Werf, Niels R. van der

A vendor-neutral Agatston score enabled consistent aortic valve calcium quantification across CT systems and improved identification of severe aortic stenosis in patients with discordant echocardiographic findings.

Introduction: This study aimed to develop and evaluate a deep learning (DL) deblurring algorithm for computed tomography (CT). Phantom testing assessed whether DL deblurring could improve CT image quality and coronary artery calcium (CAC) detectability. Materials and Methods: The algorithm consisted of 2 sequential convolutional neural networks (CNNs): one for denoising and one for deblurring. After deblurring, the estimated noise from the first model was reintroduced to preserve image texture. The CNN denoising model was trained to map high-noise images to routine low-noise ones. The CNN deblurring model was trained using natural images—blurred inputs paired with sharp targets. To assess performance, an anthropomorphic phantom with 100 small calcifications was scanned on a CT system using a CAC scoring protocol. Data were acquired at clinical and high doses, then reconstructed with and without DL deblurring, using 25% of the maximum denoising strength. Detectability was defined as the ability to calculate an Agatston score (≥3 adjacent voxels >130 HU). Results: At high dose, CAC detectability increased from 39% (standard reconstruction) to 48% with deblurring. Similar detectability (39%) was achieved at the routine dose with deblurring, representing a 6% improvement compared with the original routine dose data. Discussion: Quantitative results from phantom testing demonstrate that DL deblurring can improve image quality, based on a clinically relevant, task-based CAC detection metric at clinical dose levels.

Abstract Background Coronary artery calcium (CAC) scoring improves event prediction of coronary heart disease (CHD) and can be used to guide initiation or deferral of statin therapy in asymptomatic individuals at low-to-intermediate risk. However, there is substantial variation of CAC scores acquired with different CT scanners. Therefore, CAC scoring discrepancies may lead to suboptimal patients' treatment and harmonization is needed. Purpose The aim of our study was twofold. Our first aim was to develop a calibration tool resulting in vendor neutral Agatston Score (vnAS). Second, we aimed to assess the effect of using this calibration tool in the existing Multi-Ethnic Study of Atherosclerosis (MESA) study cohort on both risk prediction of coronary heart disease (CHD), and initiation of statin therapy. Methods Two static anthropomorphic phantoms containing multiple CAC inserts were imaged on seven different CT systems and one EBT system. For each CT system, the vnAS calculator parameters were derived using regression analysis based on Agatston scores from EBT and CT. To validate our vnAS, we used CAC scoring information and clinical data of participants from MESA study. All included participants (Cohort I, n=3181) were assigned to one out of four calcium groups, which were defined as zero-calcium (AS and vnAS of 0), low-calcium (AS and vnAS of 1–99) high-calcium (AS and vnAS ≥100), and reclassified individuals (AS <100 and vnAS ≥100). The occurrence of CHD events in each group was assessed and multivariable Cox regression models were used to assess the added value of the vnAS in the prediction of CHD events For a sub-cohort of 890 participants at intermediate cardiovascular risk, the potential benefit from statin therapy was estimated based on the number needed to treat (NNT). NNT were determined for patients with vnAS ≥100 and vnAS ≥300 with original AS below 100 (group I) or 300 (group II), respectively. Results For all CT systems, a high degree of correlation with EBT Agatston scores was shown (R2 >0.932). Using the vnAS, 85 individuals (2.7%) were reclassified from a lower to a higher risk category. For the reclassified participants, CHD event rates increased significantly from 7% to 15% (p=0.008) with a CHD hazard ratio of 3.39 (95% CI 1.82–6.35, p=0.001) (Figure 1). In intermediate risk cohort, the NNT was smaller for reclassified individuals as compared to their original (low) calcium group, at 7 vs 12 for group I and 2 vs 15 for group II, respectively. The summary of vnAS applicability is depicted in Figure 2. Conclusions We developed a calibration tool, which enables to calculate vendor neutral AS (vnAS). Based on vnAS, 85 individuals from MESA study, who were reclassified from a lower to a higher calcium category, did indeed have a higher CHD event rate. Consequently, the potential benefit from statin therapy, based on vnAS, was also increased for reclassified participants at intermediate cardiovascular risk. Funding Acknowledgement Type of funding sources: None.