Diagnostic accuracy in coronary CT angiography analysis: artificial intelligence versus human assessment

Area of Science:

• Cardiovascular imaging
• Artificial intelligence in medicine
• Medical diagnostics

Background:

• Visual assessment of coronary CT angiography (CCTA) is subjective and time-consuming.
• Reader experience and interobserver variability impact CCTA accuracy.
• A novel algorithm, atherosclerosis imaging quantitative CT (AI-QCT), was developed for coronary stenosis quantification.

Purpose of the Study:

• To evaluate the accuracy of AI-QCT in quantifying coronary artery stenosis.
• To compare AI-QCT performance against human readers and invasive quantitative coronary angiography (QCA).

Main Methods:

• 208 patients with suspected coronary artery disease (CAD) underwent CCTA.
• AI-QCT and blinded readers assessed coronary artery stenosis.
• Comparison with invasive QCA (≥50% stenosis) using area under the curve (AUC) analysis.
• Analysis performed per-patient and per-vessel, stratified by plaque volume.

Main Results:

• AI-QCT demonstrated superior concordance with QCA compared to clinical CCTA assessments.
• Per-patient AUC for AI-QCT was 0.91, outperforming level 3 (0.77) and level 2 (0.79, 0.76) readers.
• Per-vessel AUC for AI-QCT was 0.86, superior to level 2 readers (0.69).
• AI-QCT's advantage was most prominent in patients with high plaque volume.

Conclusions:

• AI-QCT shows superior agreement with invasive QCA compared to clinical CCTA assessments.
• AI-QCT particularly outperformed less experienced (level 2) readers in extensive CAD.
• AI-QCT integration into clinical practice may improve stenosis quantification accuracy in CCTA.