JMIR Publications

ABSTRACT

Background:

Artificial intelligence (AI)-enhanced imaging techniques show promise in diagnosing extracranial carotid plaques, a key cardiovascular and cerebrovascular risk factor, a systematic assessment of their diagnostic accuracy remains lacking.

Objective:

This study quantitatively assesses the diagnostic efficacy of deep learning (DL) and radiomics models in diagnosing extracranial carotid plaques, to establish a standardized framework for improving plaque detection.

Methods:

We screened PubMed, Embase, Cochrane, Web of Science, and Institute of Electrical and Electronics Engineers (IEEE) databases published before December 11, 2024. Studies quality were assessed using Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) and AI-specific (QUADAS-AI). A meta-analysis was conducted using StataMP 27.0 with a bivariate mixed-effects model to calculate pooled sensitivity (SE) and specificity (SP), generate summary receiver operating characteristic (SROC) curves, assess I²-based heterogeneity, and conduct subgroup analyses.

Results:

Among 34 studies, 25 on DL models, 8 on machine learning (ML) models (1 assessing both), and 2 evaluating models combining both approaches. The meta-analysis revealed pooled SE, SP, and area under the SROC curve (SROC AUC) from the 34 included studies (with 59 contingency tables) of 0.89, 0.91, and 0.96, respectively. For DL and ML models were 0.90 and 0.88, 0.91 and 0.90, and 0.96 and 0.95, respectively.

Conclusions:

Radiomics and DL models show promise in diagnosing extracranial carotid plaque, with DL models potentially offering slightly higher SE and SP than ML models. In the future, these results deserve further clarification. Clinical Trial: not applicable