Education, Science, Technology, Innovation and Life
Open Access
Sign In
  1. Home
  2. Journals
  3. Journal of Artificial Intelligence Practice
  4. Vol 9, Issue 1, 2026

A Pluggable Hybrid Generative Model for MVI Classification of Hepatocellular Carcinoma Based on ICG Fluorescence Imaging

Download as PDF

DOI: 10.23977/jaip.2026.090108 | Downloads: 7 | Views: 199

Author(s)

Zheng Tao 1

Affiliation(s)

1 School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, 200093, China

Corresponding Author

Zheng Tao

ABSTRACT

Microvascular invasion (MVI) of hepatocellular carcinoma (HCC) is a key factor affecting clinical diagnosis and prognosis, but the classification of MVI based on indocyanine green (ICG) fluorescence images is limited by insufficient clinical data (a total of 213 ICG laparoscopic fluorescence images were used in the experiment) and difficulty in capturing subtle features. To address these problems, this study proposes a pluggable hybrid generative model for MVI classification and verifies the value of generative data augmentation in small-sample medical image classification through multiple experiments. The proposed Hybrid module integrates the detailed fitting ability of LightGAN and the generation stability of DDIM, with a closed-loop iteration optimal interval of 5-10 rounds (excessive iteration leads to performance degradation). Experimental results show that when iterated 5 times on ResNet18, the model achieves an accuracy of 86.92% and an AUC value of 0.8846, which are 4.68% and 0.0546 higher than the peaks of LightGAN and DDIM respectively; the module not only improves the peak performance of a single backbone network but also has good generalization to different backbone networks (e.g., ConvTransResNet achieves an accuracy improvement of 7.48%, ResNet50 achieves an F1-score improvement of 7.58%), narrowing the performance gap between networks; it can effectively improve model accuracy, F1-score and AUC without additional clinical data collection, has a greater improvement effect on weak-performance networks, and reduces the dependence on high-end computing resources in medical scenarios. This study provides a feasible technical framework for small-sample medical image classification tasks and experimental support for the development of subsequent medical image auxiliary diagnosis systems.

KEYWORDS

Hepatocellular Carcinoma; Microvascular Invasion; ICG Fluorescence Imaging; Generative Data Augmentation; Hybrid Generative Model; Deep Learning

CITE THIS PAPER

Zheng Tao. A Pluggable Hybrid Generative Model for MVI Classification of Hepatocellular Carcinoma Based on ICG Fluorescence Imaging. Journal of Artificial Intelligence Practice (2026). Vol. 9, No. 1, 57-67. DOI: http://dx.doi.org/10.23977/jaip.2026.090108.

