Radiomics models for diagnosing microvascular invasion in hepatocellular carcinoma: which model is the best model?

• DOI: 10.1186/s40644-019-0249-x
• Publication Date: 2019-08-28T04:17:03Z
• Authors: Ming Ni; Xiaoming Zhou; Qian Lv; Zhiming Li; Yuanxiang Gao; Yongqi Tan; Jihua Liu; Fang Liu; Haiyang Yu; Linlin Jiao; Gang Wang
• Keywords: Lasso

To explore the feasibility of diagnosing microvascular invasion (MVI) with radiomics, to compare diagnostic performance different models established by each method, and determine best model based on radiomics.A retrospective analysis was conducted 206 cases hepatocellular carcinoma (HCC) confirmed through surgery pathology in our hospital from June 2015 September 2018. Among samples, 88 were MVI-positive, 118 MVI-negative. The radiomics process included tumor segmentation, feature extraction, data preprocessing, dimensionality reduction, modeling evaluation.A total 1044 sets texture parameters extracted, 21 methods used for analysis. All research could be diagnose MVI. Of all methods, LASSO+GBDT method had highest accuracy, LASSO+RF sensitivity, LASSO+BPNet specificity, AUC. Through Z-tests AUCs, LASSO+GBDT, LASSO+K-NN, LASSO+RF, PCA + DT, RF Z-values greater than 1.96 (p<0.05). DCA results showed that LASSO GBDT better other when threshold probability 0.22.Radiomics can preoperative, noninvasive diagnosis MVI, but reduction will affect final model. optimal greatest value