TY  - GEN 
T1  - Transmural myocardial blood flow characterization with AI-enabled parametric mapping improves prediction of major adverse cardiac events
A1  - Chevalley, Arthur
A1  - Martin Asiain, Maria
A1  - Klein, Ran
A1  - DeKemp, Robert
A1  - Prior, John O.
A1  - Jreige, Mario
A1  - Kamani, Christel H.
A1  - Moulton, Eric
A1  - Depeursinge, Adrien
Y1  - 2026
KW  - 3D imaging
KW  - Cox Model
KW  - Rubidium PET
KW  - Transmural Perfusion
N2  - Major adverse cardiac events (MACE) are among the leading causes of death worldwide. Quantitative myocardium perfusion imaging based on rubidium (Rb) PET allows measuring myocardial blood flow (MBF) in the left ventricle (LV), which proved to be predictive for MACE. Clinical routine relies on an approximate 3D LV centerline (CL) flattened to a 2D polar map (PM).

Previous studies showed the importance of transmural perfusion. In this work, we hypothesise the importance of transmural MBF in MACE prediction. We propose a fully automated LV segmentation and centreline extraction up to scanner resolution. We further extract perfusion close the epi-/endocardium to investigate the importance of multi-depth MBF for MACE prediction. 

We established a cohort of 234 patients with suspected myocardial ischemia unrolled to undergo [82Rb] cardiac Silicon PhotoMultiplier (SiPM) under approved ethics protocol. Follow-up revealed 47 events on average 572 days (2 to 862 days) after the exam. All patients were processed using FlowQuant to obtain the arterial input functions and three arterial regions. 3D MBF was produced using the model proposed in [1].

To compare the survival prognosis Cox models with elastic net penalty were fitted using the different perfusions, i.e. normed Rb activity (ACT_CL(x)) vs CL MBF (MBF_CL(x)) vs Epi-CL-Endo (ECE) MBF (MBF_ECE(x)), and dichotomised using the median predicted hazard for kaplan-meier analysis (KM). The Cox models were trained using a stratified five-fold approach with a 100-fold bootstrap C-Index estimate pooled to estimate the confidence intervals (CIs).

When comparing the C-Indexes and their CIs, the MBF-based model outperforms the normed activity one. By further including the epicardial and endocardial parts of the LV, the C-Index largely improves, reinforcing the importance of transmural perfusion. Even though the results are very encouraging, further studies with a larger cohort are still required. 

[1] Moulton, E., et al.: Artificial Intelligence-Accelerated Parametric Imaging of Myocardial Blood Flow with Rubidium-82 PET (2024)
ER  -