Prediction Of Portal Hypertension Grade Based On Quantitative Analysis Of Liver Vessel

Portal hypertension(PHT)is a group of clinical syndromes caused by abnormally high pressure in the portal vein and its branches.PHT can lead to high morbidity and mortality,causing a serious public health problem worldwide.Hepatic Venous Pressure Gradient(HVPG)is the gold standard for cirrhotic PHT,but it is invasive and expensive.This paper proposes a quantitative analysis method based on abdominal enhanced CT images for predicting the PHT level.The method is mainly divided into two steps:(1)The portal vein and hepatic vein in the abdominal enhanced CT images were segmented;(2)Vascular features related to the PHT level were extracted.Then the classification model was developed to achieve non-invasive assessment of PHT levels.Firstly,semantic segmentation of the portal vein and hepatic vein is performed.This paper proposes a novel 3D vessel segmentation network called CAU-Net.The anisotropic attention module utilizes the spatial anisotropy of vessel structures to extract vessel features from three directions and model the correlation between feature channels,enabling the learning of three-dimensional spatial information.In addition,CAU-Net adopts a mainauxiliary dual-branch model,where the b-Net performs semantic segmentation of vessels,and the a-Net learns the continuity features of vessel centerlines to constrain the vessel segmentation results of b-Net and ensure the integrity of vessel segmentation results.To validate the effectiveness of CAU-Net in the segmentation of portal veins and hepatic veins,comparative experiments were conducted on the public dataset 3D-ircadb-01.The experimental results showed that CAU-Net significantly improved the accuracy and completeness of vessel segmentation,with high accuracy and robustness in the portal vein and hepatic vein segmentation.Then,the classification model of portal hypertension is developed.This paper proposes an automatic classification method for portal hypertension based on vascular image features.First,the CAU-Net segmentation network model was used to segment the portal vein and hepatic vein from abdominal enhanced CT data of 331 patients with PHT.Then vascular radiomics features were extracted and selected.Finally,four different classifiers were trained to identify low and high PHT.The SVM model based on all liver veins showed excellent performance,with an area under the receiver operating characteristic curve,sensitivity,specificity,and accuracy of 0.928,0.896,0.875,and 0.899,respectively.The paper proposes a method for predicting the level of PHT using vascular image analysis.This method is fast,non-invasive,and highly interpretable,which makes it a useful tool for doctors in diagnosing PHT levels of their patients.Besides,this method can improve patient survival prognosis and play an important role in clinical practice.