Background: Hepatocellular carcinoma (HCC) with hilar bile duct tumor thrombus (HBDTT) often mimic hilar cholangiocarcinoma (hilar CC). The purpose of this study is to analyze the Computed Tomography (CT) characteristics of HCC with HBDTT and to identify imaging features to aid its differentiation from hilar CC on enhanced CT.Methods: We retrospectively identified 58 cases with pathologically proved HCC with HBDTT between 2011 and 2018. Seventy-seven cases of pathologically proven hilar CCs were selected during the same period. The clinical features and CT findings of the two groups were reviewed and compared.Results: HCC with HBDTTs are more commonly found in men (87.9% vs 63.6%,p = 0.001) with lower age of onset (49.84 vs 58.61 years;p < 0.001) in comparison to hilar CCs. Positive correlation were identified between HCC with HBDTTs and chronic HBV infection (72.4% vs 11.7%;p < 0.001), increased serum AFP (67.2% vs 1.3%;p < 0.001), CA19–9 level (58.6% vs 85.7%;p < 0.001) and CEA level (3.4% vs 29.9%;p = 0.001), parenchymal lesion with intraductal lesion (100% vs 18.2%;p < 0.001), washout during the portal venous phase (84.5% vs 6.5%;p < 0.001), thickened bile duct wall (8.6% vs 93.5%;p < 0.001), intrahepatic vascular embolus (44.8% vs 7.8%;p < 0.001), splenomegaly (34.5% vs 2.6%,p < 0.001). A scoring system consisting of the five parameters obtained from characteristics mentioned above was trialed. The sensitivity and specificity for diagnosing HCC with HBDTT were 96.39, 100 and 92.5% respectively when the total score was 2 or more.Conclusions: HCC with HBDTTs are often distinguishable from hilar CCs based on washout during portal venous phase without thickened bile duct wall. HBV infection and serum AFP level facilitate the differentiation.

Preoperative prediction of microvascular invasion (MVI) is essential for management decision in hepatocellular carcinoma (HCC). Deep learning-based prediction models of MVI are numerous but lack clinical interpretation due to their “black-box” nature. Consequently, we aimed to use an attention-guided feature fusion network, including intra- and inter-attention modules, to solve this problem. This retrospective study recruited 210 HCC patients who underwent gadoxetate-enhanced MRI examination before surgery. The MRIs on pre-contrast, arterial, portal, and hepatobiliary phases (hepatobiliary phase: HBP) were used to develop single-phase and multi-phase models. Attention weights provided by attention modules were used to obtain visual explanations of predictive decisions. The four-phase fusion model achieved the highest area under the curve (AUC) of 0.92 (95% CI: 0.84–1.00), and the other models proposed AUCs of 0.75–0.91. Attention heatmaps of collaborative-attention layers revealed that tumor margins in all phases and peritumoral areas in the arterial phase and HBP were salient regions for MVI prediction. Heatmaps of weights in fully connected layers showed that the HBP contributed the most to MVI prediction. Our study firstly implemented self-attention and collaborative-attention to reveal the relationship between deep features and MVI, improving the clinical interpretation of prediction models. The clinical interpretability offers radiologists and clinicians more confidence to apply deep learning models in clinical practice, helping HCC patients formulate personalized therapies.