节点文献

磁共振双参数影像组学机器学习在显著性前列腺癌诊断中的应用

Application of Magnetic Resonance Biparametric Radiomics Machine Learning in the Diagnosis of Significant Prostate Cancer

  • 推荐 CAJ下载
  • PDF下载
  • 不支持迅雷等下载工具,请取消加速工具后下载。

【作者】 李静黄宝生吴桂秀姚家喜宋泽杨晶晶王泽华

【Author】 LI Jing;HUANG Bao-sheng;WU Gui-xiu;YAO Jia-xi;SONG Ze;YANG Jing-jing;WANG Ze-hua;Imaging Research Institute of Zhangye People’s Hospital Affiliated to Hexi University;

【通讯作者】 李静;

【机构】 河西学院附属张掖人民医院影像研究所

【摘要】 目的 探究基于磁共振双参数纹理分析的机器学习模型在临床显著性前列腺癌(clinically sighificant prostate cancer,csPCa)中的诊断价值。方法 回顾性纳入2018年1月-2023年1月期间进行术前磁共振检查并通过穿刺活检经病理证实为前列腺癌的患者222例,其中临床显著性癌(Gleason≥7) 117例,非临床显著性癌(Gleason<7)105例。所有患者均采用ITK-SNAP软件勾画病灶的全部层面为感兴趣区(region of interest,ROI),通过影像组学软件FeAture Explorer (FAE)(V.0.54)提取ROI内504个影像组学特征。222例患者随机按照7:3比例分成训练组和测试组。影像组学特征采用线性判别分类器(linear discriminant analysis,LDA)、随机森林(random forest,RF)、罗杰氏回归(Logistic regression,LR)、支持向量机(support vector machine, SVM)等不同方法 对模型进行筛选。根据模型在测试集上的曲线下面积(AUC)、敏感性、特异性、阳性预测值(PPV)、阴性预测值(NPV)、95%置信区间等来选择一个最优模型。结果 基于dwi_original_firstorder_Variance、dwi_original_gIcm_ClusterProminence、adc_original_firstorder_Mean、adc_original_firstorder_Median4个特征的线性判别分类器LDA模型在验证数据集上可以得到最高的曲线下面积(AUC),AUC和精度分别达到0.764和0.769,模型的AUC和精度对测试数据集的影响分别达到了0.950和0.909。结论磁共振双参数影像组学机器学习模型在诊断临床显著性前列腺癌中的准确率更高,采用LDA方法 的机器学习模型与其他模型相比具有更高的诊断效能。

【Abstract】 Objective To explore the diagnostic value of machine learning model based on magnetic resonance two-parameter textu re analysis in clinically significant prostate cancer.Methods A retrospective inclusion of 222 patients with pathologically confirmed prostate cancer by preoperative magnetic resonance examination and needle biopsy between January 2018 and January 2023 was included.Among them,there were 117 cases of clinically significant carcinoma(Gleason≥7) and 105 cases of non-clinically significant carcinoma(Gleason<7).All patients were treated with ITK-SNAP software to delineate all levels of the lesion as region of interest(ROI),and 504 radiomics features in the ROI were extracted by the radiomics software FAE(V.0.54).The 222 patients were randomly divided into training group and test group according to the ratio of 7:3.The radiomics features were screened by different methods such as linear discriminant analysis(LDA),random forest(RF),Logistic regression(LR),and support vector machine(SVM).Select an optimal model based on the model’s AUC,sensitivity,specificity,PPV,NPV,confidence interval,etc.on the test set.Results The linear discriminant classifier LDA model based on dwi_original_firstorder Variance,dwi original_glcm_ClusterProminence,adc_original_firstorder_Mean,and adc_original_firstorder Median4 features can obtain the highest AUC on the verification dataset.AUC and accuracy reach 0.764 and 0.769,respectively,The influence of AUC and accuracy on the test data set reached 0.950 and 0.909,respectively.Conclusion The magnetic resonance biparametric radiomics machine learning model has higher accuracy in diagnosing clinically significant prostate cancer,and the machine learning model using LDA method has higher diagnostic performance than other models.

【基金】 甘肃省教育厅创新基金项目(2021B-258)(2024A-156)
  • 【文献出处】 中国CT和MRI杂志 ,Chinese Journal of CT and MRI , 编辑部邮箱 ,2024年09期
  • 【分类号】R445.2;R737.25
  • 【下载频次】13
节点文献中: 

本文链接的文献网络图示:

本文的引文网络