AbstractQuantifying aortic valve calcification is critical for assessing the severity of aortic stenosis, predicting cardiovascular risk, and guiding treatment decisions. This study evaluated the feasibility of a deep learning-based automatic quantification of aortic valve calcification using contrast-enhanced coronary CT angiography and compared the results with manual calcium scoring. A retrospective analysis of 177 patients undergoing aortic stenosis evaluation was conducted, divided into a development set (n = 97) and an internal validation set (n = 80). The DeepLab v3 + model segmented the ascending aorta, and the XGBoost model refined the aortic valve region using representative attenuation values. Calcifications were identified with a tailored threshold based on these values and quantified using a weighted scoring method analogous to the Agatston score. The automated method showed excellent agreement with manual Agatston scores derived from non-contrast CT (Pearson correlation coefficient = 0.93, 95% confidence interval [CI]: 0.89–0.95, p < 0.001, concordance correlation coefficient = 0.92, 95% CI: 0.87–0.95). For classifying severe aortic stenosis, defined by calcium scores exceeding 2000 for men and 1300 for women, the approach achieved a sensitivity of 88.6%, specificity of 91.1%, and overall accuracy of 90.0%. This deep learning model provides automated aortic valve calcification quantification with high accuracy on enhanced CT. This approach offers an alternative for measuring aortic valve calcium when non-contrast CT is unavailable, with the potential to reduce reliance on non-contrast CT, minimize operator dependency, and lower patient radiation exposure.
IntroductionQuantifying aortic valve calcification is crucial for assessing the severity of aortic stenosis, predicting cardiovascular risk, and guiding appropriate treatment strategies1,2. Aortic stenosis, a condition characterized by the narrowing of the aortic valve opening, can lead to severe cardiac complications if not properly managed3. Accurate measurement of calcification provides a pivotal role in elucidating disease progression and informs clinical decisions regarding interventions such as surgical valve replacement, transcatheter aortic valve implantation (TAVI) or medical therapy4.Traditionally, aortic valve calcification is quantified manually on non-contrast CT images by delineating the valve region, analyzing calcified areas with attenuation values ≥ 130 Hounsfield units (HU), and calculating a weighted Agatston score that incorporates HU thresholds to account for calcification density5 Alternatively, calcification volume, representing the spatial extent of calcified tissue, is measured by summing all pixels above the HU threshold. However, both methods require non-contrast CT, increasing radiation exposure, and the manual process remains time-intensive and operator-dependent, limiting its routine clinical use6.Recent advancements in medical imaging and computational techniques have opened new avenues for more efficient and precise quantification methods. Deep learning has demonstrated significant potential in various medical imaging tasks, including image segmentation and feature extraction7. In this study, we developed a deep learning-based method for the automated quantification of aortic valve calcification, utilizing contrast-enhanced coronary CT angiography (CCTA), which provides detailed imaging of the aortic valve and comprehensive assessment without the need for an additional scan. The aim of this study was to evaluate the feasibility of applying deep learning algorithms to automatically quantify aortic valve calcification from CCTA images in patients with clinically suspected aortic stenosis.MethodsThe Institutional Review Board of Seoul National University Hospital reviewed and approved the retrospective design of this study (IRB No. 2304-151-1428). The need to obtain informed consent was waived by the Institutional Review Board of Seoul National University Hospital. All experiments and methods were performed in accordance with relevant guidelines and regulations.SubjectsThis retrospective study included 177 patients who underwent both calcium scoring CT (CSCT) and CCTA for clinical suspicion of aortic stenosis on the same day at Seoul National University