Association Between AI-derived Thoracic Calcium Volume and Aortic Valve Calcification Quantified by Agatston Scoring

Abstract

Background/Aim

Aortic valve calcification is a key determinant of aortic stenosis severity. Whether AI-derived thoracic aortic calcium volume reflects valvular calcification remains unclear. This study examined the association between aortic valve calcification quantified using Agatston, volume and mass scores, and total aortic calcium volume measured with an automated AI tool.

Patients and Methods

This retrospective single-center study included 32 patients undergoing computed tomography (CT) for transcatheter aortic valve implantation planning. Aortic valve calcification was quantified on noncontrast CT using Agatston, volume and mass scores. Total aortic calcium volume was derived from contrast-enhanced CT using the AI-Rad Companion Chest CT. Correlation analyses and simple linear regression assessed associations between aortic and valvular calcification metrics.

Results

Mean total aortic calcium volume was 4.38 ml. Mean aortic valve Agatston score was 3307 units and mean valve calcium volume was 2727 mm³. Correlation between total aortic calcium and aortic valve Agatston score was weak and nonsignificant (Pearson r=0.213, p=0.259). Correlation with valve calcium volume was similarly weak (Pearson r=0.219, p=0.244). Regression models showed minimal explained variance (R²≈4-5 percent), and slopes were not statistically significant. No clinically relevant association between thoracic aortic calcium burden and aortic valve calcification was identified.

Conclusion

AI-derived thoracic aortic calcium volume did not correlate with aortic valve calcification. Valvular calcification appears to progress independently of thoracic aortic wall calcification. Automated aortic calcium measurements cannot substitute for direct quantification of aortic valve calcification in preprocedural assessment.

Introduction

Aortic valve stenosis (AVS) is the most prevalent valvular pathology worldwide and is associated with high morbidity and mortality, particularly among elderly patients (1, 2). The predominant form of AVS is calcific AVS (1), which results from lipid infiltration and inflammation of the valve leaflets, followed by progressive fibrosis and calcification (3).

Degenerative calcific aortic valve stenosis shares risk factors with systemic atherosclerosis but there are important biological differences. Ectopic cardiovascular calcification is mediated by mesenchymal cells such as vascular smooth muscle cells, valve interstitial cells and fibroblasts; these cells are able to transdifferentiate into osteoblast-like phenotypes (4). In the normal aortic valve most valve interstitial cells (VICs) display a fibroblast-like phenotype, whereas exposure to biomechanical or biochemical stimuli induces a myofibroblast phenotype and eventually an osteoblast-like phenotype (5). These differences suggest that valvular calcification is not simply a vascular phenomenon but involves distinct cellular pathways. Imaging studies indicate that the prevalence of aortic valve calcification increases sharply with age; in population-based cohorts the incidence rises from 3% under 50 years of age to about one-third of individuals ≥70 years, with a higher prevalence in men (6). The Multi-Ethnic Study of Atherosclerosis reported that calcific aortic valve disease was present in up to 55% of participants aged 80 years or older and that greater aortic valve calcification values predicted progression to severe aortic stenosis (7).A large computed tomography (CT) study of 499 patients reported that aortic valve calcification was present in 29% of the cohort and that the prevalence and degree of valve, coronary and ascending aortic calcification increased with age; however, no correlation was found between valve and coronary or aortic calcification in the whole population, and the association was only significant in subjects over 70 years (6). These findings support the notion that aortic valve and vascular calcification may progress through different mechanisms and at different rates.

Calcific AVS can lead to symptoms such as angina, exertional dyspnea, syncope, and ultimately heart failure (1). The initial steps in diagnosing AVS include physical examination and echocardiography to evaluate valve anatomy - such as the number of cusps, extent of calcification, leaflet excursion - and to assess valve hemodynamics (8). However, echocardiography can confirm the presence of calcific deposits, but it does not provide an accurate quantification of the calcium load (9).

