Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related deaths worldwide. Early detection, diagnosis, and intervention can significantly reduce mortality rates. HCC development is driven by a series of complex processes, often starting from hepatitis and cirrhosis. During this progression, subtle changes in cellular gene expression, major shifts in molecular signaling pathways, and significant changes in tissue structure and organ function collectively accelerate HCC formation and progression, leading to changes in liver imaging. Researchers have used radiomics and deep learning technologies to timely and accurately identify and intervene in the formation stage of HCC, named the Transition Stage, potentially revolutionizing early prevention and intervention strategies for HCC.To meet this need, the team led by De.Rong Fan and Jinlin Hou from the Liver Disease Center of Nanfang Hospital of Southern Medical University, leveraging a nationwide multi-center long-term follow-up cohort of high-risk liver cancer populations and corresponding multidimensional specimen banks, successfully constructed and validated an advanced tool—the ALARM model. This model, using multi-phase enhanced computed tomography (CT) and aMAP scores, can predict HCC occurrence 3 to 12 months in advance, providing an effective tool for identifying the Transition Stage of HCC. This significant achievement was recently published in eClinical Medicine, a subsidiary of The Lancet.
Research Methods
The study’s participants were primarily drawn from the national multi-center long-term cirrhosis follow-up cohort (PreCar cohort, NCT03588442). Researchers collected three-phase enhanced CT (CECT) images and common clinical indicator data from HCC patients before diagnosis or from non-HCC patients 3 to 12 months before their last follow-up. Utilizing radiomics and deep learning technologies, they constructed an early warning model in the discovery cohort at Nanfang Hospital, and subsequently validated the model in an internal validation cohort at Nanfang Hospital and in external validation cohorts from 10 other centers.
Research Results
A total of 1,858 cirrhosis patients from 11 centers nationwide were included in the study. The discovery cohort consisted of 924 patients from Nanfang Hospital, the internal validation cohort included 231 patients, and the external validation cohort covered 703 patients from 10 other centers.
By integrating deep learning-based radiomics technology with aMAP risk scores, the team constructed and validated a novel and precise HCC early warning model—the ALARM model. The model achieved AUCs of 0.929, 0.902, and 0.918 in the discovery, internal validation, and external validation cohorts, respectively.
Using optimal thresholds of 0.21 and 0.65, patients were divided into three risk groups: the high-risk group (221 cases, 11.9%) had an HCC incidence rate of 24.3% within 3 to 12 months, the medium-risk group (433 cases, 23.3%) had an HCC incidence rate of 6.4%, and the high and medium-risk groups together covered 94.4% of HCC cases. In contrast, the low-risk group comprised 1,204 patients (64.8%), with a short-term HCC incidence rate of only 0.42%.
Research Conclusion
This study successfully constructed an effective tool to accurately predict the short-term risk of HCC occurrence in cirrhosis patients using the new ALARM model based on deep learning radiomics and aMAP. This model provides robust support for identifying the Transition Stage of HCC. Integrating this tool into clinical treatment strategies in the future will significantly enhance the precision of HCC diagnosis and strongly promote the development and implementation of personalized preventive treatment plans.