Hospital. The CT images were obtained from scans performed between January 2020 and May 2023. Patients who had undergone aortic valve surgery or intervention, or whose CT image quality was insufficient for accurate analysis, were defined as exclusion criteria for the study. However, no patients met these criteria. The study included a total of 177 patients, divided into a development set (n = 97) for model optimization and an internal validation set (n = 80) for performance evaluation. The demographic and clinical characteristics of the study population are shown in Table 1.Table 1 Demographic and clinical characteristics of the study population.Full size tableCCTA image acquisition and reconstructionCT scans were performed using a third generation dual-source CT scanner (SOMATOM Force; Siemens Healthineers, Forchheim, Germany) or a wide-detector single source CT scanner (Revolution Apex; GE Healthcare). A CSCT scan was performed using prospective ECG-triggering with 70% of the R-R interval protocol as guideline8. CCTA acquisition protocols were described elsewhere in detail9. The tube voltage and tube current were individually determined based on automated kilovoltage peak (kVp) selection software (CARE kV; Siemens Healthineers) and automatic exposure control (Care Dose 4D; Siemens Healthineers) for dual-source CT scanner; 70–120 kVp were used. Reconstruction parameters were set as follows: slice thickness, 0.625–0.75 mm; increments, 0.5–0.7 mm; and a medium soft convolution kernel with iterative reconstruction. The CCTA data were collected in a motion-free manner, typically in the end-systolic or mid-diastolic phases.Manual quantification of aortic valve calcification using coronary calcium CTCalcifications were identified as contiguous pixels > 1 mm2 with CT attenuation > 130 HU in the aortic valve region. The manual Agatston score was calculated as a weighted sum of calcified plaque areas, with weights assigned according to the peak attenuation values: 130–199 HU received a weight of 1, 200–299 HU received a weight of 2, 300–399 HU received a weight of 3, and values ≥ 400 HU received a weight of 4. All CSCT images were sent to a workstation and reviewed by an experienced radiology technician using specialized software (Syngo Calcium Scoring, Siemens Healthineers).Deep Learning-based aortic valve segmentation and calcification quantificationFor the quantification of aortic valve calcification, we used prototype software, CardioLucid-AVC (Version 1.0.0, AI Medic Inc., Seoul, Korea, https://www.aimedic.kr). Notably, this software was not trained on the CCTA images used in this study. The automated segmentation of the aortic valve and the subsequent quantification of aortic valve calcification were performed using the DeepLab v3 + deep learning model in conjunction with the XGBoost model, as illustrated in Fig. 1. The DeepLab v3 + model, pre-trained on a dataset of 103,076 manually segmented thoracic aorta CT images, was applied without any additional training for this study10. The pre-trained XGBoost model extracted a calibrated representative attenuation value from the segmented ascending aorta. Instead of averaging pixel attenuation values, the model utilized these values as inputs to compute a refined attenuation metric. This calibrated value was subsequently employed to delineate the valve region and quantify aortic valve calcification11.Fig. 1An overview of the aortic valve calcium score calculation. The process involves several key steps: (1) aortic segmentation using the DeepLab v3 + model, (2) determination of representative attenuation values with the XGBoost model, (3) incorporation of anatomical location information to extract regions near the annulus, (4) application of a specific Hounsfield unit range for aortic valve detection, (5) implementation of an individualized Hounsfield unit (HU) thresholding approach to address HU inconsistencies during aortic valve calcification extraction, and (6) weighted score determination to compute accurate aortic valve calcium score. The grey ellipses highlight intermediate processes such as segmentation models, parameter adjustments, and refinement methods, while the white boxes represent the key input, output, and sequential steps in the workflow. CCTA = coronary computed tomography angiography; HU = Hounsfield unit.Full size imageInitially, the DeepLab v3 + model segmented the ascending aorta, isolating it from surrounding cardiovascular structures12 (Fig. 2a). The segmentation process then focuses on identifying the region below the annulus using the anatomical landmarks (Fig. 2b). The aortic valve was extracted from the segmented aorta based on its characteristic darker appearance relative to the aorta, with the XGBoost model identifying valve regions with attenuation values ranging from 0.8 to 1.45 times the representative attenuation value of the aorta11. The valve region was further refined through dilation using a 3 × 3 kernel (2 iterations), followed by removal of the original aorta segment, and expansion of the valve region to incorporate surrounding calcium deposits through five additional dilations using the same parameters (Fig. 2c). This approach effectively isolated the valve region while capturing larger calcium deposits for accurate quantification (Fig. 2d). Parameters for valve segmentation in the model-based algorithm were optimized using the development set, ensuring precise delineation of the aortic valve region.Fig. 2Aortic valve segmentation process. (a) The aorta is segmented using the DeepLab v3 + model, highlighting the aorta in green. Using the aorta segmentation results, (b) the region near the annulus, (c) the aortic valve, and (d) the expanded valve region to encompass surrounding calcium deposits were extracted. Figures were created using prototype software, CardioLucid-AVC (Version 1.0.0, AI Medic Inc., Seoul, Korea, https://www.aimedic.kr).Full size imageFollowing segmentation, the next critical step involved extracting calcifications within the segmented area and calculating the corresponding calcium score. This process utilized representative attenuation values derived from the XGBoost model. Unlike traditional non-contrast CT images, which typically apply a fixed HU threshold (e.g., HU 130) to identify calcium, we applied an individualized HU thresholding approach to account for variability in attenuation values of the aortic lumen and calcium caused by differences in kVp and hemodynamics. Calcifications were extracted using attenuation values exceeding 1.45 times the representative attenuation value, with this threshold optimized in a prior study to balance sensitivity and specificity13 (Fig. 3). The extracted calcification was quantified using a method analogous to the Agatston score, involving the assessment of density relative to multiples of the representative attenuation value and assigning weighted scores accordingly: values between 1.45 and 1.55 times the representative attenuation value received a weight of 1, those between 1.55 and 1.65 a weight of 1.5, those between 1.65 and 1.75 a weight of 2, and values of 1.75 or higher a weight of 2.5. These weighting parameters were adapted from a previous study on coronary artery calcium scoring, which employed relative weighting based on representative attenuation values13. In this study, the weighting parameters were further refined using the development set to ensure accurate and consistent quantification.Fig. 3Aortic valve calcification segmentation. Sample images of aortic valve calcification segmentation with (a, d, and g) non-contrast-enhanced calcium scoring CT images, (b, e, and h) contrast-enhanced coronary CT angiography images, and segmentation results highlighting the aortic valve in green and calcification in red. The sample images are shown in (a-c) axial, (d-f) coronal, and (g-i) sagittal views.Full size imageCase by case analysisEach case in the internal validation set was reviewed and categorized by two reviewers in consensus (an experienced cardiovascular board-certified radiologist and an experienced technologist) based on the extent and significance of detection inaccuracies. Cases with no detection error, where all calcified lesions within the aortic valve region were correctly identified without false positives or false negatives, were classified as no error. Insignificant error referred to cases with 1–2 minor inaccuracies, such as small false positives or omissions of calcifications, deemed unlikely to affect clinical interpretation. Cases with three or more detection errors were classified as significant error, as these errors