Caleb Haeussler

Introduction. Ischemic stroke is a leading cause of morbidity and mortality in the United States, with an incidence of 700,000 ischemic strokes per year and 4 million stroke survivors.^1,2 Ischemic stroke is a result of decreased blood flow to the brain, usually as result of thrombosis, embolism, or globally decreased perfusion³ This decreased blood flow results in hypoxic injury due to inability to generate ATP in the absence of oxygen, causing an influx of neutrophils which release hydrolytic enzymes to surrounding tissue and cause liquefactive necrosis.⁴ Presumptive stroke diagnosis can be made by history, physical exam, and clinical decision making while a definitive diagnosis depends on advanced imaging.^1,2 If symptoms are severe and patient arrives within appropriate window, tissue plasminogen activator is administered to breakdown the fibrin net and allow for lysis of the clot.¹ Mechanical thrombectomy may be performed from peripheral puncture into the occluded artery, however this treatment is only viable if sufficient collateral circulation is seen on a perfusion imaging.¹ Methods. Machine learning and artificial intelligence have been demonstrated to be useful clinical tools for diagnosis, guiding treatment, and predicting outcomes of acute ischemic stroke. For diagnosis, a deep convolutional neural network has been evaluated for detecting hyperdense MCA sign as an early predictor of middle cerebral artery infarction.⁵ Additionally, a transcranial electroencephalography trace combined with clinical data was evaluated by deep learning machine learning to determine presence of acute stroke/TIA and if large vessel occlusion was present⁶. For guiding treatment, a matrix completion algorithm was used to evaluate collateral circulation on a 4D CT angiography.⁷ To predict outcomes, a multimodal predictive model was used to determine which criteria provided the best predictive value for disability at 90 days after ischemic stroke.⁸ Additionally, a machine learning model was trained on localized voxels of brain regions of diffusion/perfusion weighted MRI to predict the size of ischemic lesion.⁹ Results. The DCNN model for hyperdense MCA sign improved diagnostic accuracy of the neuroradiologists, especially negative predictive value.⁵ The Electroencephalography model was superior to human interpretation and required less training.⁶ The Matrix completion algorithm was equivalent to radiologist accuracy with improved consistency.⁷ Localized voxel data from perfusion-weighted MRI was beneficial in estimating follow-up lesion size.⁹ Multimodal model showed improvement in predicting modified rankin score at 90 days after ischemic stroke.⁸ Conclusion. The selected articles demonstrate benefits of AI/ML including reduced user-dependent variability, improved accuracy, and reduced training needed to accurately diagnose and guide treatment of ischemic stroke.

1.  Van der Worp, H. B., & van Gijn, J. (2007). Acute ischemic stroke. New England Journal of Medicine, 357(6), 572-579.
2. Panel, Mohr, J. P., Albers, G. W., Amarenco, P., Babikian, V. L., Biller, J., … & Turpie, A. G. (1997). Etiology of stroke. Stroke, 28(7), 1501-1506.
3. Panel, Mohr, J. P., Albers, G. W., Amarenco, P., Babikian, V. L., Biller, J., … & Turpie, A. G. (1997). Etiology of stroke. Stroke, 28(7), 1501-1506.
4. Adigun, R., Basit, H., & Murray, J. (2019). Necrosis, cell (liquefactive, coagulative, caseous, fat, fibrinoid, and gangrenous). StatPearls [Internet].
5. Shinohara Y, Takahashi N, Lee Y, et al. Usefulness of deep learning-assisted identification of hyperdense MCA sign in acute ischemic stroke: comparison with readers’ performance. Jpn J Radiol. 2020;38(9):870-877. doi:10.1007/s11604-020-00986-6
6. Erani F, Zolotova N, Vanderschelden B, et al. Electroencephalography Might Improve Diagnosis of Acute Stroke and Large Vessel Occlusion. Stroke. 2020;51(11):3361-3365. doi:10.1161/STROKEAHA.120.030150
7. Aktar, M., Tampieri, D., Rivaz, H., Kersten-Oertel, M., & Xiao, Y. (2020). Automatic collateral circulation scoring in ischemic stroke using 4D CT angiography with low-rank and sparse matrix decomposition. International Journal of Computer Assisted Radiology and Surgery, 15(9), 1501-1511.
8. Brugnara G, Neuberger U, Mahmutoglu MA, et al. Multimodal Predictive Modeling of Endovascular Treatment Outcome for Acute Ischemic Stroke Using Machine-Learning. Stroke. 2020;51(12):3541-3551. doi:10.1161/STROKEAHA.120.030287
9. Grosser M, Gellißen S, Borchert P, et al. Localized prediction of tissue outcome in acute ischemic stroke patients using diffusion- and perfusion-weighted MRI datasets. PLoS One. 2020;15(11):e0241917. Published 2020 Nov 5. doi:10.1371/journal.pone.0241917