Characteristics and thresholds of coronary lesions to predict acute coronary syndrome from artificial intelligence-guided coronary computed tomography angiography
S Bar, J Knuuti, A Saraste, T Kero, J J Bax, N S Nurmohamed, A J Ruurt, P Knaapen, I Danad, T MaaniittyAbstract
Background
Recent advances with artificial intelligence (AI) have enabled rapid and accurate non-invasive plaque quantification and characterization from coronary computed tomography angiography (CCTA). AI-guided quantitative computed tomography (AI-QCT) is a novel AI-based software solution with approval from the U.S. Food and Drug Administration. However, the characteristics of future culprit-lesions according to AI-QCT are unknown.
Objectives
We aimed to determine the characteristics of future culprit-lesions according to AI-QCT and to derive and validate cut-offs to predict future acute coronary syndrome (ACS).
Methods
Two independent cohorts from two tertiary care centers in Europe were analyzed by AI-QCT on lesion-level; 2148 patients for derivation and 568 patients for validation. The median follow-up duration was 6.8 years. AI-QCT plaque characteristics of future ACS culprit-lesions were assessed in the derivation cohort, and lesion-level plaque volume thresholds were derived to predict future culprit-lesions. These thresholds were then tested for their prognostic value for patient-level ACS in the derivation and validation cohort.
Results
72/79 (91%) culprit-lesions had a matching lesion on index CCTA. Culprit (N=72) vs. non-culprit lesions (N=657) of ACS patients and vs. lesions from non-ACS patients (N=13’046) demonstrated markedly higher median total plaque volume (TPV) (106 vs. 16 vs. 10 mm3), non-calcified plaque volume (NCPV) (65 vs. 10 vs. 8 mm3), and calcified plaque volume (CPV) (24 vs. 3 vs. 1 mm3) (p<0.001 for all) (Figure 1). They depicted ≥50% stenosis in 30.6% vs. 12.8% vs. 6.1% (p≤0.003) and low-attenuation plaque in 30.6% vs. 9.0% vs. 6.5% (p<0.001). Thresholds of TPV ≥54.65 mm3 (area under the curve (AUC) 0.80) and NCPV ≥32.15 mm3 (AUC 0.80) were derived to predict future ACS culprit-lesions with similar performance (p=0.971), while both outperformed CPV in ACS prediction (AUC 0.74, p≤0.049). TPV ≥54.65 mm3 (derivation: HRadj 3.73, 95% CI 1.87-7.43, p<0.001; validation: HRadj 10.56, 95% CI 2.42-45.97, p=0.002) or NCPV ≥32.15 mm3 (derivation: HRadj 5.10, 95% CI 2.25-11.49, p<0.001 (Figure 2 A/B); validation: HRadj 8.27, 95% CI 1.89-36.12, p=0.005 (Figure 2 C/D)) were associated with significantly higher rates of patient-level ACS in the derivation and validation cohort.
Conclusions
91% of future culprit-lesions had a matching lesion on index CCTA by AI-QCT. Future culprit-lesions vs. non-culprit lesions demonstrated significantly higher plaque volumes of all types, were non-obstructive in 69% and had low-attenuation plaque in 30%. Thresholds of TPV ≥54.65 mm3 and NCPV ≥32.15 mm3 were derived and associated with increased rates of ACS in two independent cohorts.Figure 1.Stenosis and plaque volumesFigure 2.Kaplan Meier Curves