were likely to impact quantitative accuracy or clinical decision-making.For cases with errors, the issues were classified as either detection failure or erroneous detection. Detection failure referred to missed calcified lesions that were visually apparent. Erroneous detection involved false identification of calcium in non-valvular anatomical regions, including the aortic wall, coronary arteries, mitral valve, or areas of high contrast attenuation in the blood lumen. Notably, a single case could exhibit both types of errors, as detection failure and erroneous detection are not mutually exclusive.Statistical analysisThe correlation and agreement of the deep learning-based automatic quantification of aortic valve calcification with manual Agatston scores were assessed using Pearson and concordance correlation coefficients (CCCs) respectively14,15. For agreement analysis, the Bland-Altman plot was employed to assess the bias in the differences between two measurement methods16. The limits of agreement were calculated as the mean difference plus or minus 1.96 times the standard deviation (SD) of the differences, representing the upper and lower limits, respectively. Severe aortic valve calcification was defined by calcium scores exceeding 2000 for men and 1300 for women, based on diagnostic criteria for aortic stenosis as outlined by Pawade et al.17. To assess the performance of the proposed method, we calculated sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy. A p-value of 0.05 or less was considered statistically significant. All analyses were conducted using R version 4.2.0.ResultsCorrelation and agreement analysis of aortic valve calcification quantificationFor the internal validation set, the Pearson correlation coefficient between the automated and manual Agatston scores was 0.93 (95% confidence interval [CI]: 0.89–0.95, p < 0.001), and the CCC was 0.92 (95% CI: 0.87–0.95). These results indicate a strong correlation and high agreement between the automated quantification and the manual reference standard across both datasets (Fig. 4).Fig. 4Pearson correlation coefficient plots comparing deep learning-based aortic valve calcification scores with manual Agatston scores. Correlation plots for the internal validation set (a and b) show the relationship between the manual and predicted scores. A full-range plot (a) and an enlarged view focusing on the lower calcium score ranges (b) are presented. R2 = coefficient of determination.Full size imageBland-Altman analysis of aortic valve calcification quantificationThe Bland-Altman plot was used to analyze the agreement between the manual Agatston scores and the automated predictions (Fig. 5). The mean difference was 77.4 with an SD of 529.9 (95% CI: -38.7 to 193.5). Most data points fell within the limits of agreement, indicating good agreement between the manual and automated calcium quantification methods.Fig. 5Bland-Altman plot assessing the agreement between deep learning-based aortic valve calcification scores and manual Agatston scores. The analysis for the internal validation set shows the difference between the manual and predicted scores (mean difference = 77.4, SD = 529.9, 95% CI = -38.7 to 193.5). CI = confidence interval; SD = standard deviation.Full size imageClassification analysis of aortic valve calcification quantificationFor the classification of severe aortic stenosis, defined by an aortic valve calcium score exceeding 2000 in men and 1300 in women, the deep learning-based method achieved a sensitivity of 88.6%, specificity of 91.1%, PPV of 88.6%, NPV of 91.1%, and an accuracy of 90.0%.Case by case analysisAmong the 80 cases, 48 (60%) had no error, while 25 (31.3%) exhibited only minor discrepancies categorized as insignificant errors. Significant errors were observed in 7 (8.8%) cases. Detection failure was noted in 8 cases, erroneous detection in 24 cases. and both types of errors occurred in 1 case. Erroneous detection involved calcium in the aortic wall in 16 cases, blood pixels in 8 cases, coronary arteries in 4 cases, and the mitral valve in 2 cases.DiscussionThe principal finding of this study were as follows: (a) the aortic valve calcium scores, obtained using deep learning-based segmentation combined with model-based analysis, showed excellent performance with Pearson