基于CT双期增强影像组学预测甲状腺乳头状癌淋巴结转移

目的:探讨基于CT双期增强影像组学模型对甲状腺乳头状癌(PTC)淋巴结转移的预测价值.方法:回顾性分析经手术病理证实的80例PTC患者的病例资料,共搜集173个淋巴结,其中转移性淋巴结89个、未转移性淋巴结84个.患者术前均行CT平扫和双期增强扫描.采用达尔文科研平台,分别在动脉期和静脉期CT图像上于淋巴结内勾画ROI...

Prediction of lymph node metastasis in patients with papillary thyroid carcinoma:a radiomics methodbased on dual-phase contrast enhanced CT

Objective:The purposeof this study was to explore the predictive value of a radiomics model based on dual-phase contrast-enhanced CT(CECT)in lymph node metastasis of thyroid papillary carcinoma(PTC).Methods:80 patients with PTC confirmed by pathology were enrolled in our study retrospectively.173 lymph nodes were collected,including 89 metastatic lymph nodes(LMN)and 84 non-metastatic lymph nodes(ULMN).All patients underwent dual-phase CECT before surgery.ROI delineation and texture feature extraction were performed on the CECT images on arterial phase(AP)and venous phase(VP)using Darwin research platform,respectively.After pretreatment of minimum maximum normalization,optimal feature screening,model selection and iterative screening,the optimal texture parameters in AP and VP were obtained.70%patients were divided into training group and 30%into validation group,and the diagnostic model based on the optimal texture parameters of AP and VP were built using support vector machine(SVM)and cross-validation.The diagnostic performance of the models were evaluated by ROC analysis,and the sensitivity,specificity and accuracy were calculated.Results:There were six significant different texture features in the AP and 5 significant different texture features in the VP between LMN group and ULMN group respectively.The image cha-racteristics in AP were coarseness,dependence entropy(DE),short run low gray level emphasis(SRLGLE),run length non-uniformity(RLN),low gray level emphasis(LGLE)and size zone non-unifor-mity normalized(SZNN).The image characteristics in VP were coarseness,SALGLE,long run high gray level emphasis(LRHGLE),RLN and large dependence emphasis(LDE).In the training and validation groups,the areas under the ROC of the AP model and VP model were 0.903,0.915,0.895 and 0.850.In the validation group,the diagnostic accuracy of the AP model was 75.47%(40/53),the sensitivity was 88.00%,and the specificity was 80.77%;and the diagnostic accuracy of the VP model was 71.69%(38/53),the sensitivity was 80.77%,and the specificity was 81.48%.Conclusion:DCE-CT radiomics and machine learning are of good diagnostic value for prediction of lymph node metastasis in patients with thyroid papillary carcinoma,and that the diagnostic accuracy of AP texture features is relatively higher.