AI-CPA Plaque Assessment Results Consistent, Reproducible Among Observers

Inter-observer and intra-observer results of AI-informed coronary computed tomography angiography (CCTA) derived coronary plaque volume quantification (AI-CPA) are consistent with limited variation, indicating potential reproducibility, according to a recent study.1

Presented at the 21st Annual Scientific Meeting of the Society for Cardiovascular Computed Tomography (SCCT) in San Diego, California, by Biyanka Jaltotage, MD, a cardiologist at Fiona Stanley Hospital, the present study sought to optimize quantification of coronary plaque volume, given its potential to aid in monitoring disease progression and assessing the effectiveness of given therapies.1

“Reproducible results are important because we are now applying plaque analysis in patient management decisions and some clinical trials,” Ron Blankstein, MD, the associate director of the cardiovascular imaging program, director of cardiac computed tomography, and co-director of the cardiovascular imaging training program at Brigham and Women’s Hospital, as well as a professor of medicine at Harvard Medical School, told HCPLive in an exclusive interview. “We may, in the future, also think about the concept of comparing the amount of plaque for a particular person over time, perhaps to see if a treatment response is occurring. And for all those concepts, having a reproducible measure is really important.”

Jaltotage and colleagues enrolled patients from an on-label, post-market, multi-center, retrospective registry, with the intra- and inter-observer variation of 27 CCTAs assessed for AI-CPA. Each CCTA was analyzed 3 times each by 3 operators in a blinded setting. The coefficient of variation was used to compare relative variability across multiple measurements for inter- and intra-reader variability. Fleiss’ kappa was also assessed for agreement of the 243 total analyses to categorize patients into total plaque volume (TPV) stages – 0, 1-100, 101-250, 251-750, and ≥751.1

A total of 27 participants were enrolled in the study, among whom the mean age was 61 +/- 10 years and 63% were males. Mean body mass index was 27.8 +/- 4.6, and 64% of patients had obstructive coronary artery disease, defined as ≥50% stenosis.1

Jaltotage and colleagues computed plaque volume and compared it across the 243 independent AI-CPA analyses. The median value of TPV was 295 mm3 (interquartile range [IQR], 137-450), median calcified plaque volume (CPV) was 25 mm3 (IQR, 3-112), median noncalcified plaque volume (NCPV) was 238 mm3 (IQR, 124-384), and median low attenuation plaque volume (LAPV) was 0 (IQR, 0-0).1

Interobserver variability for TPV, CPV, and NCPV was 3.1%, 5.7%, and 3.9%, respectively. Investigators cited this as excellent consistency between each reader. Intra-observer quantification of TPV, CPV, and NCPV was 5.3%, 6.4%, and 5.9%, respectively, indicating good intra-reader reproducibility through each measurement. Fleiss’ kappa for agreement among the 243 analyses to categorize patients into TPV stages was 0.95.1

Ultimately, Jaltotage and colleagues concluded that these data indicate excellent inter-reader and intra-reader agreement and consistency across all 243 analyses with AI-CPA. While these results are positive, Blankstein also noted further research that he believes must be done before implementation.1

“The other thing is to see whether changes in plaque over time are predicted. There’s certainly some data suggesting that, but we need more data in that space to tell us that if we know that plaque is progressing, that’s a higher risk patient,” Blankstein said. “On the other hand, if we know that the plaque is stable and not progressing, we need to know that that’s a lower risk phenotype.”

Editors’ Note: Blankstein reports disclosures with Amgen, Heartflow, Nanox AI, Novartis, Caristo Diagnostics, Siemens, and others.

References

1. Mengesha B, Jaltotage B, Fonte T, et al. Reproducibility of AI-Informed Coronary Computed Tomography Angiography-Derived Coronary Plaque Volume Quantification. Abstract presented at the 21st Annual Scientific Meeting of the Society of Cardiovascular Computed Tomography (SCCT), San Diego, CA. July 9-12, 2026.