Pancreatic cystic neoplasm classification model integrating deep learning and radiomics features
Objective To establish a comprehensive classification model integrating deep learning and radiomics features to enhance the accuracy and consistency for identifying pancreatic cystic neoplasms.Methods Firstly,the radiomics technique and ResNet50 convolutional neural network were used to extract the features of radiomics and deep learing for pancreatic cystic neoplasms,respectively,and the key features were obtained by screening the features with the least absolute shrinkage and selection operator;secondly,three classification models were constructed by integrating only key features,deep learning features or radiomics and deep learning features,respectively;finally,logistic regression,random forest,adaptive boosting and support vector machine(SVM)classifiers were used to compare the above feature models in order to verify their efficacies for categorizing pancreatic cystic neoplasms.Results The comprehensive model integrating deep learning and radiomics features performed the best on the SVM classifier,which gained advantages over the other two models with the accuracy,recall,precision,AUC value and F1 value being 89.34％,92.13％,75.34％,0.90 and 0.83,respectively.Conclusion The comprehensive classification model integrating deep learning and radiomics features behaves well in mining relationships between various types of features and classifying pancreatic cystic neoplasms,and facilitates further precise diagnosis and treatment.[Chinese Medical Equipment Journal,2025,46(1):7-12]
magnetic resonance imagingdeep learningradiomicspancreatic cystic neoplasmneoplasm classification
万方数据
维普
