Comparative analysis of the results of predicting diameters of daughter segments of coronary artery bifurcations under normal conditions, obtained using graph neural networks and known numerical modeling methods

Background. The accurate determination of human coronary artery (HCA) diameters plays a pivotal role in the diagnosis, risk assessment, and treatment planning of ischemic heart disease. Given that vascular networks possess a fractal structure and consist of arterial bifurcations (AB), a critical task is the prediction of the diameters of daughter (distal) arterial segments (AS) based on the known diameter of the mother (proximal) AS. Existing numerical modeling methods (ENMM) for predicting arterial diameters rely on approximate analytical dependencies derived empirically.

Aim. To conduct a comparative analysis of the results of predicting diameters of daughter segments (bifurcation components) obtained using Graph Neural Networks and known numerical modeling methods based on morphometric data of real healthy human coronary arteries.

Methods. To achieve this objective, various GNN architectures were implemented and tested: GCN, GraphSAGE, GAT, and TransformerConv. Additionally, ENMM and a Multilayer Perceptron (MLP) were evaluated. A comparative statistical analysis of the results was conducted. The study utilized previously obtained morphometric data on the internal diameters of AS comprising 30 corrosion cast graphs of healthy HCAs. Results. This study proposes the use of GNNs for this specific task for the first time. Comparative analysis demonstrated that all GNN models surpassed both ENMM and the Multilayer Perceptron in accuracy when predicting the diameters of smaller-caliber daughter AS. The TransformerConv architecture demonstrated the best performance, achieving a coefficient of determination R2 = 0.96 and RMSE = 0.199 for predicting the larger daughter segment diameter, and R2 = 0.76 and RMSE = 0.221 for the smaller segment.

Conclusion. The developed models may serve as the foundation for creating clinical decision support systems, enabling more precise assessments of the structural status of human coronary arteries.

1. Y.-li Chen, G.-Q. Bai, L.-xing Ren, Y. Bai, M.-y Sun, T. Shang, et al. Blood physiological and flow characteristics within coronary artery circulatory network for human heart based on vascular fractal theory. Advances in Mechanical Engineering. 2020; 12(7): 1–13. doi: 10.1177/1687814020933385

2. Zenin OK, Kafarov ES, Miltiadis I. Comparative Analysis of Morphometric Data on Internal Diameters of Segments Forming Bifurcations in Corrosion Casts of Human Coronary Arteries and Their Calculation Using Contemporary Methods. Acta biomedical scientifica. 2025; 10(5): 100-106. (In Russ.). doi: 10.29413/ABS.2025-10.5.11

3. Huo Y, Kassab GS. Intraspecific scaling laws of vascular trees. Journal of the Royal Society Interface. 2012; 9(66): 190–200. doi: 10.1098/rsif.2011.0270

4. Olufsen MS, Peskin CS, Kim WY, Pedersen EM, Nadim A, Larsen J. Numerical Simulation and Experimental Validation of Blood Flow in Arteries with Structured-Tree Outflow Conditions. Ann Biomed Eng. 2000; 28(11): 1281-1299. doi: 10.1114/1.1326031

5. Labonne M. Hands-On Graph Neural Networks Using Python: Practical techniques and architectures for building powerful graph and deep learning apps with PyTorch. Birmingham-Mumbai: Packt Publishing; 2023.

6. Certificate of state registration of database No. 2021623049 Russian Federation. Quantitative anatomy of the intraorgan arterial channel of the human heart: No. 2021622956: submitted. 14.12.2021: publ. 20.12.2021 / O. Zenin, AV Dmitriev, IS Miltykh; applicant Federal State Budgetary Educational Institution of Higher Education “Penza State University”. (In Russ.).

7. PyG documentation — pytorch_geometric documentation. URL: https://pytorch-geometric.readthedocs.io/en/latest/ [date of access: December 25, 2025].

8. NetworkX — NetworkX documentation. URL: https://networkx.org/en/ [date of access: December 25, 2025].

9. Scikit-learn: machine learning in Python — scikitlearn 1.8.0 documentation. URL: https://scikit-learn.org/stable/ [date of access: December 25, 2025].

10. AdamW — PyTorch 2.9 documentation. URL: https://docs.pytorch.org/docs/stable/generated/torch.optim.AdamW.html [date of access: December 25, 2025].

11. HuberLoss — PyTorch 2.9 documentation. URL: https://docs.pytorch.org/docs/stable/generated/torch.nn.HuberLoss.html [date of access: December 25, 2025].

12. Akiba T, Sano S, Yanase T, Ohta T, Koyama M. Optuna: a next-generation hyperparameter optimization framework. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2019. doi: 10.1145/3292500.3330701

13. Hamilton WL, Ying R, Leskovec J. Inductive representation learning on large graphs. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. NIPS’17. Curran Associates Inc.; 2017: 1025-1035.

14. Veličković P, Cucurull G, Casanova A, Romero A, Liò P, Bengio Y. Graph Attention Networks. arXiv. Preprint posted online. 2017. doi: 10.48550/ARXIV.1710.10903

15. Shi Y, Huang Z, Feng S, Zhong H, Wang W, Sun Y. Masked label prediction: unified message passing model for semi-supervised classification. Preprint posted online September 8, 2020. doi: 10.48550/arXiv.2009.03509

16. Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need. arXiv. Preprint posted online. 2023. doi: 10.48550/arXiv.1706.03762