Back to all papers

Deep learning model integrating contrast-enhanced ultrasound spatiotemporal imaging with clinical data for the differential diagnosis between hepatocellular carcinoma and intrahepatic cholangiocarcinoma.

October 28, 2025pubmed logopapers

Authors

Li W,Liu Z,Cheng M,Huang B,Hou C,Luo Y,Yang K,Lu M,Chen X,Wang W

Affiliations (4)

  • Department of Medical Ultrasonics, Institute of Diagnostic and Interventional Ultrasound, The First Affiliated Hospital of Sun Yat-Sen University, 58 Zhongshan Road 2, Guangzhou, 510080, People's Republic of China.
  • National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Shenzhen University Medical School, Shenzhen University, 1066 Xueyuan Road, Shenzhen, 518055, People's Republic of China.
  • National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Shenzhen University Medical School, Shenzhen University, 1066 Xueyuan Road, Shenzhen, 518055, People's Republic of China. [email protected].
  • Department of Medical Ultrasonics, Institute of Diagnostic and Interventional Ultrasound, The First Affiliated Hospital of Sun Yat-Sen University, 58 Zhongshan Road 2, Guangzhou, 510080, People's Republic of China. [email protected].

Abstract

This study aimed to develop a deep learning model, capable of extracting both spatial and temporal features from contrast-enhanced ultrasound (CEUS) data and integrating with patient clinical parameters, for the differential diagnosis between hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC). We retrospectively analyzed the CEUS data (ultrasound contrast agent: SonoVue®-sulfur hexafluoride microbubbles) from 165 ICC patients and 140 date-matched HCC patients. A deep learning model, namely CEUS-CD-Net, was developed to extract spatial-temporal features from dynamic CEUS data and integrate them with patient clinical parameters for the differential diagnosis between HCC and ICC. The performance of CEUS-CD-Net was evaluated using the area under the receiver operating characteristic curve (AUC), with comparisons against other methods including the single-source data-based models (CEUS-Net and CD-Net, based merely on dynamic CEUS or patient clinical data), CEUS static image-based model (sCEUS-Net), time-intensity curve-based model (TIC-Model), and the assessment by radiologists. CEUS-CD-Net achieved an AUC of 0.884 (95% CI, 0.794-0.938) on the test cohort, significantly outperforming the single-source data-based models of CEUS-Net (0.827 [0.730-0.896]) and CD-Net (0.812 [0.718-0.887]), as well as sCEUS-Net (0.772 [0.669-0.851]) and TIC-Model (0.731 [0.633-0.823]). In the subset of determinate cases, CEUS-CD-Net achieved an AUC of 0.893 [0.806-0.950], which was better than the one obtained by radiologists' assessment (0.790 [0.683-0.868]). Model visualization results revealed that CEUS-CD-Net surpassed radiologists in discerning subtle patterns reflected by CEUS. The integration of spatial and temporal features of dynamic CEUS data, coupled with clinical parameters of patients in CEUS-CD-Net, significantly improved the differential diagnosis between HCC and ICC.

Topics

Journal Article

Ready to Sharpen Your Edge?

Join hundreds of your peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.