Advanced Multi-Stream Deep Learning Framework for Accurate Ischemic Stroke Diagnosis and Segmentation Using MRI Images

Advanced Multi-Stream Deep Learning Framework for Accurate Ischemic Stroke Diagnosis and Segmentation Using MRI Images R. Bhuvana Department of Biomedical Engineering, GKM College of Engineering & Technology, Chennai 600063, Tamil Nadu, India https://orcid.org/0000-0002-1477-3640 R. J. Hemalatha Department of Biomedical Engineering, Vels Institute of Science Technology and Advanced Studies, Chennai 600117, Tamil Nadu, India https://orcid.org/0000-0003-4712-7482

The significant amount of inherent noise and complexity in magnetic resonance imaging (MRI) images is causing difficulty in detecting and segmentation of Ischemic Stroke (IS) caused using an advanced integrated solution. Modern high-end deep learning techniques are shown to be capable of rendering improvement in diagnosing IS in terms of incorporating MRI images in terms of scientific strokes. In this context, we propose a comprehensive decision-making system based on four keys. The first is the use of a Convolution-LSTM-Deep Neural Network (CLDNN) for denoising MRI images specific to the IS diagnosis. Next, fusion techniques will leverage the combined utilities of multi-MRI modalities and clinical data to enable much broader and more precise diagnoses of the IS. This would be implemented through an MM-DLA-UM-M model made up of multi-stream deep learnings with mechanisms of attention feature extraction and interpretation. In addition to that, complex hierarchical deep learning models were set up in order to empower precise segmentation of locations in the brain that involvement stroke events. A segmented database is then made consisting of modular coarse-to-fine segmentation and sophisticated data augmentation techniques for improving segmentation performance and generalizability. Finally, uncertainty estimation techniques are integrated to provide clinicians with uncertainty-aware diagnostic recommendations, supported by a user-friendly interface for visualization and clinical decision-making. By conducting comprehensive validation tests on large-scale clinical datasets, the proposed methodology is assessed and found to significantly outperform existing methods in terms of brain stroke segmentation and IS diagnostic accuracy. The findings indicate that the suggested strategy has the potential to improve diagnosis accuracy and facilitate efficient patient care in actual clinical settings.
08 27 2025 12 2025 2550049 10.1142/S021947752550049X 10.1142/S021947752550049X https://www.worldscientific.com/doi/10.1142/S021947752550049X https://www.worldscientific.com/doi/pdf/10.1142/S021947752550049X 10.24018/ejece.2023.7.1.483 Rev. Intell. Artif. Bandi V. 753 34 6 2020 10.1155/2022/5373585 10.1109/JBHI.2021.3097591 10.1155/2022/2456550 10.1155/2022/7725597 10.1007/s00521-019-04041-y 10.1016/j.clineuro.2020.105892 Mobile Inf. Syst. Alotaibi N. S. 2022 2022 10.1002/aisy.202200240 10.1109/ACCESS.2023.3312046 10.18280/ts.390401 10.3389/fneur.2023.1105616 10.3390/app10186337 10.1186/s12874-022-01672-z 10.1063/5.0193935 10.1016/j.ijcard.2021.11.005 10.1155/2021/3628179 10.1093/ehjci/ehaa946.2418 10.1016/j.bspc.2022.103948 10.1038/s43856-021-00062-8 10.1016/j.eij.2020.07.005 10.3390/rs12162509 10.3390/s22030867 10.1016/j.compbiomed.2021.104699 10.1016/j.bspc.2024.106550 10.1016/j.bspc.2024.106286 10.1016/j.bspc.2024.106872 10.1016/j.bspc.2024.106811 10.1016/j.bspc.2025.107637 10.1016/j.bspc.2024.106390 J. Intell. Fuzzy Syst. Gupta S. 9785 45 6 2023 Multimed. Tools Appl. Raju K. N. 1 2024 Multimed. Tools Appl. Syamala M. 1 2024 10.1007/s42835-023-01654-1 10.1016/j.rinam.2024.100504 10.1142/S0129183123500705 10.1109/ICICT57646.2023.10134487