TRANSFORMER BASED ENSEMBLE FOR ISCHEMIC STROKE SEGMENTATION ON 3D CT SCANS

Ischemic stroke is one of the leading causes of mortality and disability. Accurate segmentation of damaged regions in brain CT images is critical for timely diagnosis and clinical decision-making. In this study, an ensemble approach is proposed, combining SE-UNETR and Swin UNETR transformer models via weighted voting. The Dice coefficient was used for evaluation, measuring the overlap between predicted lesion regions and reference annotations. Unlike single-model approaches, ensemble neural network methods provide higher reliability and segmentation accuracy by integrating predictions from multiple architectures. Three-dimensional CT scans of 98 patients with acute ischemic stroke, provided by the International Tomography Center of the Siberian Branch of the Russian Academy of Sciences, were used. The results demonstrated that the proposed ensemble outperforms individual models. The average Dice coefficient was 0.7983, indicating the high effectiveness of the method in segmenting ischemic lesions. Analysis showed that the ensemble approach more accurately delineates lesion boundaries in brain CT images and reduces segmentation errors. The proposed method can be applied not only to stroke but also to other pathologies requiring precise medical image analysis in automated diagnostic systems.

1. Bakator, M., & Radosav, D. Deep learning and medical diagnosis: A review of literature. Multimodal Technologies and Interaction, 2(3), 47 (2018). https://doi.org/10.3390/mti2030047

2. Zhu, S., Xia, X., Zhang, Q., & Belloulata, K. An image segmentation algorithm in image processing based on threshold segmentation. Proceedings of the Third International IEEE Conference on Signal-Image Technologies and Internet-Based Systems, Shanghai, China, 673–678 (2007). https://doi.org/10.1109/SITIS.2007.116

3. Chen, X., Williams, B.M., Vallabhaneni, S.R., Czanner, G., Williams, R., & Zheng, Y. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 11632–11640 (2019).

4. Khan, M.Z., Gajendran, M.K., Lee, Y., & Khan, M.A. Deep neural architectures for medical image semantic segmentation: Review. IEEE Access, 9, 83002–83024 (2021). https://doi.org/10.1109/ACCESS.2021.3086530.

5. Tursynova, A., & Omarov, B. 3D U-Net for brain stroke lesion segmentation on ISLES 2018 dataset. Proceedings of the 16th International Conference on Electronics Computer and Computation (ICECCO), Kazakhstan, 1–4 (2021). https://doi.org/10.1109/ICECCO53203.2021.9663825.

6. Khan, W.R., Madni, T.M., Janjua, U.I., Javed, U., Khan, M.A., Alhaisoni, M., Tariq, U., & Cha, J.-H. A hybrid attention-based residual U-Net for semantic segmentation of brain tumor. Computers, Materials & Continua, 76(1), 647–664 (2023). https://doi.org/10.32604/cmc.2023.039188.

7. Zhang, B., Qiu, S., & Liang, T. Dual attention-based 3D U-Net liver segmentation algorithm on CT images. Bioengineering, 11, 737 (2024). https://doi.org/10.3390/bioengineering11070737.

8. Siddique, N., Paheding, S., Elkin, C.P., & Devabhaktuni, V. U-Net and its variants for medical image segmentation: A review of theory and applications. IEEE Access, 9, 82031–82057 (2021). https://doi.org/10.1109/ACCESS.2021.3086020.

9. Pinheiro, G.R., Voltoline, R., Bento, M., & Rittner, L. V-Net and U-Net for ischemic stroke lesion segmentation in a small dataset of perfusion data. In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries (BrainLes 2018). Cham: Springer, 301–309 (2019).

10. Atika, L., Nurmaini, S., Partan, R.U., & Sukandi, E. Image segmentation for mitral regurgitation with convolutional neural network based on U-Net, ResNet, V-Net, FractalNet, and SegNet: A preliminary study. Big Data and Cognitive Computing, 6(4), 141 (2022). https://doi.org/10.3390/bdcc6040141.

11. Amirgaliyev, Y.N., Buribayev, Z.A., Melis, Z.M., & Ataniyazova, A.S. On one approach to recognizing fuzzy images of faces based on an ensemble. Proceedings of the 25th International Conference on Circuits, Systems, Communications and Computers (CSCC) (2021). https://doi.org/10.1109/CSCC53858.2021.00011.

