Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated Prediction of Paravalvular Regurgitation before Transcatheter Aortic Valve Implantation

Michele Cannito, Riccardo Renzulli, Adson Duarte, Farzad Nikfam, Carlo Alberto Barbano, Enrico Chiesa, Francesco Bruno, Federico Giacobbe, Wojciech Wanha, Arturo Giordano, Marco Grangetto, Fabrizio D'Ascenzo

TL;DR

The paper addresses predicting paravalvular regurgitation (PVR) after transcatheter aortic valve implantation (TAVI) using preoperative CT, a clinically important yet challenging task. It proposes a 3D convolutional neural network framework trained on isotropic pre-TAVI CT volumes, comparing baselines and domain-specific pretraining (internal coronary calcium scoring) against out-of-domain pretraining (COCA), and evaluating the impact of focusing on heart/aorta regions via segmentation. Domain-specific pretraining yields the best balanced accuracy ($BA\approx71.7\pm3.3\%$), while segmentation-based inputs offer more stable performance with slightly lower accuracy, illustrating a trade-off between discriminative power and robustness. The work demonstrates feasibility for preprocedural risk stratification, enabling more informed valve selection and procedural planning, and highlights avenues for improvement through larger datasets, multimodal data, and attention-guided methods.

Abstract

Severe aortic stenosis is a common and life-threatening condition in elderly patients, often treated with Transcatheter Aortic Valve Implantation (TAVI). Despite procedural advances, paravalvular aortic regurgitation (PVR) remains one of the most frequent post-TAVI complications, with a proven impact on long-term prognosis. In this work, we investigate the potential of deep learning to predict the occurrence of PVR from preoperative cardiac CT. To this end, a dataset of preoperative TAVI patients was collected, and 3D convolutional neural networks were trained on isotropic CT volumes. The results achieved suggest that volumetric deep learning can capture subtle anatomical features from pre-TAVI imaging, opening new perspectives for personalized risk assessment and procedural optimization. Source code is available at https://github.com/EIDOSLAB/tavi.

Automated Prediction of Paravalvular Regurgitation before Transcatheter Aortic Valve Implantation

TL;DR

The paper addresses predicting paravalvular regurgitation (PVR) after transcatheter aortic valve implantation (TAVI) using preoperative CT, a clinically important yet challenging task. It proposes a 3D convolutional neural network framework trained on isotropic pre-TAVI CT volumes, comparing baselines and domain-specific pretraining (internal coronary calcium scoring) against out-of-domain pretraining (COCA), and evaluating the impact of focusing on heart/aorta regions via segmentation. Domain-specific pretraining yields the best balanced accuracy (), while segmentation-based inputs offer more stable performance with slightly lower accuracy, illustrating a trade-off between discriminative power and robustness. The work demonstrates feasibility for preprocedural risk stratification, enabling more informed valve selection and procedural planning, and highlights avenues for improvement through larger datasets, multimodal data, and attention-guided methods.

Abstract

Severe aortic stenosis is a common and life-threatening condition in elderly patients, often treated with Transcatheter Aortic Valve Implantation (TAVI). Despite procedural advances, paravalvular aortic regurgitation (PVR) remains one of the most frequent post-TAVI complications, with a proven impact on long-term prognosis. In this work, we investigate the potential of deep learning to predict the occurrence of PVR from preoperative cardiac CT. To this end, a dataset of preoperative TAVI patients was collected, and 3D convolutional neural networks were trained on isotropic CT volumes. The results achieved suggest that volumetric deep learning can capture subtle anatomical features from pre-TAVI imaging, opening new perspectives for personalized risk assessment and procedural optimization. Source code is available at https://github.com/EIDOSLAB/tavi.
Paper Structure (15 sections, 2 equations, 2 figures, 1 table)

This paper contains 15 sections, 2 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Materials and Methods
  4. Dataset
  5. Preprocessing
  6. Architectures and Pretraining
  7. Loss and Metrics
  8. Model Interpretability
  9. Results
  10. Training Setup
  11. Baseline Experiments
  12. Pretraining Experiments
  13. Segmentation-based Evaluation
  14. Discussion and Conclusion
  15. Compliance with Ethical Standards

Figures (2)

  • Figure 1: Overview of the proposed pipeline and model performance. A) 3D rendering of a representative preoperative CT showing the automatically segmented heart (red) and aorta (blue). B) Grad-CAM activation map.
  • Figure 2: Confusion matrices of the DenseNet model after pretraining and fine-tuning. (Left) Training set. (Right) Test set.