湖北医药学院学报2023，Vol.42Issue(6)：620-625,6.DOI:10.13819/j.issn.2096-708X.2023.06.008
基于临床及超声影像组学联合深度学习模型预测甲状腺结节良恶性
Prediction of Benign and Malignant Thyroid Nodules Based on a Combined Deep Learning Model of Clinical and Ultrasound Radiomics
摘要
Abstract
Objective To construct a joint clinical and ultrasound image-based radiomics combined with deep learning(DL)model and assess its value in predicting the malignant risk of thyroid nodules.Methods A retrospective analysis was conducted on 164 cases with surgically confirmed initial thyroid nodule ultrasound image data,which were divided into training and test sets in a 7 ∶ 3 ratio.Independent risk factors for malignant thyroid nodules were selected through t-tests and χ2 tests and used as clinical features for the risk model.Based on the ultrasound image,the best radiomics features were extracted and screened to construct the radiomics label score(Rad-score),the best DL features were extracted and screened to construct the DL label score(DL-score),and the radiomics-DL model was constructed.Based on clinical risk indicators,Rad-score and DL-score,we tried to use 9 algorithms to build a joint model.Draw the receiver operating char-acteristic(ROC)curve,calculate the area under the curve(AUC),evaluate the predictive efficacy of each model and the accuracy of benign and malignant thyroid nodules,and apply decision curve analysis(DCA)to compare the clinical bene-fits of different models.Results Age,gender,shape of the border,and echo category were identified as independent risk factors for malignant thyroid nodules.DCA showed that the XGBoost joint model had higher clinical benefits than other diag-nostic models.The GBoost joint model performed the best among the nine models,surpassing logistic regression,Naive-Bayes,Support Vector Machine(SVM),K-Nearest Neighbor(KNN),Light Gradient Boosting Machine(LightGBM),Gradient Boosting,AdaBoost,and Multi-Layer Perceptron(MLP)models,with a test set AUC of 0.93.Subsequently,a radiomics chart was established,facilitating accurate judgment of the benign and malignant nature of thyroid nodules.Con-clusion The joint clinical and ultrasound image-based radiomics combined with deep learning model can effectively predict the benign and malignant nature of thyroid nodules.关键词
甲状腺癌/影像组学/深度学习Key words
Thyroid cancer/Imagingomics/Deep learning引用本文复制引用
万义墨,宋鑫洋,郭建峰,郭红志,段鹏,廉凯..基于临床及超声影像组学联合深度学习模型预测甲状腺结节良恶性[J].湖北医药学院学报,2023,42(6):620-625,6.基金项目
湖北省自然科学基金创新发展基金联合基金(2022CFD010) （2022CFD010）
湖北省卫生健康委面上项目(WJ2023M162) （WJ2023M162）
襄阳市科技局重点项目(2022YL24A) （2022YL24A）
湖北医药学院研究生科技创新项目(YC2021027,YC2022049) （YC2021027,YC2022049）
