A Radiomics Model Based on Gd-EOB-DTPA-Enhanced MRI for the Prediction of Microvascular Invasion in Solitary Hepatocellular Carcinoma ≤ 5 cm

Abstract

AimThe aim of this study is to establish and validate a radiomics-based model using preoperative Gd-EOB-DTPA-enhanced MRI to predict microvascular invasion (MVI) in patients with hepatocellular carcinoma ≤ 5 cm.MethodsClinicopathologic and MRI data of 178 patients with solitary hepatocellular carcinoma (HCC) (≤5 cm) were retrospectively collected from a single medical center between May 2017 and November 2020. Patients were randomly assigned into training and test subsets by a ratio of 7:3. Imaging features were extracted from the segmented tumor volume of interest with 1-cm expansion on arterial phase (AP) and hepatobiliary phase (HBP) images. Different models based on the significant clinical risk factors and/or selected imaging features were established and the predictive performance of the models was evaluated.ResultsThree radiomics models, the AP_model, the HBP_model, and the AP+HBP_model, were constructed for MVI prediction. Among them, the AP+HBP_model outperformed the other two. When it was combined with a clinical model, consisting of tumor size and alpha-fetoprotein (AFP), the combined model (AP+HBP+Clin_model) showed an area under the curve of 0.90 and 0.70 in the training and test subsets, respectively. Its sensitivity and specificity were 0.91 and 0.76 in the training subset and 0.60 and 0.79 in the test subset, respectively. The calibration curve illustrated that the combined model possessed a good agreement between the predicted and the actual probabilities.ConclusionsThe radiomics-based model combining imaging features from the arterial and hepatobiliary phases of Gd-EOB-DTPA-enhanced MRI and clinical risk factors provides an effective and reliable tool for the preoperative prediction of MVI in patients with HCC ≤ 5 cm.

Update on the Use of Artificial Intelligence in Hepatobiliary MR Imaging

“… 50 Qu et al evaluated three radiomics models based on arterial phase images, hepatobiliary phase images, and all images of a gadoxetic-acid-enhanced MRI. 51 Results showed that the radiomics model using all images outperformed the other two. The combined model based on clinical and radiomics analysis showed an AUC of 0.90 and 0.70 in the training and test subsets, respectively.…”

Current state of radiomics in hepatobiliary and pancreatic malignancies

“…Moreover, microinvasion, encompassing infiltration of portal vein, hepatic vein, or bile duct, is considered a marker of poor prognosis after hepatic resection and transplantation [95] . Several studies have also demonstrated promising results of select CT and U/S-based radiomics signatures in accurately predicting the presence of microvascular tumor invasion [96][97][98][99] . For ICC, in a study of 203 cases, radiomics-based models demonstrated the ability to predict futile resections with greater AUC (0.804) than traditional clinical risk factor-based models (AUC: 0.590) [100] .…”