An imaging-based artificial intelligence model for non-invasive grading of hepatic venous pressure gradient in cirrhotic portal hypertension

Qian Yu,Yifei Huang,Xiaoguo Li,Michael Pavlides,Dengxiang Liu,Hongwu Luo,Huiguo Ding,Weimin An,Fuquan Liu,Changzeng Zuo,Chunqiang Lu,Tianyu Tang,Yuancheng Wang,Shan Huang,Chuan Liu,Tianlei Zheng,Ning Kang,Changchun Liu,Jitao Wang,Seray Akçalar,Emrecan Çelebioğlu,Evren Üstüner,Sadık Bilgiç,Qu Fang,Chi-Cheng Fu,Ruiping Zhang,Chengyan Wang,Jingwei Wei,Jie Tian,Necati Örmeci,Zeynep Ellik,Özgün Ömer Asiller,Shenghong Ju,Xiaolong Qi
DOI: https://doi.org/10.1016/j.xcrm.2022.100563
IF: 16.988
2022-03-01
Cell Reports Medicine
Abstract:The hepatic venous pressure gradient (HVPG) is the gold standard for cirrhotic portal hypertension (PHT), but it is invasive and specialized. Alternative non-invasive techniques are needed to assess the hepatic venous pressure gradient (HVPG). Here, we develop an auto-machine-learning CT radiomics HVPG quantitative model (aHVPG), and then we validate the model in internal and external test datasets by the area under the receiver operating characteristic curves (AUCs) for HVPG stages (≥10, ≥12, ≥16, and ≥20 mm Hg) and compare the model with imaging- and serum-based tools. The final aHVPG model achieves AUCs over 0.80 and outperforms other non-invasive tools for assessing HVPG. The model shows performance improvement in identifying the severity of PHT, which may help non-invasive HVPG primary prophylaxis when transjugular HVPG measurements are not available.
What problem does this paper attempt to address?