Improving the accuracy of coronary artery calcium scoring in computed tomography images using a fast nonlocal means noise-reduction algorithm

초록

Purpose: Computed tomography (CT) image-based coronary artery calcium (CAC) scoring is widely used in the field of diagnosis. The noise level, which occurs relatively frequently when acquiring CT images for calcium scoring examinations in obese patients, affects the accuracy of the score. The purpose of this study was to improve the accuracy of CAC scoring by applying a fast nonlocal means (FNLM) noise-reduction algorithm to CT images. Methods: A total of 30 subjects undergoing normal-dose CAC CT were analyzed (mean age, 65.7 years; 12 men and 18 women). Noise parameters were derived from clinical scans of overweight (n = 3) and obese (n = 15) patients and applied to generate virtual overweight and obese images using a Poisson-Gaussian noise model. FNLM was applied at three smoothing levels. The primary endpoint was agreement between FNLM-processed CAC scores and subject-specific normal-dose CAC scores, assessed using absolute score differences, Bland-Altman analysis, and Agatston risk-category concordance. Results: Compared with reference normal-dose CAC scores, virtual overweight and obese images showed CAC overestimation of 27.5% (95% CI: 18.2-36.7%) and 34.0% (95% CI: 22.9-45.1%), respectively. After FNLM processing, the mean CAC difference was reduced to-0.3% (95% CI:- 5.8-5.2%) in the overweight group and 8.1% (95% CI: 1.9-14.3%) in the obese group, with the highest agreement observed at moderate smoothing levels. Risk-category concordance improved without clinically significant reclassification. Conclusion: We demonstrated that the FNLM improves agreement with reference normal-dose CAC measurements and reduces noise-induced quantification variability in overweight and obese patients.

키워드