Severe AVS is associated with a poor prognosis and has a negative impact on quality of life (10). Therapeutic options for severe calcific AVS with corresponding symptoms include aortic valve replacement, either through surgical aortic valve replacement (SAVR) or transcatheter aortic valve implantation (TAVI) (3, 11). In recent years, TAVI has emerged as the preferred modality, particularly in elderly or high-risk patients, due to its minimally invasive nature and favorable clinical outcomes (12, 13). A 2025 meta-analysis of younger low-risk patients undergoing TAVI reported lower rates of disabling stroke and shorter hospital stays compared with surgical replacement, supporting the expanding indications for TAVI (14).

Given the limited ability of echocardiography to accurately quantify calcium load, combined with the increasing use of TAVI, the role of CT in assessing aortic valve calcification (AVC) became more important (11). Based on the well-established use of coronary calcium scoring, a comparable approach can be employed to quantify AVC (15). The quantification of coronary artery calcification using the Agatston-Score in non-contrast enhanced CT is a known method in clinical practice for determining the coronary calcium load and is widely accepted as a predictor of future cardiovascular events (16). A similar methodology can be applied to quantify AVC, offering a non-invasive and reproducible tool for evaluating the extent of calcific involvement of the aortic valve (17). Moreover, CT imaging is already an essential component of preprocedural planning for TAVI, as it allows for the determination of the optimal access route, the selection of the appropriate valve prosthesis, and the precise measurement of key anatomical structures (18-20).

The integration of artificial intelligence (AI) continues to be a transformative factor in the field of radiology (21, 22). Recently, advances in AI have provided radiologists with tools promising increasing efficiency and diagnostic accuracy in clinical practice (23, 24). One such tool is the Siemens AI-Rad Companion Chest CT which provides automated volumetric analysis of aortic calcification.

As it is well established that severe AVS is associated with an impaired prognosis and the degree of AVC has been shown to be a strong predictor of clinical outcomes, accurate quantification of AVC is of significant clinical relevance (10, 15, 25). In summary, we consider the correlation between AVC and generalized calcification of the aorta to be a clinically relevant information for the early diagnosis of patients with potential AVS.

In contrast to the wealth of data on coronary or ascending aorta calcification, there have been few studies on the relationship between aortic valve calcification and overall calcium burden of the thoracic aorta. A significant correlation between vale, coronary and thoracic aortic calcification has been previously assessed in two large cohorts, using score-based comparison, but no volumetric analysis.

Since total aortic calcification reflects systemic atherosclerotic burden, accurate correlation with valve calcium may improve risk stratification for TAVI. The present study therefore aimed to examine the association between aortic valve calcification quantified using Agatston, volume and mass scores and the volumetric calcium burden of the entire thoracic aorta.

Patients and Methods

Ethics approval. The study was conducted in accordance with the Declaration of Helsinki and approved by the local institutional ethics committee (EKNZ 2023-02016).

Study population. Assessment of aortic calcification load and the assessment of AVC (determination of Agatston-Score) were used as primary outcome parameters. A high effect size (R²=0.26) was estimated due to the common risk factors. Power analysis with one predictor (aortic calcification load), significance level of 0.05, and power of 0.9 resulted in a sample size of 32.

This retrospective single-center study included 32 patients who underwent chest and abdominal CT as part of preprocedural planning for TAVI between 01/10/2022 and 01/10/2023. A signed general consent was obligatory for all patients included. Patients with prior aortic valve replacement were excluded. The data was retrospectively extracted from the radiology information system (RIS).

CT imaging protocol. CT scans were performed using SOMATOM Drive and SOMATOM X.ceed scanners (Siemens Healthineers AG, Forchheim, Germany). All images had been acquired as part of routine clinical practice, following institutional protocols to ensure consistent patient positioning and exposure parameters. Acquisition parameters for non-contrast scans were 120 kV, collimation 0.6 mm and 3 mm reconstructed slice thickness. Contrast-enhanced scans were acquired at 100 kV with automatic tube current modulation.

