Lower Attenuation and Higher Kurtosis of Coronary Artery Calcification Associated with Vulnerable Plaque – An Agatston Score Propensity-matched CT Radiomics study

Abstract

Abstract Background: Coronary artery calcification (CAC) burden assessed by Agatston score (AS) is currently recommended to stratify patients at risk for future acute coronary syndrome (ACS). Besides the CAC burden, the biostructure of CAC may also play a vital role in the vulnerability of CAC, which CT radiomics could reveal. Propensity-score matching of the traditional risk factors and CAC burden between the ACS and asymptomatic groups could radically remove biases and allow exploration of characteristic features of CAC in ACS. Methods: We retrospectively identified 77 patients with ACS who had a CAC scan before percutaneous coronary intervention between 2016–2019. These 77 patients were one-to-two propensity-score matched for traditional risk factors of ACS and AS ranks to select 154 subjects from 2890 asymptomatic subjects. A validation cohort of 30 subjects was also enrolled. Radiomics features of each plaque were extracted and averaged in each person. Conditional logistic regression and area under curve analysis was used for statistical analysis. Results: A higher number of coronary segments involved, lower mean, median, first quartile, and standard deviation of attenuation, and increased kurtosis of attenuation of CAC were associated with the ACS group compared to the control group (p < 0.05 for all). Multivariable analysis showed that the lower median attenuation (OR = 0.969, p < 0.001) and higher Kurtosis (OR = 18.7, p < 0.001) were associated with the ACS group. The median attenuation and kurtosis significantly increase across AS ranks 1 to 4 (p = 0.001). The AUC of kurtosis (0.727) and median attenuation (0.66) were both significantly higher than that of the standard AS (AUC = 0.502) and the number of TRF (AUC = 0.537). The best cut-off of kurtosis at 2.74 yielded an accuracy of 74%, and the cut-off of median attenuation at 196 yielded an accuracy of 68%. The accuracy of kurtosis was 64%, and the accuracy of median attenuation was 55% in the validation cohort. Conclusion: After propensity-matching traditional risk factors and CAC burden, CT radiomics highlighted that lower median attenuation and higher kurtosis were the CAC characteristics of vulnerable plaques. These features improve the understanding of the biomechanics of CAC evolution and enhance the value of CAC scan in ACS risk assessment.