融合临床特征与超声影像组学的机器学习模型用于原发性与转移性肝癌鉴别诊断研究OA

Objective:To investigate application of machine learning-based ultrasound radiomics in preoperative classification for primary and metastatic liver cancer.Methods:A retrospective analysis was conducted on ultrasound images from 70 patients with primary liver cancer and 38 patients with metastatic liver cancer,who admitted at The 910th Hospital of the Joint Logistics Support Force,Chinese PLA between March 2020 and November 2024.The patients were randomly divided into training and validation sets as a ratio of 7:3.Firstly,ultrasound radiomic features were extracted on the Pyradiomics platform.Subsequently,the Least Absolute Shrinkage and Selection Operator(LASSO)was used to conduct dimensionality reduction on features,so as to screen the most relevant features.Finally,four classifier algorithms,including logistic regression(LR),support vector machine(SVM),extreme gradient boosting(XGBoost),and k-nearest neighbor(KNN),were applied to train the model.The efficiencies of the four classifiers were assessed by using area under curve(AUC)of the receiver operating characteristic(AUC)curve,accuracy,sensitivity,and specificity.Results:There were significant differences in clinical indicators,including Aspartate Aminotransferase(AST),Alkaline Phosphatase(ALP),Alpha-Fetoprotein(AFP),and Carcinoembryonic Antigen(CEA),between primary and metastatic liver cancer.The study employed the LASSO method to obtain eight optimal features by screening,and constructed a machine learning model integrating clinical features and ultrasound radiomics.The results of ROC analysis indicated that this model exhibited favorably diagnostic efficiency in differentiating primary liver cancer and metastatic liver cancer.Among of these models,the performance of LR model was slightly better than that of others,which achieved an AUC value of 0.975 in the training set,and the accuracy,sensitivity and specificity of that were respectively 95.65%,96.99%and 95.01%.Conclusion:The machine learning model integrating clinical features and ultrasound radiomics can more accurately differentiate primary and metastatic liver cancer,which has favorable application value for clinical diagnosis.