12. Berikov, V.B., & Cherikbayeva, L.S. Searching for optimal classifier using a combination of cluster ensemble and kernel method. CEUR Workshop Proceedings, 2098, 45–60 (2018).

13. Dobshik, A.V., Verbitskiy, S.K., Pestunov, I.A., Sherman, K.M., Sinyavskiy, Y.N., Tulupov, A.A., & Berikov, V.B. Acute ischemic stroke lesion segmentation in non-contrast CT images using 3D convolutional neural networks. Computer Optics, 47(5), 770–777 (2023). https://doi.org/10.18287/2412-6179-CO-1233.

14. Hatamizadeh, A., Tang, Y., Nath, V., Yang, D., Myronenko, A., Landman, B., Roth, H., & Xu, D. UNETR: Transformers for 3D medical image segmentation. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) (2022). https://doi.org/10.1109/WACV51458.2022.00109.

15. Hatamizadeh, A., Yang, D., Roth, H.R., & Xu, D. Swin UNETR: Swin transformers for semantic segmentation of brain tumors in MRI images. arXiv preprint arXiv:2201.01266 (2022). https://arxiv.org/abs/2201.01266.

16. Tang, Y., Yang, D., Li, W., Roth, H.R., Landman, B., & Xu, D. Self-supervised pre-training of Swin transformers for 3D medical image analysis. arXiv preprint arXiv:2111.14791 (2022).

17. Zhao, W., Li, Y., Lin, S., et al. 3D medical image segmentation using hybrid Swin transformers: A comparative study. Medical Imaging with Deep Learning (MIDL) (2023).

18. Sriramakrishnan, P., Kalaiselvi, T., Padmapriya, S.T., Shanthi, N., Ramkumar, S., & Kalaichelvi, N. An medical image file formats and digital image conversion. International Journal of Engineering and Advanced Technology, 9(1S3), 74–78 (2019).

19. Wang, Y., Wang, H., Shen, K., Chang, J., & Cui, J. Brain CT image segmentation based on 3D slicer. Journal of Complexity in Health Sciences, 3(1), 34–42 (2020). https://doi.org/10.21595/chs.2020.21263.

20. Kleesiek, J., Urban, G., Hubert, A., Schwarz, D., Maier-Hein, K., Bendszus, M., & Biller, A. Deep MRI brain extraction: A 3D convolutional neural network for skull stripping. NeuroImage, 129, 460–469 (2016). https://doi.org/10.1016/j.neuroimage.2016.01.024.

21. Lima, F.T., & Souza, V.M.A. A large comparison of normalization methods on time series. Big Data Research, 34, 100407 (2023). https://doi.org/10.1016/j.bdr.2023.100407.

22. Jin, X., Xie, Y., Wei, X.-S., Zhao, B.-R., & Chen, Z.-M. Delving deep into spatial pooling for squeeze and excitation networks. Pattern Recognition, 121, 108159 (2021). https://doi.org/10.1016/j.patcog.2021.108159.

23. Ogundokun, R.O., Maskeliunas, R., Misra, S., & Damaševičius, R. Improved CNN based on batch normalization and Adam optimizer. In: International Conference on Computational Science and Its Applications. Cham: Springer, 593–604 (2022). https://doi.org/10.1007/978-3-031-10545-6_43.

24. Choquette, J., Gandhi, W., Giroux, O., Stam, N., & Krashinsky, R. NVIDIA A100 Tensor Core GPU: Performance and innovation. IEEE Micro, 41(2), 29–35 (2021). https://doi.org/10.1109/MM.2021.3053039.

25. Wang, J., Wang, S., & Liang, W. METrans: Multi-encoder transformer for ischemic stroke segmentation. Electronics Letters, 58, 340–342 (2022). https://doi.org/10.1049/ell2.12444.

26. Omarov, M., Ibragimov, M., Kaldybekov, K., & Aytmahanov, M. Modified 3D U-Net for brain stroke lesion segmentation on computed tomography images. Computers, 11, 23 (2022). https://doi.org/10.3390/computers11020023.