Accuracy of visual estimation of coronary artery calcifications on low-dose CT from oncologic FDG PET/CT

To evaluate the feasibility and accuracy of visual ordinal coronary artery calcium (CAC) scoring on low-dose, non-gated CT scans acquired as part of oncologic PET/CT, using CAC scoring from ECG-gated non-contrast cardiac CT as the reference standard. 210 oncological patients who underwent both PET/CT and single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) with ECG-gated non-contrast cardiac CT within 6 months were retrospectively included. Four independent readers graded image quality and assessed CAC burden using a visual ordinal scoring system (scores ranging from 0 to 12) on non-gated CT images from PET/CT. CAC scores as well as information on presence of myocardial ischemia and/or scars were recorded from the dedicated cardiac diagnostic workup and were used for comparison. Image quality was considered diagnostic in 99.6% of scans. Interreader agreement of ordinal scores was excellent among all four readers (Krippendorff's alpha = 0.95). Non-gated CT demonstrated favorable diagnostic performance for detecting CAC per-patient, with a sensitivity/specificity/positive predictive value/negative predictive value of 92.2%/95%/99.3%/62.1%. Ordinal scores showed a strong correlation with quantitative reference standard CAC scores (Spearman's rho = 0.9, p < 0.001). Receiver operating characteristic (ROC) analysis indicated that ordinal scores effectively differentiated patients based on reference standard CAC score cutoffs of 100, 400, and 1000, with AUC values of 0.96, 0.92, and 0.95, respectively. Additionally, using ordinal scores, ROC analysis yielded an AUC of 0.7 for differentiating patients with and without ischemia/scars. Visual ordinal scoring of CAC burden on low-dose non-gated CT of oncologic PET/CT scans is accurate, effective and strongly correlates with reference standard measurements. The lower negative predictive value indicates that (mild) coronary calcifications are occasionally underestimated.