A Proposed AI/ML-Based Detection Framework for Early Detection of Hepatocellular Carcinoma (HCC) and Cirrhosis
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Hepatocellular carcinoma (HCC) is the leading cause of liver cancer-related mortality, with cirrhosis being a major risk factor. Current blood-based early detection tests rely on targeted sequencing assays, limiting their adaptability to additional conditions such as cirrhosis. Here, we propose an AI/ML-driven framework leveraging low- to moderate-depth whole-genome bisulfite sequencing (WGBS) to detect HCC and extend this methodology to cirrhosis. Our approach, trained on PRJCA001372 (low-depth liquid biopsy WGBS), was validated independently against PRJNA984754 (moderate-depth matched liver tissue WGBS), achieving an overall accuracy of 92% in blind validation. Key methylation biomarkers, particularly hypomethylation around HBV integration sites (MethylHBV5K) and SENP5 gene hypomethylation, were found to effectively differentiate HCC from cirrhosis. Our validation confirms that these biomarkers generalize across independent datasets, demonstrating robustness beyond the original PRJCA001372 study. The alignment of tissue-based and liquid biopsy WGBS methylation patterns, even without batch effect correction, underscores the robustness of this approach and strengthens its potential for clinical translation as a non-invasive diagnostic tool. Additionally, we propose extending this framework to cirrhosis detection using the same WGBS-based methodology. Peroxisome Proliferator-Activated Receptor Gamma (PPARγ 4 ) hypermethylation in plasma cfDNA has been associated with liver fibrosis severity, making it a promising biomarker for non-invasive stratification of cirrhosis. PRJCA001372 will serve as a training dataset for HBV-related cirrhosis, while PMC5031527 (NAFLD fibrosis dataset) will be explored for training a model on non-HBV cirrhosis detection. PRJNA984754, containing a mix of HCC and cirrhosis patients, will be utilized for independent validation of the cirrhosis model. Given the absence of a standard-of-care blood-based test for cirrhosis, this study highlights the potential of providing a scalable solution for both HCC and cirrhosis detection.