correlation coefficients of 0.93 (95% CI: 0.89–0.95, p < 0.001), and the CCC was 0.92 (95% CI: 0.87–0.95) compared to manual Agatston score of aortic valve from CSCT, (b) severe aortic valve stenosis, defined by calcium scores exceeding 2000 for men and 1300 for women, was accurately classified by the method, achieving 88.6% sensitivity, 91.1% specificity, 88.6% PPV, 91.1% NPV, and an overall accuracy of 90.0% and (c) the deep learning model successfully detected aortic valve calcium with no error or only minor error in 91.3% of cases on CCTA.The reference standard for quantifying aortic valve calcification involves manual measurement on non-contrast CT images18,19. However, this approach has significant limitations, including additional radiation exposure and the time-consuming nature of manual analysis. To address these challenges, some researchers have attempted manual quantification of aortic valve calcification using contrast-enhanced images20,21,22,23. These studies used fixed HU thresholds (e.g., 500 HU or 800 HU) for manual calcium detection, typically derived from specific datasets and applied to coronary or aortic regions. However, such fixed thresholds may vary depending on the degree of enhancement in the ascending aorta, potentially missing lower levels of calcium in the aortic valve. This limitation can lead to reduced accuracy and inconsistency across different CT scan settings. Moreover, these fixed-threshold approaches are restricted to calculating calcium volume rather than the Agatston score. Alternatively, two studies proposed dynamic threshold methods for manual calcium detection, setting thresholds at 3 or 4 SD above the mean attenuation of the ascending aorta6,24. The calcium values detected using these thresholds were then converted into Agatston scores using regression models6,24. These studies reported high intraclass correlation coefficients (ICC) for Agatston score measurement on contrast-enhanced CCTA: an ICC of 0.897 (95% CI: 0.840–0.934) in one study by Eberhad et al.6 and an ICC of 0.915 (95% CI: 0.786–0.966) in the study by Laohachewin et al.24 In comparison, our study demonstrated strong agreement with CCC of 0.92 (95% CI, 0.87 to 0.95) against CSCT-derived Agatston scores. These findings suggest that our deep learning-generated aortic valve calcium scores offer comparable or improved consistency in quantifying aortic valve calcification.To date, no study has fully automated the quantification of aortic valve calcification using contrast-enhanced CCTA. Accurate automatic quantification of aortic valve calcification from these images requires two key processes: precise segmentation of the aortic valve and accurate calcium extraction from the enhanced aortic lumen. In this study, we developed a method that combines the DeepLab v3 + deep learning model with the XGBoost model to achieve precise aortic valve segmentation. For calcium extraction, we employed an individualized HU-based thresholding method. A representative attenuation value was derived from the segmented ascending aorta and used as the reference for calcium detection. Regions with attenuation exceeding 1.45 times the representative value were identified as calcifications. These calcified regions were assigned weighted scores based on the degree to which their attenuation exceeded the threshold, simulating the Agatston scoring methodology. This individualized approach enabled precise segmentation and quantification, accommodating variations in patient-specific attenuation characteristics and CT acquisition parameters. By simulating the Agatston scoring system, our method not only reproduces the advantages of the traditional approach but also extends its applicability to contrast-enhanced CCTA. Unlike calcium volume measurement, which often underrepresents the clinical significance of dense calcifications, Agatston calcium scoring incorporates the density of calcifications, making it particularly effective for predicting valve stiffness and hemodynamic impairment in aortic stenosis17. This density-weighted approach has proven to be a superior predictor of disease progression and procedural outcomes, such as those observed in TAVI17. Our performance in classifying severe aortic valve stenosis using enhanced CCTA is comparable to a previous study of deep