CT-Based Deep Learning Assessment of Aortic Valve Fibrosis and Calcification: Development and Multicenter External Validation Against Experts, Histology, and Outcomes.
Authors
Affiliations (20)
Affiliations (20)
- Departments of Biomedical Sciences and Medicine, Cedars-Sinai Medical Center, Biomedical Imaging Research Institute, Los Angeles, California, USA; 1st Department of Cardiology, Medical University of Warsaw, Warsaw, Poland.
- Departments of Biomedical Sciences and Medicine, Cedars-Sinai Medical Center, Biomedical Imaging Research Institute, Los Angeles, California, USA.
- Department of Interventional Cardiology and Angiology, Institute of Cardiology, Warsaw, Poland.
- Departments of Biomedical Sciences and Medicine, Cedars-Sinai Medical Center, Biomedical Imaging Research Institute, Los Angeles, California, USA; Department of Cardiology, Universitair Ziekenhuis Brussel (UZ Brussel), Vrije Universiteit Brussel (VUB), Brussels, Belgium.
- Department of Cardiac Surgery and Transplantology, Poznan University of Medical Sciences, Poznan, Poland.
- Department of Cardiology, Wroclaw Medical University, Wroclaw, Poland.
- Division of Cardiology and Structural Heart Diseases, Medical University of Silesia, Katowice, Poland.
- 2nd Department of Radiology, Medical University of Gdansk, Gdansk, Poland.
- 1st Department of Cardiology, Medical University of Warsaw, Warsaw, Poland.
- Department of Cardiac Surgery, Medical University of Gdansk, Gdansk, Poland.
- Department of Cardiac Surgery, Medical University of Lodz, Lodz, Poland.
- Division of Medical Sciences, Wroclaw Medical University, Wroclaw, Poland.
- Department of Cardiac Surgery, Medical University of Silesia, Katowice, Poland.
- Department of Cardiology, Poznan University of Medical Sciences, Poznan, Poland.
- Department of Cardiology and Angiology, Medical Center-University of Freiburg, Freiburg, Germany.
- Division of Cardiology and Angiology II, University Heart Centre Freiburg-Bad Krozingen, Bad Krozingen, Germany.
- Discipline of Cardiology, Saolta Healthcare Group, University of Galway, Royal Wolverhampton NHS Trust, Galway, United Kingdom.
- BHF Centre for Cardiovascular Science, University of Edinburgh, Edinburgh, United Kingdom.
- Cedars-Sinai Smidt Heart Institute, Cedars-Sinai Medical Center, Los Angeles, California, USA.
- Departments of Biomedical Sciences and Medicine, Cedars-Sinai Medical Center, Biomedical Imaging Research Institute, Los Angeles, California, USA. Electronic address: [email protected].
Abstract
Computed tomography angiography (CTA) enables assessment of aortic valve tissue composition, providing insights into disease progression and helping predict outcomes in patients undergoing transcatheter aortic valve replacement for severe aortic stenosis (AS). The authors aimed to develop a deep learning system for CTA-based quantification of aortic fibrotic and calcific tissue, validate its agreement with expert manual analysis and histopathology, and evaluate its value for outcome prediction. This international, multicenter study included 1,674 patients who underwent CTA at 9 sites. A deep learning model based on the nnU-Net architecture was trained in 468 patients and applied to an external test set of 1,206 patients, including 29 with valve histology and 12 with serial CTA. In 1,165 patients with clinical outcomes, prognostic value was assessed adjusting for clinical, echocardiographic, and procedural covariates. Deep learning-enabled tissue composition analysis was rapid (1.35 ± 0.4 min vs 7.2 ± 1.2 min for experts) with excellent agreement vs expert assessment (r = 0.915; intraclass correlation coefficient [ICC] = 0.913), histology (r = 0.909; ICC = 0.905), and scan-rescan reproducibility (r = 0.999; ICC = 0.999). In the external cohort, 469 patients (40%) died over 4.1 ± 1.1 years of follow-up. Fibrotic tissue volume (adjusted HR [aHR] 1.42 per IQR increase; P < 0.001), calcific tissue attenuation (aHR 1.02 per 25 HU increase; P = 0.005), left ventricular ejection fraction (aHR 1.10 per 5% decrease; P < 0.001), life-threatening bleeding (aHR: 1.96; P = 0.003), major vascular complication (aHR: 1.51; P = 0.042) and postprocedural peak pressure gradient (aHR 1.03 per 1 mm Hg increase; P < 0.001) independently predicted mortality. Patients in the highest tertile of fibrotic tissue volume (>0.940 cm<sup>3</sup>) had a 2-fold higher risk of death (aHR: 2.0; P < 0.001), whereas patients with dense calcific tissue attenuation (>885 HU) had a 1.5-fold higher risk of death. Deep learning enables rapid quantification of aortic valve fibrotic and calcific tissue from CTA, demonstrating excellent agreement with both expert CTA and histological assessments. Quantitative analysis of aortic tissue composition provides additional prognostic value for all-cause mortality after transcatheter aortic valve replacement in multicenter external evaluation.