Role of Machine Learning Algorithms in Stroke Diagnosis and Treatment
Background: Stroke is one of the leading causes of morbidity, mortality, and disability in the United States. 1, 2 It has an incidence of 700,000 cases per year, prevalence in 2.5% of adults, and 4 million survivors. Ischemic stroke is the result of decreased blood flow to the brain due to thrombosis, embolism, or reduced overall perfusion.1 The oxygen and nutrient deficiency caused by these strokes leads to hypoxic injury, thereby inhibiting ATP generation and instigating an influx of neutrophils. This cascade prompts the release of hydrolytic enzymes, causing liquefactive necrosis or tissue death.3 Diagnosis is typically suspected based on history, physical examination, and clinical judgment, and then confirmed through advanced imaging like CT and MRI. CT, despite its high rate of false positives, is often the first choice due to its availability, time-efficiency, and cost-effectiveness. Therapeutically, tissue plasminogen activator is administered to dissolve the clot in severe cases within a specific time window of 4.5 hours.1 Alternatively, mechanical thrombectomy, may also be performed.
Objective: Availability of stroke treatment options are time dependent. Therefore, multiple artificial intelligence (AI) models are in development. An analysis of multiple study results was done to gain data on the effectiveness of AI on stroke diagnosis and treatment.
Search Methods: GoogleScholar search was performed for studies published in the last 10 years. Studies used were related to the use of machine learning and artificial intelligence in stroke detection.
Results: ML algorithms using logistic regression, support vector machines, and random forest all had significantly greater sensitivities at detecting stroke using a DWI-FLAIR mismatch than human readers. 5 Specificity was comparable between all groups. Another study showed that ML algorithm using a deep convolutional neural network was able to identify the hyperdense middle cerebral artery sign to accurately predict the presence of an MCA stroke. The identification rate was comparable to neuroradiologists.6 ML algorithms were also able to predict effectiveness of endovascular treatment using modified Rankin Scale and imaging characteristics. 7
Conclusions: Artificial intelligence is a new promising method of detecting stroke that may be more effective than human readers. Currently, data supports higher sensitivity with comparable specificity. Future directions involve further research and development and increase in specificity.
- Van der Worp, H. B., & van Gijn, J. Acute ischemic stroke. New England Journal of Medicine, (2007). 357(6), 572-579.
- Feske SK. Ischemic Stroke. American Journal of Medicine; (2021). 134(12):1457-1464.
- Adigun, R., Basit, H., & Murray, J. Necrosis, cell (liquefactive, coagulative, caseous, fat, fibrinoid, and gangrenous). (2019) StatPearls
- Kaothanthong N, Atsavasirilert K, Sarampakhul S, Chantangphol P, Songsaeng D, Makhanov S. Artificial intelligence for localization of the acute ischemic stroke by non-contrast computed tomography. PLoS One. 2022;17(12):e0277573.
- Lee, H., Lee, E.-J., Ham, S., Lee, H.-B., Lee, J. S., Kwon, S. U., Kim, J. S., Kim, N., & Kang, D.-W. (2020). Machine Learning Approach to Identify Stroke Within 4.5 Hours. In Stroke (Vol. 51, Issue 3, pp. 860–866). Ovid Technologies (Wolters Kluwer Health)
- Shinohara Y, Takahashi N, Lee Y, Ohmura T, Umetsu A, Kinoshita F, Kuya K, Kato A, Kinoshita T. Usefulness of deep learning-assisted identification of hyperdense MCA sign in acute ischemic stroke: comparison with readers’ performance. Jpn J Radiol. 2020 Sep;38(9):870-877. doi: 10.1007/s11604-020-00986-6. Epub 2020 May 12. PMID: 32399602.
- Brugnara, G., Neuberger, U., Mahmutoglu, M. A., Foltyn, M., Herweh, C., Nagel, S., Schönenberger, S., Heiland, S., Ulfert, C., Ringleb, P. A., Bendszus, M., Möhlenbruch, M. A., Pfaff, J. A. R., & Vollmuth, P. (2020). Multimodal Predictive Modeling of Endovascular Treatment Outcome for Acute Ischemic Stroke Using Machine-Learning. In Stroke (Vol. 51, Issue 12, pp. 3541–3551). Ovid Technologies (Wolters Kluwer Health). https://doi.org/10.1161/strokeaha.120.030287