REFERENCES

[1] Bray, F. et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA A Cancer J Clinicians 74, 229–263 (2024).
[2] Cheng, Z. et al. Risk factors and management for early and late intrahepatic recurrence of solitary hepatocellular carcinoma after curative resection. HPB 17, 422–427 (2015).
[3] Raoul, J.-L. & Edeline, J. Systemic treatment of hepatocellular carcinoma: standard of care in China and elsewhere. The Lancet Oncology 21, 479–481 (2020).
[4] Tang, Y. et al. Identification and Validation of a Prognostic Model Based on Three MVI-Related Genes in Hepatocellular Carcinoma. Int. J. Biol. Sci. 18, (2022).
[5] Wang, F. et al. Predicting microvascular invasion in small (≤ 5 cm) hepatocellular carcinomas using radiomics-based peritumoral analysis. Insights Imaging 15, 90 (2024).
[6] Zheng, R. et al. Comparison of non-radiomics imaging features and radiomics models based on contrast-enhanced ultrasound and Gd-EOB-DTPA-enhanced MRI for predicting microvascular invasion in hepatocellular carcinoma within 5 cm. Eur Radiol 33, 6462–6472 (2023).
[7] Xu, C., Jiang, D., Tan, B., Shen, C. & Guo, J. Preoperative diagnosis and prediction of microvascular invasion in hepatocellularcarcinoma by ultrasound elastography. BMC Med Imaging 22, 88 (2022).
[8] Wang, G. Prediction of Microvascular Invasion of Hepatocellular Carcinoma Based on Preoperative Diffusion-Weighted MR Using Deep Learning. Academic Radiology (2020).
[9] Zhou, W. Prediction of Microvascular Invasion of Hepatocellular Carcinoma Based on Contrast-Enhanced MR and 3D Convolutional Neural Networks. Frontiers in Oncology 11, (2021).
[10] Cen Y-Y, et al. Computed tomography-based deep learning and multi-instance learning for predicting microvascular invasion and prognosis in hepatocellular carcinoma. World Journal of Gastroenterology, 2025, 31(30): 109186. doi:10.3748/wjg.v31.i30.109186.
[11] Cao, L. et al. MVI-TR: A Transformer-Based Deep Learning Model with Contrast-Enhanced CT for Preoperative Prediction of Microvascular Invasion in Hepatocellular Carcinoma. https://doi.org/10.3390/cancers15051538 (2023) doi:10.3390/cancers15051538.
[12] Zhou, M. A rock joint roughness coefficient determination method incorporating the SE-Net attention mechanism and CNN. Environmental Earth Sciences https://doi.org/10.1007/s12665-025-12414-x (2025) doi:10.1007/s12665-025-12414-x.
[13] Wang, Q. et al. ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks. https://doi.org/10.1109/CVPR42600.2020.01155 doi:10.1109/CVPR42600.2020.01155.
[14] Anshelevich, D. Synthetic tabular data generation using a VAE-GAN architecture. https://doi.org/10.1016/j.knosys.2025.113997 (2025) doi:10.1016/j.knosys.2025.113997.
[15] Radford, A., Metz, L. & Chintala, S. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. Preprint at http://arxiv.org/abs/1511.06434 (2016).
[16] Isola, P., Zhu, J.-Y., Zhou, T. & Efros, A. A. Image-to-Image Translation with Conditional Adversarial Networks. https://doi.org/10.1109/CVPR.2017.632 doi:10.1109/CVPR.2017.632.
[17] Ho, J., Jain, A. & Abbeel, P. Denoising Diffusion Probabilistic Models. Preprint at http://arxiv.org/abs/2006.11239 (2020).
[18] Song, J., Meng, C. & Ermon, S. DENOISING DIFFUSION IMPLICIT MODELS. https://doi.org/10.48550/arXiv.2010.02502 (2021).
[19] Kose, E. et al. A comparison of indocyanine green fluorescence and laparoscopic ultrasound for detection of liver tumors. HPB 22, 764–769 (2020).
[20] Szostek, A. K., Grimes, J. A. & Wallace, M. L. Average fluorescence intensity is affected by distance and penetration depth with near-infrared fluorescence imaging with indocyanine green in an ex vivo model. https://doi.org/10.2460/ajvr.25.01.0001 doi:10.2460/ajvr.25.01.0001.
[21] Cai, X. Does Using Indocyanine Green Fluorescence Imaging for Tumors Help in Determining the Safe Surgical Margin in Real-Time Navigation of Laparoscopic Hepatectomy? A Retrospective Study. https://doi.org/10.1245/s10434-022-12893-3 doi:10.1245/s10434-022-12893-3.
[22] Krizhevsky, A., Sutskever, I. & Hinton, G. E. ImageNet classification with deep convolutional neural networks. Commun. ACM 60, 84–90 (2017).
[23] Simonyan, K. & Zisserman, A. VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION. https://doi.org/10.48550/arXiv.1409.1556 (2015).
[24] Szegedy, C. et al. Going Deeper with Convolutions. https://doi.org/10.1109/CVPR.2015.7298594 doi:10.1109/CVPR.2015.7298594.
[25] He, K., Zhang, X., Ren, S. & Sun, J. Deep Residual Learning for Image Recognition. Preprint at http://arxiv.org/abs/1512.03385 (2015).

Subscription
E-Mail Alert
Most popular papers
Call for Special Issue
Publication Ethics & OA Statement
Plagiarism Policy
Contact Us
Downloads: 25773
Visits: 725898

Sponsors, Associates, and Links

All published work is licensed under a Creative Commons Attribution 4.0 International License.

Copyright © 2016 - 2031 Clausius Scientific Press Inc. All Rights Reserved.