Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Sep;45(9):1188-1193.
doi: 10.1002/cac2.70043. Epub 2025 Jun 26.

A serum exosomal microRNA-based artificial intelligence diagnostic model for highly accurate detection of hepatocellular carcinoma

Jin-Seong Hwang  1   2 , Sugi Lee  1 , Gyeonghwa Kim  3 , Hoibin Jeong  4 , Kiyoon Kwon  1 , Eunsun Jung  1 , Yuna Roh  1   2 , Taesang Son  1   2 , Hana Lee  1 , Moo-Seung Lee  1   2 , Kyoung-Jin Oh  1   2 , Hye Won Lee  5 , Yu Rim Lee  6 , Soo Young Park  6 , Won Young Tak  6 , Hyun Seung Ban  1   2 , Hyun-Soo Cho  1   2   7 , Mi-Young Son  1   2   7 , Jang-Seong Kim  1   2 , Keun Hur  3   8 , Dae-Soo Kim  1   2 , Tae-Su Han  1   2   7
Affiliations

A serum exosomal microRNA-based artificial intelligence diagnostic model for highly accurate detection of hepatocellular carcinoma

Jin-Seong Hwang et al. Cancer Commun (Lond). 2025 Sep.
No abstract available

PubMed Disclaimer

Conflict of interest statement

The authors declare no competing interests.

Figures

FIGURE 1
FIGURE 1
Diagnostic performance of the AI‐based multi‐marker model utilizing serum exo‐miRNAs for HCC detection. (A‐C) Scheme of the development of mouse models of liver diseases using combinations of diet and CCl4 treatment. After 40 weeks, all mice were sacrificed and analyzed for liver morphology and pathology using H&E and Sirius red staining (A). The WD + oil group exhibited ballooning and immune responses with moderate fibrosis and increased liver weight. The ND + CCl4 group exhibited severe fibrosis and a cirrhotic phenotype. The WD + CCl4 group exhibited a cirrhotic phenotype, with tumors observed. White arrows indicate the tumor nodules. Scale bars indicate 100 µm. Average of liver weight (B) and tumor number (C) for each group. * P < 0.05, ** P < 0.01, *** P < 0.001. (D) Characterization of exosomes from mouse and human serum samples using nanoparticle tracking and transmission electron microscopy analyses. White arrows indicate the exosomes. (E) Identification of 8 exo‐miRNAs by exosome profiling in mouse models of liver diseases and human serum samples using Nanostring. Venn diagrams showing the results of an integrated analysis that revealed four commonly upregulated miRNAs (miRNA set 1) in serum exosomes from mouse models of liver diseases and human samples (left). Four additional miRNAs (miRNA set 2), which were upregulated in mouse models of liver diseases, were included. Finally, 8 exo‐miRNAs (miRNA set 3) are listed (right). (F) ROC curves showing the diagnostic performance of each selected exo‐miRNA for healthy controls versus patients with HCC (top) and cirrhosis versus patients with HCC (bottom) in the training cohort. (G) ROC curve analyses of the performance of each selected exo‐miRNA for healthy controls versus patients with HCC (top) and cirrhosis versus patients with HCC (bottom) in the validation cohort. (H) The radar chart summarizes the AUC values for each exo‐miRNA in the training (violet) and validation (blue) sets. (I) Scheme of the development of the AI‐based, multi‐marker model using eight exo‐miRNA signatures in the training set (n = 195), which included controls (n = 129, consisting of healthy controls [n = 30], those with gastric cancer [n = 49], and those with colorectal cancer [n = 50]), those with cirrhosis (n = 16), and those with HCC (n = 50). The training set was divided into training and test groups to develop the AI diagnostic model (upper panel). The training group, which was used to develop the model, consisted of controls (n = 45), those with cirrhosis (n = 8), and those with HCC (n = 25). The test group, which was used to evaluate the model, included controls (n = 84), those with cirrhosis (n = 8), and those with HCC (n = 25). The developed AI diagnostic model was verified in the validation set (n = 175) containing healthy controls (n = 30), those with hepatitis (n = 30), those with NASH (n = 20), those with cirrhosis (n = 15), and those with HCC (n = 80; bottom panel). (J‐K) The ROC curves with AUC values showing the ability of AFP level alone, a combination of miRNA set 1 with and without AFP level, a combination of miRNA set 2 with and without AFP level, and a combination of miRNA set 3 with and without AFP level, in distinguishing between patients with all stages of HCC and healthy controls and (J) between patients with early‐stage (stages I and II) HCC and healthy controls (K) in the validation set. (L‐M) The ROC curves and AUC values showing the diagnostic values of AFP level alone, a combination of miRNA set 1 with and without AFP level, a combination of miRNA set 2 with and without AFP level, and a combination of miRNA set 3 with and without AFP level in distinguishing between all stages of HCC and cirrhosis (L), and between early‐stage (stages I and II) HCC and cirrhosis (M). (N) Analysis of ROC curves with AUC values using the miRNA set 3 with AFP model for the training and validation sets after applying data augmentation techniques using the synthetic minority over‐sampling technique (SMOTE): the top table showing the original sample number and the SMOTE‐applied sample number in the training and validation sets; ROC curve with AUC values for distinguishing between healthy controls and HCC in the training and validation sets (bottom left); ROC curve with AUC values for distinguishing between cirrhosis and HCC in the training and validation sets (bottom right). (O) ROC curves with AUC values demonstrating the diagnostic performance of AI models based on different miRNA panels using raw samples (solid lines) and SMOTE‐augmented samples (dotted lines) from the GSE83977 dataset: the eight exo‐miRNAs combined with AFP level, the three exo‐miRNAs combined with AFP level, as proposed by Wang et al. [9], and the eight exo‐miRNA panel proposed by Sohn et al. [10]. The table summarizes the miRNA combinations used in each study, along with the corresponding AUC, sensitivity, specificity, and accuracy values. Abbreviations: AFP, alpha‐fetoprotein; AI, artificial intelligence; AUC, area under the curve; CCl4, carbon tetrachloride; CI, confidence interval; exo‐miRNA, exosomal microRNA; HCC, hepatocellular carcinoma; miRNA, microRNA; NASH, non‐alcoholic steatohepatitis; ND, normal diet; ROC, receiver operating characteristic; SMOTE, synthetic minority oversampling technique; WD, western diet.
See this image and copyright information in PMC

References

    1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021; 71(3):209‐49. - PubMed
    1. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin. 2024; 74(1):12‐49. - PubMed
    1. Hanif H, Ali MJ, Susheela AT, Khan IW, Luna‐Cuadros MA, Khan MM, et al. Update on the applications and limitations of alpha‐fetoprotein for hepatocellular carcinoma. World J Gastroenterol. 2022; 28(2):216‐29. - PMC - PubMed
    1. Chan YT, Zhang C, Wu J, Lu P, Xu L, Yuan H, et al. Biomarkers for diagnosis and therapeutic options in hepatocellular carcinoma. Mol Cancer. 2024;23(1):189. - PMC - PubMed
    1. Arzumanyan A, Reis HM, Feitelson MA. Pathogenic mechanisms in HBV‐ and HCV‐associated hepatocellular carcinoma. Nat Rev Cancer. 2013; 13(2):123‐35. - PubMed

LinkOut - more resources