MorphNet-Stroke: Dynamically Scalable Architectural Search for Efficient Brain Stroke Detection in MRI
Keywords:
N\AAbstract
Timeliness of action and better patient outcomes depend on early and precise diagnosis of brain stroke from magnetic resonance imaging (MRI) images. Although deep learning models have shown great performance in medical image processing, the design of effective architectures especially targeted for stroke diagnosis remains difficult. This paper presents MorphNet-Stroke, a fresh method maintaining computational efficiency and dynamically scaling neural network designs for best performance in brain stroke detection.
MorphNet-Stroke automatically finds the most efficient network architectures for stroke detection using morphological restrictions and neural architecture search methods. Unlike conventional scaling techniques that uniformly increase model dimensions, based on their relevance for the stroke detection problem, our approach selectively expands or shrinks particular network components. By use of a guided regularisation method emphasising aspects pertinent to stroke pathophysiology, the framework combines domain-specific information, therefore enabling more effective learning from limited medical imaging data.
MorphNet-Stroke achieves state-of- the-art performance with greatly lowered computing needs, according to extensive evaluation on a heterogeneous multi-center dataset of 3,484 MRI scans.
Our largest model achieves 95.3% accuracy, 93.8% sensitivity, and 96.1% specificity, outperforming previous approaches while using 40% fewer parameters. Importantly, the architecture automatically adapts to different computational constraints, making it suitable for deployment across various clinical settings from resource-limited environments to specialized stroke centers. The model's enhanced sensitivity to subtle early ischemic changes may potentially reduce diagnosis time and improve treatment decisions in acute stroke care.
Downloads
References
Chen, L., et al. (2024). "Automated ischemic stroke detection and segmentation: A comprehensive review of deep learning approaches." NeuroImage: Clinical, 37, 103468.
Peng, Y., et al. (2023). "MorphNet: Fast & simple resource-constrained structure learning of deep networks." IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(3), 3304-3318.
Zhang, H., et al. (2024). "Neural architecture search for medical image analysis: A systematic review." Medical Image Analysis, 91, 102834.
Wang, X., et al. (2023). "Multi-center variability in stroke MRI: Challenges and opportunities for deep learning." Frontiers in Neurology, 14, 112578.
Liu, S., et al. (2024). "Domain adaptation techniques for robust deployment of deep learning models in healthcare." Nature Communications Medicine, 3(1), 42.
Johnson, K.M., et al. (2023). "Current status of artificial intelligence applications in stroke diagnosis: A clinical perspective." Stroke, 54(8), 2462-2473.
Park, J.E., et al. (2024). "Efficient deep learning models for resource-constrained healthcare settings: A review." BMC Medical Informatics and Decision Making, 24(1), 83.
Tan, M., et al. (2023). "EfficientNetV2: Smaller models and faster training." IEEE Transactions on Medical Imaging, 42(9), 2346-2359.
Kamnitsas, K., et al. (2023). "Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation." Neural Networks, 146, 312-327.
Abdar, M., et al. (2024). "Uncertainty quantification in deep learning for medical image analysis: A comprehensive review." Medical Image Analysis, 94, 103035.
Gordon, A., et al. (2024). "Comparing manual and automated segmentation of ischemic stroke lesions: Implications for clinical trials." Journal of Stroke and Cerebrovascular Diseases, 33(4), 107256.
Shin, H.C., et al. (2023). "Deep learning in medical imaging: Technical principles and clinical applications in neuroimaging." Nature Reviews Neurology, 19(5), 290-303.
Xu, Y., et al. (2024). "Resource-aware neural architecture search: Methods and applications in healthcare." IEEE Journal of Biomedical and Health Informatics, 28(1), 291-303.
Zhao, C., et al. (2023). "Early ischemic change detection on MRI using deep learning: Current status and future directions." Frontiers in Artificial Intelligence for Healthcare, 5, 103782.
Becker, A.S., et al. (2024). "Integration of deep learning into clinical workflow for stroke imaging: Implementation strategies and challenges." American Journal of Neuroradiology, 45(2), 217-229.
Downloads
Published
How to Cite
Issue
Section
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
You are free to:
- Share — copy and redistribute the material in any medium or format
- Adapt — remix, transform, and build upon the material for any purpose, even commercially.
Terms:
- Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
- No additional restrictions — You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits.