learning-based automatic quantification of aortic valve calcium using non-enhanced gated images, which reported a sensitivity of 94.6%, specificity of 93.2%, and accuracy of 93.8%19. The strong concordance of automated calcium scores with manual Agatston scores reinforces the clinical relevance of this method and its potential to streamline aortic stenosis assessment while reducing operator dependency. Furthermore, the clinical utility of the Agatston score is underscored by its reproducibility and standardization across studies17. These characteristics make our automated approach a promising tool for advancing the precision and efficiency of aortic stenosis evaluation in routine clinical practice.Quantification of aortic valve calcification is of paramount importance, as it serves as a well-established marker for assessing aortic stenosis, its severity, disease progression, and acts as a powerful prognostic indicator of adverse cardiovascular events17. Additionally, accurate quantification of aortic valve calcification is essential in guiding clinical decision-making for therapeutic interventions, such as surgical valve replacement and TAVI. In the context of TAVI, precise calcium quantification within the landing zone is critically important, as calcium burden has been strongly correlated with procedural complications, including paravalvular leakage, conduction disturbances, and even catastrophic outcomes such as aortic annular rupture25. Moreover, further investigations focusing on the automated quantification of calcification in anatomically adjacent structures, such as the left ventricular outflow tract and the aortic root, as well as the precise localization of calcification within specific regions of the aortic valve, would offer significant improvements in optimizing TAVI management and enhancing patient outcomes.However, there are several limitations to this study. First, this study was conducted as a single-center investigation, which may limit the generalizability of the findings. To enhance applicability and ensure reproducibility, future studies incorporating multi-center validation are necessary. Second, our study focused on segmenting the aortic valve region as a single unit within a “one-label” framework, followed by calcium segmentation and scoring specific to the valve. Unlike previous studies that primarily emphasized multi-label segmentation of the aortic valve and its three leaflets, our approach prioritized calcium detection and quantification26,27. This methodology underscores the clinical relevance of calcium scoring, which was validated through comparisons with expert manual annotations. Although precise three-leaflet segmentation was not the focus of our study, incorporating such an approach in future research could further enhance calcium detection and analysis. Third, our model successfully detected aortic valve calcium with no error or only minor error in 91.3% of cases on CCTA, demonstrating strong overall performance. However, some limitations remain, including missed low-attenuation calcium deposits and false identification of non-valvular structures, such as the aortic wall, coronary ostium, mitral valve, or high-attenuation blood pixels, as calcifications. These inaccuracies highlight the need for further refinement to improve sensitivity and specificity in calcification detection.In conclusion, the deep-learning-based automatic quantification of aortic valve calcification using contrast-enhanced CCTA demonstrated excellent performance and high concordance with manual Agatston scores of aortic valve calcium derived from CSCT. This approach offers an alternative for measuring aortic valve calcium on enhanced CT scans when non-contrast CT scans are unavailable, with the potential to reduce reliance on non-contrast CT, improve efficiency, minimize operator dependency, and lower patient radiation exposure.
Data availability
The data set analyzed during the current study are not publicly available due to medical confidentiality but are available from the corresponding author on reasonable request summarized form pending the approval of the IRB.
ReferencesKoos, R. et al. Aortic valve calcification as a marker for aortic stenosis severity: Assessment on 16-MDCT. AJR Am. J. Roentgenol. 183, 1813–1818. https://doi.org/10.2214/ajr.183.6.01831813 (2004).Article
