Retrospective Study to Develop a Model Using Artificial Intelligence to Predict Hepatocellular Carcinoma Risk in Cirrhosis

National Institute Of Diabetes And Digestive And Kidney Diseases

Investigators

Linked publications, trials & patents

Abstract

The protocol was approved by the Institutional Review Board in September 2022. We have recruited approximately 429 patients with advanced fibrosis/cirrhosis into our NIH clinical cohort, of whom 52 patients developed HCC (hepatocellular carcinoma). We have approximately 226 Georgetown patients who underwent liver transplant for indications of HCC and cirrhosis. We also have 200 cirrhosis patients from University of Alabama at Birmingham, of whom all 200 developed HCC as well. We have since built a clinical dataset of 429 NIH subjects with extracted demographic, clinical, and radiologic data. We have since presented an abstract at DDW 2024 showing abdominal varices to be an independent predictor of HCC development and building a time-vary clinical model based on labs, imaging, and clinical data to predict HCC development. We are currently working on manuscript. We have since built a model using Artificial Intelligence or AI to predict the presence of liver lesions in cirrhosis patients using MRI images. The model is currently able to predict the presence of a LI-RADS 3, 4, or 5 lesion (LI-RADS 5 is diagnostic for HCC) with good sensitivity (89%) and accuracy (59%). We have submitted this abstract since to American Roentgen Ray Society annual meeting. We have also explored imaging characteristics and its ability to predict liver cancer recurrence after liver transplant. We will present this abstract in AASLD (American Association for the Study of Liver Disease) 2025 meeting. Ultimately, we want to develop the best model to predict HCC development. Thus, we are evaluating various experimental biomarkers in our cohort of patients and have collaborated with UT Southwestern to evaluate PLSec (prognostic liver secretome signature) to predict HCC development and will present that abstract at AASLD 2025.