Quantification of aortic valve calcification.Native non-contrast scans of the heart were processed with the Syngo.via^® software for multimodality reading (Syngo.via CaScoring, Siemens Healthineers AG 2009-2018, Version 05.01.000.0030). Lesions with ≥130 HU and contiguous voxels ≥1 mm³ were identified. Three metrics were recorded for each patient:

● Volume score (mm³): number of calcified voxels multiplied by voxel volume (26).

● Mass score (mg): mean attenuation of each lesion multiplied by its volume and corrected for a water phantom (26).

● Agatston score (AU): area of each lesion multiplied by a weighting factor based on peak attenuation (26).

Two readers with 2.5 and 15 years of experience in cross-sectional cardiothoracic imaging performed the evaluation of the AVC Agatston score in a consensus reading to minimize interobserver variability. An exemplary assessment of AVC using syngo.via is depicted inFigure 1.

Figure 1.

Quantitative assessment of AVC using syngo.via software indicating heavy aortic valve calcifications.

Quantification of aortic calcification.Total aortic calcification was quantified on contrast-enhanced scans using the AI-Rad Companion Chest CT (Cardiovascular module, VS13, Siemens Healthineers AG) (27). The algorithm automatically segments the thoracic aorta from the sino-tubular junction to the diaphragm and identifies voxels above the calcium threshold. For each patient the software provided: i) Total aortic calcium volume (ml) – the absolute volume of calcified voxels within the aortic wall; ii) Luminal volume (ml) – the aortic blood pool volume; iii) Percentage of calcification (%) – total calcium volume divided by the sum of luminal and wall volumes, expressing relative calcification burden.

Assessment of aortic calcification is depicted in Figure 2. For comparability with valve measurements, valve volumes in mm³ were converted to milliliters by dividing by 1,000. Calcium mass score was not converted, as it reflects both calcium volume and density.

Figure 2.

Quantitative assessment of aortic calcification using syngo.via demonstrating heavy segmental calcifications.

Statistical analysis.Continuous variables are reported as mean±standard deviation or median with inter-quartile range (IQR). Categorical variables are presented as counts and percentages. Correlation between valve and aortic calcification metrics was assessed using the Pearson correlation coefficient for normally distributed variables and the Spearman rank correlation for non-normal distributions. Agreement between native Agatston scores and those estimated from volume was evaluated using intraclass correlation coefficients and Bland-Altman plots as recommended in previous validation studies (28). Linear regression was performed with valve calcification as the dependent variable and total aortic calcification as the independent variable. The goodness of fit was reported using the coefficient of determination (R²). A p-value <0.05 was considered statistically significant. Statistical analyses were carried out using IBM SPSS Statistics (Version 30.0.0.0, IBM, Chicago, IL, USA).

Results

The dataset consisted of 32 complete cases for analysis. For each case, total aortic calcification was defined as the sum of calcium volumes across eight aortic segments. Aortic valve calcification was reported separately as Agatston score and valve calcium volume, respectively. The study population consisted of n=26 men and n=6 women. Mean age was 78±10 years.

Total aortic calcification averaged 4.38±4.59 ml (median=2.92 ml; IQR=1.32-4.20 ml). The aortic valve calcifications showed a mean Agatston score of 3,307±2,134 units (median=2,741; IQR=1,839-4,741). The mean valve calcium volume was 2,727±1,755 mm³ (median=2,243 mm³; IQR=1,530-3,956 mm³).

Pearson and Spearman correlations were used to assess the association between total aortic calcification and aortic valve calcification. Only weak positive associations were observed, and none reached statistical significance. Total aortic calcification and Agatston Score of the aortic valve demonstrated a Pearson r=0.213 (p=0.259) and Spearman ρ=0.296 (p=0.112).

Total aortic calcification and valve calcium volume demonstrated a Pearson r=0.219 (p=0.244) and Spearman ρ=0.311 (p=0.095).

Simple linear regression models indicated that an additional ml of calcium in the aortic wall increased the Agatston score by roughly 98.9 and valve calcium volume by 83.9 mm³. However, these slopes were not statistically significant (p=0.259 and p=0.244, respectively), and the explained variance (R²) was low at ≈ 4-5 %. Scatter plots are depicted in Figure 3.

