Can AI Assessment of CT Attenuation Correction Mapping Be a Viable Alternative for CAC Risk Stratification?

An automated deep learning model for evaluating coronary artery calcium (CAC) burden may provide a low-dose alternative to ECG-gated calcium-scoring computed tomography (CSCT), according to a new study involving nearly 1,100 patients undergoing myocardial perfusion PET/CT.

For the retrospective study, presented at the Society of Nuclear Medicine and Molecular Imaging (SNMMI) conference, researchers assessed the the aforementioned deep learning model to evaluate low-dose, non-gated CT attenuation correction (CTAC) maps for CAC risk stratification. Employing Agatston scores derived from CSCT, the study authors noted the 1,088-patient cohort was comprised of 296 patients with low-risk “0” CAC scoring (27 percent), 247 patients with CAC scoring of 1-100 (23 percent), 181 people with CAC scoring of 101-400 (17 percent) and 364 patients with > 400 CAC scoring (33 percent).

For patients with low-risk CAC scoring, the deep learning model offered an 83.5 percent AUC, according to the study authors. The model provided an 88.9 percent AUC in the high-risk CAC scoring cohort. The researchers pointed out that automated deep learning evaluation of CTAC maps alone identified 69.3 percent of people with high-risk CAC.

“By enabling automated identification of high-risk CAC burden without additional radiation or imaging time, the proposed approach advances opportunistic cardiovascular risk assessment and redefines the clinical utility of CTAC within hybrid PET/CT imaging,” emphasized Hao and colleagues,” noted lead study author Jinkui Hao, PhD, who is affiliated with the radiology department at Northwestern University, and colleagues.

(Editor’s note: For additional content from the SNMMI conference, click here.)

However, the study authors cautioned that the deep learning model had only a 64.5 percent AUC for those with CAC scoring between 1 and 100.

“ … (This was) likely due to motion-induced blurring of small calcifications and confounding adjacent non-coronary calcium,” suggested Hao and colleagues.

Reference

1. Hao J, Avery R, Leonard S, Weinberg R, Shah NS, Zhou B.Coronary artery calcium stratification on low-dose CT attenuation correction maps during PET/CT imaging via deep learning. Presented at the Society for Nuclear Medicine and Molecular Imaging (SNMMI) conference, May 30-June 2, 2026. Available at: https://snmmi.org/AM/AM/Home.aspx .

(Editor’s note: For related content, see “Five Takeaways from New Consensus Recommendations for CT Imaging and Reporting in Patients with CAD,” “Nine Takeaways from New Guidelines on PET Myocardial Perfusion Imaging” and “Can Deep Learning Provide a CT-Less Alternative for Attenuation Compensation with SPECT MPI?”)