PubMed
MATH
Google Scholar
Cowell, S. J. et al. Aortic valve calcification on computed tomography predicts the severity of aortic stenosis. Clin. Radiol. 58, 712–716. https://doi.org/10.1016/s0009-9260(03)00184-3 (2003).Article
PubMed
MATH
Google Scholar
Yan, A. T. et al. Association between cardiovascular risk factors and aortic stenosis: The CANHEART aortic stenosis study. J. Am. Coll. Cardiol. 69, 1523–1532. https://doi.org/10.1016/j.jacc.2017.01.025 (2017).Article
PubMed
MATH
Google Scholar
Otto, C. M. & Prendergast, B. Aortic-valve stenosis—from patients at risk to severe valve obstruction. N. Engl. J. Med. 371, 744–756 (2014).Article
PubMed
Google Scholar
Williams, M. C. et al. Reporting incidental coronary, aortic valve and cardiac calcification on non-gated thoracic computed tomography, a consensus statement from the BSCI/BSCCT and BSTI. Br. J. Radiol. 94, 20200894. https://doi.org/10.1259/bjr.20200894 (2021).Article
PubMed
Google Scholar
Eberhard, M. et al. Quantification of aortic valve calcification on contrast-enhanced CT of patients prior to transcatheter aortic valve implantation. EuroIntervention 13, 921–927. https://doi.org/10.4244/EIJ-D-17-00060 (2017).Article
PubMed
Google Scholar
Litjens, G. et al. A survey on deep learning in medical image analysis. Med. Image Anal. 42, 60–88. https://doi.org/10.1016/j.media.2017.07.005 (2017).Article
PubMed
MATH
Google Scholar
Azour, L., Kadoch, M. A., Ward, T. J., Eber, C. D. & Jacobi, A. H. Estimation of cardiovascular risk on routine chest CT: ordinal coronary artery calcium scoring as an accurate predictor of Agatston score ranges. J. Cardiovasc. Comput. Tomogr. 11, 8–15. https://doi.org/10.1016/j.jcct.2016.10.001 (2017).Article
PubMed
Google Scholar
Hong, J. H., Park, E. A., Lee, W., Ahn, C. & Kim, J. H. Incremental image noise reduction in coronary CT angiography using a deep Learning-Based technique with iterative reconstruction. Korean J. Radiol. 21, 1165–1177. https://doi.org/10.3348/kjr.2020.0020 (2020).Article
PubMed
PubMed Central
MATH
Google Scholar
Park, D., Park, E. A., Jeong, B. & Lee, W. A comparative analysis of deep learning-based location-adaptive threshold method software against other commercially available software. Int. J. Cardiovasc. Imaging. 40, 1269–1281. https://doi.org/10.1007/s10554-024-03099-7 (2024).Article
PubMed
PubMed Central
MATH
Google Scholar
Chen, T. & Guestrin, C. in Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining. 785–794.Chen, L. C. et al. Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected Crfs. IEEE Trans. Pattern Anal. Mach. Intell. 40, 834–848 (2017).Article
PubMed
MATH
Google Scholar
Lee, J. O., Park, E. A., Park, D. & Lee, W. Deep Learning-Based automated quantification of coronary artery calcification for Contrast-Enhanced coronary computed tomographic angiography. J. Cardiovasc. Dev. Dis.https://doi.org/10.3390/jcdd10040143 (2023).Article
PubMed
PubMed Central
MATH
Google Scholar
Cohen, I. et al. Pearson correlation coefficient. Noise Reduct. Speech Process., 1–4 (2009).Lawrence, I. & Lin, K. A concordance correlation coefficient to evaluate reproducibility. Biometrics, 255–268 (1989).Bland, J. M. & Altman, D. G. Measuring agreement in method comparison studies. Stat. Methods Med. Res. 8, 135–160 (1999).Article
PubMed
MATH
Google Scholar
Pawade, T., Sheth, T., Guzzetti, E., Dweck, M. R. & Clavel, M. A. Why and how to measure aortic valve calcification in patients with aortic stenosis. JACC Cardiovasc. Imaging. 12, 1835–1848. https://doi.org/10.1016/j.jcmg.2019.01.045 (2019).Article
PubMed
Google Scholar
Haensig, M. et al. Aortic valve calcium scoring is a predictor of significant paravalvular aortic insufficiency in transapical-aortic valve implantation. Eur. J. Cardiothorac. Surg. 41, 1234–1240. https://doi.org/10.1093/ejcts/ezr244 (2012). discussion 1240 – 1231.Article
PubMed
Google Scholar
Chang, S. et al. Development of a deep learning-based algorithm for the automatic detection and quantification of aortic valve calcium. Eur. J. Radiol. 137, 109582. https://doi.org/10.1016/j.ejrad.2021.109582 (2021).Article
PubMed
MATH
Google Scholar
Koos, R. et al. Association of aortic valve calcification severity with the degree of aortic regurgitation after transcatheter aortic valve implantation. Int. J. Cardiol. 150, 142–145. https://doi.org/10.1016/j.ijcard.2010.03.004 (2011).Article