Figure 3.

Scatter plots with fitted regression lines and 95% confidence bands for each pair of variables demonstrating no statistical significance.

Discussion

This study aimed to explore the relationship between calcific burden along the thoracic aorta and calcification of the aortic valve. Despite measuring total calcium volume across eight aortic segments with high reproducibility, we found no significant association between the total aortic calcification and the aortic valve Agatston score or aortic valve calcium volume. These findings reinforce earlier observations that the pathophysiological drivers of aortic wall calcification and valvular calcification differ (6, 30).

Previous work has shown that aortic valve calcification is a marker of adverse outcomes in aortic stenosis. Marrero et al. reported that AVC did not correlate with coronary or ascending aortic calcification in the overall study cohort, as a significant association only emerged in patients older than 70 years (6). Our study expands upon these findings by using volumetric rather than score-based quantification of aortic wall calcium and by applying an automated segmentation tool. Even with this sensitive approach we observed only weak, non-significant correlations between aortic wall calcium volume and aortic valvular calcification, suggesting that degeneration of the valve leaflets may proceed independently of systemic atherosclerotic burden.

Recent literature emphasizes the clinical relevance of AVC beyond correlation with aortic wall disease. An observational study of 75 patients with severe AS and reduced left-ventricular ejection fraction found AVC volume to be an independent predictor of mortality associated with negative outcomes after TAVI, as higher AVC was linked to reduced baseline left-ventricular function and predicted post-procedural recovery (30). In a larger cohort of 464 patients undergoing TAVI, severe aortic arch calcification carried a markedly higher three-year all-cause mortality (39.6%) compared with no calcification (6.7%) (31). These data highlight the prognostic importance of calcification distribution and support comprehensive calcium assessment when planning interventions.

Advances in imaging technology further underscore the need for precise calcium quantification. A 2025 narrative review summarized that degenerative aortic valve disease is now the third most common cause of heart disease and that calcium scoring on CT serves as a valuable adjunct to echocardiography for grading AS and predicting post-procedural outcomes (32). Modern deep-learning algorithms can automatically quantify AVC on contrast-enhanced CT with excellent agreement to manual Agatston scoring and high accuracy in identifying severe AS, suggesting that automated techniques may soon become routine (33). However, our findings indicate that such automated measures of aortic wall calcification should not be used as a surrogate for valvular calcification.

AI-driven volumetric measurement of aortic calcification may provide a comprehensive assessment of calcific burden along the aorta. In our cohort, aortic wall calcification did not correlate with aortic valve calcification, implying that valvular disease represents a distinct pathobiological process. These results corroborate earlier studies showing minimal interplay between aortic and valvular calcification and emphasize that quantifying aortic wall calcium cannot substitute for direct evaluation of the valve. Future research should validate these findings in larger populations and explore whether integrating AI-derived calcium volumes into prognostic models could improve risk stratification for interventions such as TAVI.

Conflicts of Interest

DB received scientific funding from Guerbet AG Switzerland.The Department of Radiology of Kantonsspital Baden has research agreements with Siemens Healthineers.

Authors’ Contributions

Conceptualization, GGA, DB, MG, SE, AE, RAK, TN; methodology, GGA, DB, FK, TN; formal analysis, DB, MK, FK, SE, MT, TN; resources, RAK, SE, MT; data curation, GGA, DB, MG, FK, TN; writing – original draft preparation, MG, DB, TN; writing – review and editing, GGA, DB, MG, AE, FK, TN; visualization, MG, DB, GGA, TN; supervision, TN, RAK, MT. All Authors have read and agreed to the published version of the manuscript.

Artificial Intelligence (AI) Disclosure

During the preparation of this manuscript, a large language model (ChatGPT 5.2, OpenAI) was used solely for language editing and stylistic improvements in select paragraphs. No sections involving the generation, analysis, or interpretation of research data were produced by generative AI. All scientific content was created and verified by the authors. Furthermore, no figures or visual data were generated or modified using generative AI or machine learning-based image enhancement tools.