Open Collections

Gidda, Harseerat

Introduction: Strokes are a time-sensitive disease that rely heavily on efficient detection, diagnosis, and intervention. Artificial intelligence (AI) is a quickly advancing technology that has made its way into healthcare. Over the last decade, there is a significant increase in evidence related to the application of AI into stroke detection and diagnosis. However, the extent, range, and nature of this evidence has not been mapped comprehensively. The purpose of this scoping review is to identify the types of AI being used for stroke detection or diagnosis in adults presenting with stroke-like symptoms, as well as to present the timeliness, accuracy, sensitivity, and specificity of the current AI tools in reporting stroke. By mapping the available literature, this scoping review aims to provide a foundation for understanding the current available evidence, and identifying potential gaps so that the research is effectively able to inform practice. Methods: This Scoping Review was conducted with the guidance of Arksey & O’Malley’s (2005) framework for scoping reviews. The literature search was carried out using two databases, Cumulative Index of Nursing and Allied Health Literature (CINAHL) and Medical Literature Analysis and Retrieval System Online (MEDLINE [PubMed]), for articles related to the use of AI tools for the detection or diagnosis phase of stroke, from symptom onset to formal diagnosis. Adults aged 18 years and older were included of studies published within the last 10 years written in English. This data collected was reported as a descriptive analysis. Results: Twenty studies met the inclusion criteria and were included for full review. Three types of AI were identified in the studies, these were machine learning, deep learning, and natural language processing. Four of the studies used machine learning AI, 14 of the studies used deep learning AI, one study compared machine learning to deep learning, and one study used natural language processing AI. Each study explored a different type of AI to aid in the detection or diagnosis of stroke. Out of the 20 studies, only one was specific to diagnosis of stroke. The sensitivity, specificity, and accuracy of each AI stroke tool varied, although most studies reported positive results comparative to those of clinicians in detecting stroke by significantly improving the speed and accuracy in detecting large vessel occlusions (LVOs) and intracranial hemorrhages (ICHs). Reduction in time to detect stroke was evident in many of the studies, but none recommended the substitution of clinicians’ judgement with an AI tool. Only a single study identified diagnosis of a stroke by AI. Conclusions: The findings demonstrated that AI tools show promise in supporting the detection and diagnosis of stroke. However, majority of the studies recommend using AI tools to support decision-making and not to replace the clinical judgement of health care professionals. Findings also indicate a need for further research to validate the reliability and effectiveness of AI tools across diverse patient populations.

Text

University of British Columbia. NURS_V 595

Vancouver : University of British Columbia Library

10.14288/1.0448126

Applied Science, Faculty of; Nursing, School of

Unreviewed

http://creativecommons.org/licenses/by-nc-nd/4.0/