Physics-informed Neural Networks Enable High Fidelity Shear Wave Viscoelastography across Multiple organs
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912448588546048 |
|---|---|
| author | Yin, Ziying Guo, Yuxi Pu, Jiayi Jiang, Yuxuan Ma, Shiyu Li, Guo-Yang Cao, Yanping |
| author_facet | Yin, Ziying Guo, Yuxi Pu, Jiayi Jiang, Yuxuan Ma, Shiyu Li, Guo-Yang Cao, Yanping |
| contents | Tissue viscoelasticity has been recognized as a crucial biomechanical indicator for disease diagnosis and therapeutic monitoring. Conventional shear wave elastography techniques depend on dispersion analysis and face fundamental limitations in clinical scenarios. Particularly, limited wave propagation data with low signal-to-noise ratios, along with challenges in discriminating between dual dispersion sources stemming from viscoelasticity and finite tissue dimensions, pose great difficulties for extracting dispersion relation. In this study, we introduce SWVE-Net, a framework for shear wave viscoelasticity imaging based on a physics-informed neural network (PINN). SWVE-Net circumvents dispersion analysis by directly incorporating the viscoelasticity wave motion equation into the loss functions of the PINN. Finite element simulations reveal that SWVE-Net quantifies viscosity parameters within a wide range (0.15-1.5 Pa*s), even for samples just a few millimeters in size, where substantial wave reflections and dispersion occur. Ex vivo experiments demonstrate its applicability across various organs, including brain, liver, kidney, and spleen, each with distinct viscoelasticity. In in vivo human trials on breast and skeletal muscle tissues, SWVE-Net reliably assesses viscoelastic properties with standard deviation-to-mean ratios below 15%, highlighting robustness under real-world constraints. SWVE-Net overcomes the core limitations of conventional elastography and enables reliable viscoelastic characterization where traditional methods fall short. It holds promise for applications such as grading hepatic lipid accumulation, detecting myocardial infarction boundaries, and distinguishing malignant from benign tumors. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_03935 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Physics-informed Neural Networks Enable High Fidelity Shear Wave Viscoelastography across Multiple organs Yin, Ziying Guo, Yuxi Pu, Jiayi Jiang, Yuxuan Ma, Shiyu Li, Guo-Yang Cao, Yanping Medical Physics Soft Condensed Matter Biological Physics Tissue viscoelasticity has been recognized as a crucial biomechanical indicator for disease diagnosis and therapeutic monitoring. Conventional shear wave elastography techniques depend on dispersion analysis and face fundamental limitations in clinical scenarios. Particularly, limited wave propagation data with low signal-to-noise ratios, along with challenges in discriminating between dual dispersion sources stemming from viscoelasticity and finite tissue dimensions, pose great difficulties for extracting dispersion relation. In this study, we introduce SWVE-Net, a framework for shear wave viscoelasticity imaging based on a physics-informed neural network (PINN). SWVE-Net circumvents dispersion analysis by directly incorporating the viscoelasticity wave motion equation into the loss functions of the PINN. Finite element simulations reveal that SWVE-Net quantifies viscosity parameters within a wide range (0.15-1.5 Pa*s), even for samples just a few millimeters in size, where substantial wave reflections and dispersion occur. Ex vivo experiments demonstrate its applicability across various organs, including brain, liver, kidney, and spleen, each with distinct viscoelasticity. In in vivo human trials on breast and skeletal muscle tissues, SWVE-Net reliably assesses viscoelastic properties with standard deviation-to-mean ratios below 15%, highlighting robustness under real-world constraints. SWVE-Net overcomes the core limitations of conventional elastography and enables reliable viscoelastic characterization where traditional methods fall short. It holds promise for applications such as grading hepatic lipid accumulation, detecting myocardial infarction boundaries, and distinguishing malignant from benign tumors. |
| title | Physics-informed Neural Networks Enable High Fidelity Shear Wave Viscoelastography across Multiple organs |
| topic | Medical Physics Soft Condensed Matter Biological Physics |
| url | https://arxiv.org/abs/2505.03935 |