PubMed
MATH
Google Scholar
Mauri, V. et al. Predictors of paravalvular regurgitation and permanent pacemaker implantation after TAVR with a next-generation self-expanding device. Clin. Res. Cardiol. 107, 688–697. https://doi.org/10.1007/s00392-018-1235-1 (2018).Article
PubMed
MATH
Google Scholar
Pollari, F. et al. Risk factors for atrioventricular block after transcatheter aortic valve implantation: a single-centre analysis including assessment of aortic calcifications and follow-up. Europace 21, 787–795. https://doi.org/10.1093/europace/euy316 (2019).Article
PubMed
Google Scholar
Ewe, S. H. et al. Location and severity of aortic valve calcium and implications for aortic regurgitation after transcatheter aortic valve implantation. Am. J. Cardiol. 108, 1470–1477. https://doi.org/10.1016/j.amjcard.2011.07.007 (2011).Article
PubMed
MATH
Google Scholar
Laohachewin, D. et al. Quantification of aortic valve calcification in Contrast-Enhanced computed tomography. J. Clin. Med. 13 https://doi.org/10.3390/jcm13082386 (2024).Article
PubMed
PubMed Central
Google Scholar
Milhorini Pio, S., Bax, J. & Delgado, V. How valvular calcification can affect the outcomes of transcatheter aortic valve implantation. Expert Rev. Med. Devices. 17, 773–784. https://doi.org/10.1080/17434440.2020.1789456 (2020).Article
PubMed
Google Scholar
Pak, D. H., Caballero, A., Sun, W. & Duncan, J. S. in 2020 IEEE 17th international symposium on biomedical imaging (ISBI). 1738–1742 (IEEE).Aoyama, G. et al. Automatic aortic valve cusps segmentation from CT images based on the cascading multiple deep neural networks. J. Imaging. 8 https://doi.org/10.3390/jimaging8010011 (2022).Article
PubMed
PubMed Central
MATH
Google Scholar
Download referencesFundingNone.Author informationAuthors and AffiliationsDepartment of Radiology, Seoul National University Hospital, 101 Daehak-ro, Jongno-gu, Seoul, 03080, KoreaBaren Jeong, Whal Lee & Eun-Ah ParkDepartment of Radiology, Seoul National University College of Medicine, Seoul, KoreaWhal Lee & Eun-Ah ParkInstitute of Radiation Medicine, Seoul National University Medical Research Center, Seoul, KoreaWhal LeeDepartment of Clinical Medical Sciences, Seoul National University College of Medicine, Seoul, KoreaDaebeom Park & Whal LeeAI Medic Inc, Seoul, KoreaDaebeom Park, Soon-Sung Kwon, Yoona Song & Yoon A KimAuthorsDaebeom ParkView author publicationsYou can also search for this author in
PubMed Google ScholarSoon-Sung KwonView author publicationsYou can also search for this author in
PubMed Google ScholarYoona SongView author publicationsYou can also search for this author in
PubMed Google ScholarYoon A KimView author publicationsYou can also search for this author in
PubMed Google ScholarBaren JeongView author publicationsYou can also search for this author in
PubMed Google ScholarWhal LeeView author publicationsYou can also search for this author in
PubMed Google ScholarEun-Ah ParkView author publicationsYou can also search for this author in
PubMed Google ScholarContributionsConception and design: E.P. Acquisition of data: D.P., B.J., and W.L. Analysis and interpretation of data: All authors. Writing of the manuscript: D.P. and E.P. Study supervision E.P. All authors have reviewed and approved the final version of the manuscript.Corresponding authorCorrespondence to
Eun-Ah Park.Ethics declarations
Competing interests
Daebeom Park, Soon-Sung Kwon, Yoona Song, and Yoon A Kim are employees of AI Medic Inc. Daebeom Park is also a Ph.D. graduate at Seoul National University of College Medicine. The other authors declare that they have no conflicts of interest.
Additional informationPublisher’s noteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissionsAbout this articleCite this articlePark, D., Kwon, SS., Song, Y. et al. Deep learning based automatic quantification of aortic valve calcification on contrast enhanced coronary CT angiography.
Sci Rep 15, 8472 (2025). https://doi.org/10.1038/s41598-025-93744-5Download citationReceived: 11 October 2024Accepted: 10 March 2025Published: 12 March 2025DOI: https://doi.org/10.1038/s41598-025-93744-5Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard
Provided by the Springer Nature SharedIt content-sharing initiative