乳腺炎(non-lactating mastitis,NLM)的鉴别诊断价值。材料与方法回顾性收集2020年6月至2024年6月于新疆医科大学附属中医医院经病理证实为非肿块型乳腺癌和NLM的患者MRI资料共193例,其中非肿块型乳腺癌100例,NLM 93例。两组患者病灶总数225个,其中乳腺癌110个(48.89%),NLM 115个(51.11%)。按7∶3随机划分为训练集(157例)和测试集(68例),采用支持向量机(support vector machines,SVM)机器学习算法对动态增强磁共振成像(dynamic contrast-enhanced magnetic resonance imaging,DCE-MRI)第1、4、7期(即CE1、CE4、CE7)、T2加权成像(T2-weighted imaging,T2WI)和扩散加权成像(diffusion weighted imaging,DWI)这5个序列的数据分别构建单序列模型、多参数MRI模型,并联合5个序列数据和临床特征建立融合模型,通过受试者工作特征(receiver operating characteristic,ROC)曲线、校准曲线、决策曲线分析(decision curve analysis,DCA)评价不同模型的性能,并使用SHAP图形对模型进行解释及可视化。结果单序列模型进行对比,CE1、CE4、CE7、T2WI和DWI序列在测试集的曲线下面积(area under the curve,AUC)分别为0.768、0.804、0.746、0.769、0.812,DWI在测试集的AUC最高,其次是CE4;多参数MRI模型在测试集的AUC为0.840(95%置信区间:0.749~0.932),而融合模型在测试集的AUC为0.866(95%置信区间:0.783~0.948),与CE1、CE4、CE7、T2WI单序列模型相比差异均有统计学意义(P<0.01)。结果显示,融合模型的准确度最高(77.94%);融合模型敏感度最高(90.00%);融合模型和CE4序列的特异度最高(均为68.42%)。结论多参数MRI联合临床特征的融合模型准确度、敏感度和特异度较高,与单序列模型、多参数MRI模型相比预测性能更优,可以为非肿块型乳腺癌和NLM的鉴别诊断提供较高价值。
Objective:To investigate the value of imaging omics model based on multimodal magnetic resonance imaging(MRI)in the differential diagnosis of non mass breast cancer and non lactating mastitis(NLM).Materials and Methods:The MRI data of 193 patients with non mass breast cancer and NLM confirmed by pathology in the First Affiliated Hospital of Traditional Chinese Medicine,Xinjiang Medical University from June 2020 to June 2024 were retrospectively collected,including 100 cases of non mass breast cancer and 93 cases of NLM.The total number of lesions in the two groups was 225,including 110 breast cancer(48.89%)and 115 NLM(51.11%).It is randomly divided into training set(157 cases)and test set(68 cases)according to 7∶3.The support vector machines(SVM)learning algorithm was used to construct single sequence models and multi parameter MRI models for the first,fourth,and seventh phases of dynamic contrast-enhanced magnetic resonance imaging(CE1,CE4,CE7),T2 weighted imaging(T2WI)and diffusion weighted imaging(DWI).The fusion model was established by combining the data of five sequences and clinical characteristics.The performance of different models was evaluated by receiver operating characteristic(ROC)curve,calibration curve and decision curve analysis(DCA),and the model was interpreted and visualized using shap graphics.Results:The area under the curve(AUC)of CE1,CE4,CE7,T2WI and DWI sequences in the test set were 0.768,0.804,0.746,0.769 and 0.812,respectively.The AUC of DWI in the test set was the highest,followed by CE4;the AUC of the multi parameter MRI model in the test set was 0.840(95%confidence interval was 0.749 to 0.932),while the AUC of the fusion model in the test set was 0.866(95%confidence interval was 0.783 to 0.948),which was significantly different from CE1,CE4,CE7 and T2WI single-mode models(P<0.01).The results showed that the accuracy of the integrated model was the highest(77.94%);the sensitivity of the integrated model was the highest(90.00%);and the specificity of the integrated model and the CE4 sequence was the highest(both at 68.42%).Conclusions:The fusion model of multi parameter MRI combined with clinical features has higher accuracy,sensitivity and specificity,and better prediction performance than the single sequence model and multi-parametric MRI models,which can provide higher value for the differential diagnosis of non mass breast cancer and NLM.
作者
宋丽俊
薛志伟
田兄玲
贾毅
马依迪丽·尼加提
SONG Lijun;XUE Zhiwei;TIAN Xiongling;JIA Yi;MAYIDILI·Nijiati(Xinjiang Key Laboratory of Artificial Intelligence Assisted Imaging Diagnosis,Kashi 844000,China;Imaging Center of Xinjiang Medical University Affiliated Traditional Chinese Medicine Hospital,Urumqi 830000,China)
出处
《磁共振成像》
北大核心
2025年第8期73-79,94,共8页
Chinese Journal of Magnetic Resonance Imaging
基金
新疆人工智能影像辅助诊断重点实验室开放课题资助项目(项目编号:XJRGZN2024009)。
关键词
非哺乳期乳腺炎
非肿块型乳腺癌
多参数
磁共振成像
鉴别诊断
non lactation mastitis
non mass breast cancer
multiparameter
magnetic resonance imaging
differential diagnosis
