The Prediction Of Deep Learning Based On Contrast-enhanced Ultrasound In The Classifi Cation Of Benign And Malignant Cervical Lymphadenopathy

Objective Lymph node enlargement,a disease of the lymph nodes,has various etiologies.It may be benign or malignant.We aimed to evaluate the efficacy of deep learning and radiomics combined with seven models respectively based on continuous contrast-enhanced ultrasound(CEUS)videos for the classification of benign and malignant cervical lymphadenopathy.Further,we compared the performance of the two models with that of radiologists’ diagnoses.We also identified the best prediction model by comparing the prediction performance metrics of the two types of modelsMethods We collected CEUS videos of 192 benign and 70 malignant cases of cervical lymph node enlargement from department of Ultrasonography,Affiliated Hangzhou Chest Hospital.Radiomics combined with XGBoost,Kneighbors,RandomForest,GradientBoosting,SVM,AdaBoost,and LogisticRegression models respectively as well as deep learning model were used to classify benign and malignant cervical lymphadenopathy based on CEUS videos.Deep learning uses a pretrained 3D residual network(ResNet)model utilizing 34 and 50 layers.The classification results of the two types of models were compared with the diagnoses made by two radiologists who manually segmented lesions.An area under curve(AUC)was used to evaluate the diagnostic performance of the models.Results Among the deep learning models,the 3D-ResNet-50 model showed better efficacy in the test cohort(with an AUC of 0.899)than the 3D-ResNet-34 model.Radiomics combined with XGBoost,Kneighbors,RandomForest,GradientBoosting,SVM,AdaBoost,and LogisticRegression models respectively showed AUCs of 0.833,0.774,0.752,0.851,0.804,0.831,and 0.863,respectively,in the test cohort.Among these,radiomics combined with LogisticRegression model had the highest AUC of 0.863.The AUCs of the two radiologists were 0.714 and 0.747.Conclusion Among all models in this study,the 3D-ResNet-50 model was superior to radiomics combined with seven models respectively and radiologists’ diagnoses in classifying benign and malignant cervical lymphadenopathy using CEUS videos.