From November 15–19, 2024, the American Association for the Study of Liver Diseases (AASLD) convened in San Diego, USA. A study led by Dr. Huiguo Ding’s team from the Beijing You’an Hospital, Capital Medical University, was recognized as a Poster of Distinction. Conducted primarily by Dr. Yanna Liu and PhD candidate Yanglan He, this research developed a machine learning-based CT imaging model for distinguishing portal-sinusoidal vascular disease (PSVD) with portal hypertension (PH) from cirrhotic PH. Dr. Liu was also honored with the 2024 International Early Career Investigator Award.
Cirrhosis is the primary cause of portal hypertension (PH), but 15-20% of PH cases stem from non-cirrhotic conditions, notably idiopathic non-cirrhotic portal hypertension (INCPH). INCPH is now categorized under the broader term of portal sinusoidal vascular disease (PSVD), which frequently presents with PH. A significant challenge is that up to 80% of PSVD patients are initially misdiagnosed with cirrhotic PH, underscoring the urgency for improved diagnostic methods.
This study aimed to assess the efficacy of a CT image-based machine learning model in differentiating PSVD with PH from cirrhotic PH. Patients diagnosed with PSVD and PH, based on European VALDIG and Baveno VII criteria, were enrolled between January 2012 and December 2021. A control group with cirrhotic PH was matched at a 1:2 ratio. All participants underwent contrast-enhanced abdominal CT scans.
The study cohort comprised 291 patients: 97 with PSVD and 194 with cirrhotic PH. Patients were randomly split into training (80%) and testing (20%) cohorts. Utilizing radiomics and machine learning techniques, researchers developed a non-invasive diagnostic model known as the PSVD Score. This score integrated 31 features extracted from CT images, with 15 from the liver and 16 from the spleen.
In the training cohort of 232 patients, the PSVD Score achieved an area under the receiver operating characteristic curve (AUC) of 0.913, indicating excellent diagnostic performance. The model’s validity was further confirmed in the testing cohort of 59 patients, with an AUC of 0.819.
The CT-based PSVD Score demonstrated high accuracy in distinguishing PSVD with PH from cirrhotic PH. This study marks the first to report the feasibility of using a CT-based radiomics model for this differentiation in PH patients. To further validate the model, larger, multi-center studies are necessary. Additionally, exploring the model’s potential to predict clinical outcomes in PSVD patients, such as variceal bleeding or portal vein thrombosis, is worth pursuing in future research.
About the Authors
Dr. Huiguo Ding
• Director, Department of Hepatology and Gastroenterology, Beijing You’an Hospital, Capital Medical University
• Renowned expert in vascular liver diseases and portal hypertension
• Published extensively in hepatology, with pioneering contributions to the diagnosis and treatment of PSVD
Dr. Yanna Liu
• Physician, Beijing You’an Hospital
• Recipient of the 2024 International Early Career Investigator Award
• Lead author of the study, focusing on non-invasive diagnostic tools for vascular liver diseases
Yanglan He
• PhD candidate, Beijing You’an Hospital
• Key contributor to the study, specializing in radiomics and machine learning applications in liver diseases
