The prognostic value of automated coronary calcium derived by a deep learning approach on non-ECG gated CT images from ⁸²Rb-PET/CT myocardial perfusion imaging

Background: Assessment of both coronary artery calcium(CAC) scores and myocardial perfusion imaging(MPI) in patients suspected of coronary artery disease(CAD) provides incremental prognostic information. We used an automated method to determine CAC scores on low-dose attenuation correction CT(LDACT) images gathered during MPI in one single assessment. The prognostic value of this automated CAC score is unknown, we therefore investigated the association of this automated CAC scores and major adverse cardiovascular events(MACE) in a large chest-pain cohort.

Method: We analyzed 747 symptomatic patients referred for ⁸²RubidiumPET/CT, without a history of coronary revascularization. Ischemia was defined as a summed difference score≥2. We used a validated deep learning(DL) method to determine CAC scores. For survival analysis CAC scores were dichotomized as low(<400) and high(≥400). MACE was defined as all cause death, late revascularization (>90 days after scanning) or nonfatal myocardial infarction. Cox proportional hazard analysis were performed to identify predictors of MACE.

Results: During 4 years follow-up, 115 MACEs were observed. High CAC scores showed higher cumulative event rates, irrespective of ischemia (nonischemic: 25.8% vs 11.9% and ischemic: 57.6% vs 23.4%, P-values <0.001). Multivariable cox regression revealed both high CAC scores (HR 2.19 95%CI 1.43-3.35) and ischemia (HR 2.56 95%CI 1.71-3.35) as independent predictors of MACE. Addition of automated CAC scores showed a net reclassification improvement of 0.13(0.022-0.245).

Conclusion: Automatically derived CAC scores determined during a single imaging session are independently associated with MACE. This validated DL method could improve risk stratification and subsequently lead to more personalized treatment in patients suspected of CAD.