Selection of Optimal Pulse Sequences and Enhancement Phases of MRI Study for Radiomics Analysis in the Diagnosis of Early Hepatocellular Carcinoma

Purpose: To compare the importance of different MRI sequences and enhancement phases in creation of a diagnostic radiomics model in MRI diagnostics of early hepatocellular carcinoma (HCC). Material and methods: Data from 72 patients with 93 masses who underwent Gadoxetic acid-enhanced MRI scans was retrospectively analyzed, a comparative assessment of the indicators of four sequences and enhancement phases of MRI studies was performed. Results: As a result of the study, machine- learning radiomics based models on various MRI sequences and enhancement phases with high discriminatory capabilities were created. The area under the ROC curve (Area Under Curve, AUC) ranged from 0.58 to 0.94 in various models; the best results were performed in Random Forest model based on MRI-hepatobiliary enhancement phase — AUC 0.949684, the combination of different enhancement sequences — AUC 0.914342. Conclusion: The hepatobiliary phase of MRI study independently, as well as the combination of four enhancement phases and sequences of MRI study, have the greatest discriminatory capabilities for creating machine- learning radiomics based models on enhanced MR images in diagnostics of early HCC.