Findings from a new meta-analysis demonstrate that artificial intelligence (AI) can enhance computed tomography angiography (CTA) detection of coronary artery stenosis and calcified plaque in comparison to radiologist assessment.
For the meta-analysis, recently published in Academic Radiology, researchers reviewed data from 17 studies (5,560 total patients) to compare the use of AI and radiologist interpretation of CTA for detecting calcified plaque and coronary artery stenosis > 50 percent.
Artificial intelligence software offered a 96 percent area under the receiver operating characteristic curve (AUC), 92 percent sensitivity and 87 percent specificity for detecting coronary artery stenosis > 50 percent on CTA, according to the meta-analysis authors. In contrast, the authors found that radiologist assessment of CTA images resulted in a 91 percent AUC, 85 percent sensitivity and 84 percent specificity.
When assessing AI for patients with > 70 percent coronary artery stenosis, the researchers noted a 98 percent AUC, 88 percent sensitivity and 96 percent specificity. For the detection of calcified plaques on CTA, AI offered a 98 percent AUC, 93 percent sensitivity and 94 percent specificity, according to the meta-analysis authors.
“Our results indicated that CTA-based AI demonstrates high diagnostic performance in detecting coronary artery stenosis of ≥50% and ≥70%, as well as calcified plaques. The findings consistently show that this advanced technology has comparable or potentially superior performance to radiologists in accurately identifying ≥50% coronary artery stenosis,” wrote lead meta-analysis author Ming Du, M.D., who is affiliated with the Department of Cardiology at the Liaoning Provincial People’s Hospital and Dalian Medical University in Liaoning, China, and colleagues.
Three Key Takeaways
1. AI offers superior diagnostic accuracy for coronary artery stenosis detection. CTA-based AI demonstrated a higher AUC (96 percent vs. 91 percent), sensitivity (92 percent vs. 85 percent), and specificity (87 percent vs. 84 percent) compared to radiologists for detecting ≥50% coronary artery stenosis.
2. Strong performance in detecting severe stenosis and calcified plaque. AI maintained excellent performance for detecting ≥70% stenosis (AUC 98 percent) and calcified plaque detection (AUC 98 percent).
3. Consistency across diverse study conditions. The AI tools showed uniform diagnostic performance across various study designs, AI methods, and patient populations, supporting their potential for broad clinical implementation.
The meta-analysis authors also noted no significant differences with AI interpretation of CTA across studies with varying AI methodologies, study designs and patient populations.
“The uniform performance across regions suggests that the AI systems maintain their efficacy despite potential differences in patient demographics and clinical practices. These findings further validate the widespread clinical applicability of AI in detecting coronary artery stenosis and calcified plaques, highlighting its potential for broad implementation in diverse clinical settings,” added Du and colleagues.
(Editor’s note: For related content, see “FDA Clears AI-Powered 3D CTA Reconstructions for Head and Neck Imaging,” “Can Multimodality AI Enhance CT Detection of Medium Vessel Occlusions?” and “Can Deep Learning Provide a CT-Less Alternative for Attenuation Compensation with SPECT MPI?”)
In regard to limitations with the meta-analysis, the authors acknowledged substantial variability with the imaging protocols and technologies in the reviewed studies, and the exclusion of non-English studies. The meta-analysis authors also noted possible overestimation of the pooled AUC values stemming from the combination of patient-based and image-based analyses.
