AI-enabled cardiac chamber volumetry from non-gated chest CT versus echocardiography for prediction of heart failure and atrial fibrillation: an AI-CVD study
R Mirjalili, K Atlas, C Zhang, A Reeves, A Azimi, M MozafaryBazargany, A Hashemi, A Ghaffari Jolfayi, H Mohammadi, H Ghoshouni, H Zarei, Z Fayad, D Yankelevitz, N Wong, M NaghaviAbstract
Background
Approximately 19 million non-gated chest computed tomography (CT) scans are performed annually in the United States. Artificial intelligence enables opportunistic extraction of cardiovascular phenotypes from these scans. AI-derived left atrial (LA) and left ventricular (LV) volumes from non-gated chest CT predict atrial fibrillation (AF) and heart failure (HF), but direct comparison with echocardiography has not been systematically evaluated.
Purpose
To compare AI-enabled cardiac chamber volumetry from non-gated chest CT with echocardiography for detection and prediction of HF and AF.
Methods
We analyzed 1,675 participants from the Multi-Ethnic Study of Atherosclerosis who underwent both non-gated chest CT and echocardiography at Exam 6 (2016–2018; mean age 73.6±8.4 years; 44.6% male). AI-CVD cardiac chambers volumetry automatically segmented all four chambers and LV mass on non-contrast CT. LA and LV volumes were indexed to body surface area (LAVI, LVVI). Echocardiographic volumes were derived using the biplane Simpson method and similarly indexed. Discrimination for prevalent HF and AF was assessed using AUC. Incident events were evaluated using age- and sex-adjusted Cox models with comparison of C-indices.
Results
At baseline, 158 participants had AF and 37 had HF. During follow-up (median 2.6 years for HF; 1.6 years for AF), 16 HF and 49 AF events occurred. LVVI and LAVI were higher on CT than echocardiography (LVVI 53.0±12.1 vs 41.0±8.8 cc/m²; LAVI 36.2±13.3 vs 28.2±9.2 cc/m²). For prevalent HF, CT-derived LVVI showed comparable discrimination (AUC 0.76 vs 0.70; p=0.12). For prevalent AF, CT-derived LAVI demonstrated superior discrimination (AUC 0.77 vs 0.72; p=0.01). For incident HF, C-indices were 0.82 (CT) vs 0.79 (echo; p=0.36). For incident AF, C-indices were identical (0.73 vs 0.73; p=0.66). Risk gradients across tertiles were similar between modalities.
Conclusion
AI-enabled opportunistic cardiac chamber volumetry from non-gated chest CT provides discrimination for prevalent and incident HF and AF comparable to echocardiography. Routine chest CT, without contrast or ECG gating, may serve as a scalable cardiovascular screening and triage tool, extending risk assessment beyond coronary artery calcium and enabling population-level detection of subclinical structural heart disease without additional imaging burden.