AI bests radiologists at computed tomography angiography reads
Artificial intelligence-enabled tools have proven to be more accurate than radiologists when it comes to detecting certain signs of coronary artery disease (CAD).
Computed tomography angiography (CTA) is becoming the modality of choice for many providers tasked with assessing coronary plaques and stenosis—two telltale signs of CAD. Although CTA offers enhanced visualization for radiologists to assess patients’ risk of CAD, the manual process can be time-consuming and cumbersome. What’s more, CTA reads are known to vary between readers, creating issues with inconsistency.
AI has been proposed as a solution to the variable and laborious process of interpreting CTA exams. And while much of the research into its potential for CAD detection and quantification has been positive, different methodologies and sample sizes have produced mixed results.
“Since AI systems can quickly process large amounts of imaging data, they may be able to identify subtle patterns that are difficult for human observers to discover. Recent studies have shown that AI can automatically detect and quantify coronary artery stenosis and calcified plaques,” Long Yuan, with the department of cardiology at Liaoning Provincial People's Hospital in China, and colleagues write. “However, the diagnostic performance of AI systems is still controversial, with results reported differently across studies.”
The team recently pooled data from 17 studies that included nearly 6,000 patients to conduct a meta-analysis on AI in CTA reads. AI’s cumulative performance was compared against that of radiologists.
For coronary artery stenosis of 50% or higher, AI outperformed radiologists, yielding a sensitivity of 0.92, specificity of 0.87 and AUC of 0.96. In comparison, radiologists achieved a sensitivity of 0.85, specificity of 0.84 and AUC of 0.91. AI’s performance remained consistent with stenosis of 70% or higher and for calcified plaque detection.
“The superior diagnostic performance of AI can be attributed to its capability to simultaneously process and analyze large amounts of imaging data while detecting subtle differences in density and morphological changes,” the authors suggest. “Small pathogenic characteristics that the human eye could miss can be found by AI algorithms.”
The group acknowledges that additional prospective assessments are needed, but they signal optimism for AI’s ability to eliminate subjectivity in CTA interpretations.
The study abstract is available here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.