Prediction of Microvascular Invasion of Hepatocellular Carcinoma Based on Contrast-Enhanced MR and 3D Convolutional Neural Networks

Background and purpose: Preoperative prediction of MVI of hepatocellular carcinoma (HCC) can be significant and helpful for deciding treatment strategy and patient management before hepatic resection or liver transplantation. In the present study, we propose to demonstrate that a deep learning framework based on 3D Convolutional Neural Network (CNN) can extract essential information from Contrast-enhanced MR images for better microvascular invasion (MVI) prediction in patients with HCC. Materials and methods:114consecutive patients with 117 histologically proven HCCs after surgical resection from October 2012 to October 2018 were included for this retrospective study. MR sequences including 3.0T/LAVA(Liver acquisition with volume acceleration) and 3.0T/e-THRIVE (Enhanced T1 high resolution isotropic volume excitation) were used for the image acquisition of each patient. First, numerous 3D patches were separately extracted from the region of each lesion for data augmentation. Then, 3D CNN is adopted to extract high-level deep features of HCC from Contrast-enhanced MR separately. Furthermore, a deeply supervised loss function is designed to improve the prediction performance by taking advantage of the discriminative features from multiple phases of Contrast-enhanced MR. The dataset was divided into two parts, in which 77 lesions were randomly chosen as the training set, and the remaining 40 lesions were used for independent testing. Receiver operating characteristic curve (ROC) and area under the curve (AUC) were used to assess the performance of MVI prediction. The output probability of the model was assessed by the independent student's t-test or Mann-Whitney U test. Results: The mean AUC values of MVI prediction of HCC using 3D CNN in the pre-contrast phase, arterial phase and portal vein phase were 0.793(p=0.001), 0.855 (p=0.000) and 0.817 (p=0.000), respectively. Simple concatenation of deep features using 3D CNN derived from all the three phases improved the performance with the AUC value of 0.906 (p=0.000). Finally, the proposed deeply supervised deep learning model yielded the best performance with the AUC value of 0.926 (p=0.000). Conclusion: A deep learning framework based on 3D CNN and deeply supervised net with CE-MR images could be effective for MVI prediction.