AI-Driven Coronary Artery Calcium Grading Reduces Scoring Time by 81%

In a groundbreaking study presented at the Society of Cardiovascular Computed Tomography (SCCT) annual scientific meeting, researchers have demonstrated that an artificial intelligence (AI)-driven automated grading system for coronary artery calcium (CAC) significantly reduces the time needed for scoring while maintaining accuracy. The study, led by Dr. Harendra Kumar, MBBS, from Dow University of Health Sciences in Pakistan, involved a retrospective analysis of 1,442 patients who underwent non-contrast cardiac computed tomography (CT) at two hospitals in Florida.

The study revealed that AI-driven CAC grading accurately categorized 79% of patients into the same risk group as assessments made by radiologist readers. More impressively, it reduced the average scoring time from 8.6 minutes to just 1.6 minutes, representing a time savings of approximately 81%. This efficiency could potentially alleviate the workload on healthcare professionals, allowing them to focus on more complex cases or patient interactions.

According to the findings, the automated system utilized a deep-learning algorithm based on a Mask Region-Based Convolutional Neural Network (Mask R-CNN). This technology was trained using manually annotated CAC scores provided by expert radiologists, ensuring high precision in correlating AI findings with human evaluations. The study reported an 85% agreement between AI grading and human grading, with statistical significance (p < 0.001).

Dr. Kumar and his team highlighted the critical role of CAC grading in cardiovascular risk stratification, noting that manual evaluations are often labor-intensive and can suffer from inter-reader variability. The introduction of AI into this process aims to enhance clinical integration and productivity, making it a scalable solution for healthcare systems.

In addition to improving grading efficiency, the AI system demonstrated its capability to identify previously unreported high CAC scores (100 or higher) in 14.6% of patients who underwent non-gated chest CT scans. This early risk assessment could be crucial for patient management and preventive care.

The researchers employed the intraclass correlation coefficient and Bland-Altman analysis to evaluate the agreement between AI-derived and manual Agatston scores, finding a mean difference of 2.3 with limits of agreement from -4.1 to 8.7, indicating little bias in AI scores compared to those assigned by radiologists. Inter-reader variability, a common challenge in manual CAC scoring, decreased significantly from 12% (k = 0.79) with manual grading to just 4% (k = 0.92, p < 0.001) with the AI system.

The SCCT recognized this innovative research among its 19th Annual Young Investigator Awards, supported by an educational grant from Canon Medical Systems USA. This acknowledgment underscores the potential impact of AI technologies in enhancing cardiovascular imaging processes and patient outcomes.

As AI continues to evolve in the medical field, its applications may extend beyond CAC grading, potentially transforming various aspects of cardiovascular care. The findings from this study pave the way for future research and integration of AI solutions into routine clinical practice, promising to improve both efficiency and accuracy in cardiovascular risk assessment.