Shape-informed cardiac mechanics surrogates in data-scarce regimes via geometric encoding and generative augmentation
#cardiac mechanics #surrogate models #geometric encoding #generative augmentation #machine learning #computational cardiology #neural networks #data-scarce regimes
📌 Key Takeaways
- Researchers developed a two-step framework for cardiac mechanics surrogates that work with limited data
- The approach decouples geometric representation from physics response learning
- The framework uses neural networks to encode heart geometries and predict ventricular displacement
- The method shows strong generalization across diverse anatomies and robustness to noisy inputs
📖 Full Retelling
A team of researchers led by Davide Carrara and including Marc Hirschvogel, Francesca Bonizzoni, Stefano Pagani, Simone Pezzuto, and Francesco Regazzoni developed a novel two-step framework for shape-informed cardiac mechanics surrogates, submitting their findings to arXiv on February 23, 2026, to address the challenge of creating computational heart models that can work effectively with limited data across diverse anatomies. The research addresses a significant challenge in computational cardiology: while high-fidelity models of cardiac mechanics provide valuable insights into heart function, they are computationally too expensive for routine clinical use. Surrogate models can accelerate these simulations, but they struggle to generalize across different anatomical variations, particularly in medical settings where patient data is limited. The researchers' innovative approach decouples the geometric representation from the physics response learning, enabling more effective modeling under data-scarce conditions. The framework operates in two main stages: first, a shape model learns a compact latent representation of left ventricular geometries, effectively encoding anatomical variations and enabling the generation of synthetic geometries for data augmentation; second, a neural field-based surrogate model, conditioned on this geometric encoding, is trained to predict ventricular displacement under external loading. The researchers tested their approaches on both idealized and patient-specific datasets, demonstrating that their methods allow for accurate predictions and generalization to unseen geometries, while maintaining robustness to noisy or sparsely sampled inputs.
🏷️ Themes
Medical AI, Computational modeling, Data augmentation
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
--> Computer Science > Machine Learning arXiv:2602.20306 [Submitted on 23 Feb 2026] Title: Shape-informed cardiac mechanics surrogates in data-scarce regimes via geometric encoding and generative augmentation Authors: Davide Carrara , Marc Hirschvogel , Francesca Bonizzoni , Stefano Pagani , Simone Pezzuto , Francesco Regazzoni View a PDF of the paper titled Shape-informed cardiac mechanics surrogates in data-scarce regimes via geometric encoding and generative augmentation, by Davide Carrara and 5 other authors View PDF HTML Abstract: High-fidelity computational models of cardiac mechanics provide mechanistic insight into the heart function but are computationally prohibitive for routine clinical use. Surrogate models can accelerate simulations, but generalization across diverse anatomies is challenging, particularly in data-scarce settings. We propose a two-step framework that decouples geometric representation from learning the physics response, to enable shape-informed surrogate modeling under data-scarce conditions. First, a shape model learns a compact latent representation of left ventricular geometries. The learned latent space effectively encodes anatomies and enables synthetic geometries generation for data augmentation. Second, a neural field-based surrogate model, conditioned on this geometric encoding, is trained to predict ventricular displacement under external loading. The proposed architecture performs positional encoding by using universal ventricular coordinates, which improves generalization across diverse anatomies. Geometric variability is encoded using two alternative strategies, which are systematically compared: a PCA-based approach suitable for working with point cloud representations of geometries, and a DeepSDF-based implicit neural representation learned directly from point clouds. Overall, our results, obtained on idealized and patient-specific datasets, show that the proposed approaches allow for accurate predictions and generalization...
Read full article at source