| After receiving liver resection or interventional therapy,development of subsequent treatment for the patients with hepatocellular carcinoma(HCC)will be limited by occurrence of macrovascular invasion(MaVI).Therefore,MaVI is an important indicator of poor prognosis.Clinically,several therapies,such as targeted and immune drugs,can be used to treat MaVI.However,the safety and efficacy of these therapies are highly dependent on early diagnosis and treatments of MaVI.Therefore,early prediction of MaVI is essential to improve the survival rate of HCC.Diagnosis and treatment of HCC is independent on the pathological results.As such,it is reasonable to explore deep learning methods based on medical images for early prediction of MaVI to avoid the extra bleeding risk caused by needle biopsy.To achieve this goal,two key points need to be solved.First,accurately automatic segmentation of HCC is required,which can be used to extract HCC as the region of interests(ROIs)for early prediction of MaVI.Although manual segmentation of HCC is accurate,it is time-consuming and labor-intensive,which is impractical for early prediction of MaVI.Second,an early prediction method for MaVI is needed.Based on the results of HCC segmentation,it is crucial to fully extract information with deep learning methods for early prediction of MaVI.Automatic segmentation of HCC is a challenging task since the size,shape,location,and gray level ranged a lot in HCC.Moreover,the unclear boundaries of HCC increase the difficulty when segmenting HCC.To overcome these challenges,an HCC segmentation method,namely Swin Transformer UNet(STUNet),was proposed in this study.First,a U-shaped network was used to include multi-scale information,which can be used to explore information in HCC with varied sizes,shapes,and gray levels.Second,Swin Transformer blocks were applied in the skip connection layers to obtain global information,which can be used to learn information of HCC with different locations.Finally,a robust segmentation loss was introduced to solve the problem of unclear boundaries in HCC.The proposed STUNet is comparable to some state-of-theart segmentation algorithms,which demonstrates the effectiveness of the proposed STUNet.It is difficult to predict MaVI since inter-class similarity and intra-class variation of HCC in CT images.Moreover,existing methods failed to incorporate clinical priori knowledge associated with HCC,leading to poor prediction accuracy.In this study,we proposed a prior knowledge-aware fusion network(PKAFnet)to accurately achieve MaVI prediction in CT images.First,a perception module was presented to extract features related to tumor marginal heterogeneity in the graph domain,which contributed to rotation invariance and captured intensity variations of tumor margin.Second,a tumor segmentation network was built to obtain global information of a 3D tumor image and information associated with tumor internal heterogeneity in the image domain.Finally,multi-domain features associated with the tumor margin and tumor region were combined by using a multi-domain attentional feature fusion module.Thus,by incorporating MaVI-related prior knowledge,our PKAFnet can alleviate overfitting,which can improve the discriminative ability.Moreover,the performance of our PKAFnet is better than some classification methods,which illustrated the effectiveness of PKAFnet.STUNet and PKAFnet were assessed on multicenter datasets,which can further demonstrate the good generalized ability of the proposed method.Moreover,the practicability and accuracy issues for early prediction of MaVI can be addressed by using these two methods.Therefore,the proposed method showed great application potential for MaVI prediction. |
