Artificial intelligence diagnostic performance in image-based vulnerable carotid plaque detection: a systematic review and meta-analysis.
Authors
Affiliations (4)
Affiliations (4)
- Department of Vascular Surgery, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, 100730, China.
- 4+4 Medical Doctor Program, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, 100730, China.
- Department of Vascular Surgery, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, 100730, China. [email protected].
- Department of Vascular Surgery, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, 100730, China. [email protected].
Abstract
Atherosclerosis in the carotid artery significantly contributes to embolic events leading to ischemic stroke. Precise identification of unstable carotid plaques through non-invasive imaging, is pivotal for stroke prevention. Artificial intelligence (AI) has demonstrated promise in enhancing the accuracy of plaque risk stratification. This review aims to assess the diagnostic performance of AI algorithms in distinguishing unstable carotid plaques from stable plaques using medical imaging. We conducted comprehensive searches in Medline, Embase, Web of Science, IEEE, PubMed, and the Cochrane Library up to June 6, 2023. Eligible studies included those that utilized AI algorithms for identifying unstable carotid plaques from medical images. Binary diagnostic accuracy metrics, including sensitivity, specificity, and Area Under the Curve (AUC), were extracted. QUADAS-AI was used to assess risk of bias of the included studies. Among the 31 studies reviewed, 14 were subjected to meta-analysis, revealing a pooled sensitivity of 91% (95%CI: 86 - 95%), specificity of 84% (79 - 89%), and AUC of 0.94 (0.91 - 0.95). However, only one study reported external validation, limiting the generalizability of these findings, and substantial heterogeneity was observed (I<sup>2</sup> > 90%). Subgroup analyses indicated performance variations based on factors such as sample size, type of AI algorithms (machine learning or deep learning), segmentation methods (manual or automatic), and publication year. Despite observed publication bias and study heterogeneity, the findings underscore the promise of AI-driven approaches in carotid plaque risk stratification. AI algorithms demonstrated favorable diagnostic performance in identifying unstable carotid plaques. Future research should focus on rigorous validation, ensuring generalizability, and enhancing the explainability of AI algorithms to facilitate their translational use. Not applicable.