Dataset Information

Predicting 3D soft tissue dynamics from 2D imaging using physics informed neural networks.

ABSTRACT: Tissue dynamics play critical roles in many physiological functions and provide important metrics for clinical diagnosis. Capturing real-time high-resolution 3D images of tissue dynamics, however, remains a challenge. This study presents a hybrid physics-informed neural network algorithm that infers 3D flow-induced tissue dynamics and other physical quantities from sparse 2D images. The algorithm combines a recurrent neural network model of soft tissue with a differentiable fluid solver, leveraging prior knowledge in solid mechanics to project the governing equation on a discrete eigen space. The algorithm uses a Long-short-term memory-based recurrent encoder-decoder connected with a fully connected neural network to capture the temporal dependence of flow-structure-interaction. The effectiveness and merit of the proposed algorithm is demonstrated on synthetic data from a canine vocal fold model and experimental data from excised pigeon syringes. The results showed that the algorithm accurately reconstructs 3D vocal dynamics, aerodynamics, and acoustics from sparse 2D vibration profiles.

SUBMITTER: Movahhedi M

PROVIDER: S-EPMC10199019 | biostudies-literature | 2023 May

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Predicting 3D soft tissue dynamics from 2D imaging using physics informed neural networks.

Movahhedi Mohammadreza M Liu Xin-Yang XY Geng Biao B Elemans Coen C Xue Qian Q Wang Jian-Xun JX Zheng Xudong X

Communications biology 20230518 1

Tissue dynamics play critical roles in many physiological functions and provide important metrics for clinical diagnosis. Capturing real-time high-resolution 3D images of tissue dynamics, however, remains a challenge. This study presents a hybrid physics-informed neural network algorithm that infers 3D flow-induced tissue dynamics and other physical quantities from sparse 2D images. The algorithm combines a recurrent neural network model of soft tissue with a differentiable fluid solver, leverag ...[more]

PMID: 37208428

Similar Datasets

Project description:In this work an innovative approach was developed to address a significant challenge in the field of radiation dosimetry: the accurate measurement of spatial dose distributions using Fricke gel dosimeters. Hydrogels are widely used in radiation dosimetry due to their ability to simulate the tissue-equivalent properties of human tissue, making them ideal for measuring and mapping radiation dose distributions. Among the various gel dosimeters, Fricke gels exploit the radiation-induced oxidation of ferrous ions to ferric ions and are particularly notable due to their sensitivity. The concentration of ferric ions can be measured using various techniques, including magnetic resonance imaging (MRI) or spectrophotometry. While Fricke gels offer several advantages, a significant hurdle to their widespread application is the diffusion of ferric ions within the gel matrix. This phenomenon leads to a blurring of the dose distribution over time, compromising the accuracy of dose measurements. To mitigate the issue of ferric ion diffusion, researchers have explored various strategies such as the incorporation of additives or modification of the gel composition to either reduce the mobility of ferric ions or stabilize the gel matrix. The computational method proposed leverages the power of artificial intelligence, particularly deep learning, to mitigate the effects of ferric ion diffusion that can compromise measurement precision. By employing Physics Informed Neural Networks (PINNs), the method introduces a novel way to apply physical laws directly within the learning process, optimizing the network to adhere to the principles governing ion diffusion. This is particularly advantageous for solving the partial differential equations that describe the diffusion process in 2D and 3D. By inputting the spatial distribution of ferric ions at a given time, along with boundary conditions and the diffusion coefficient, the model can backtrack to accurately reconstruct the original ion distribution. This capability is crucial for enhancing the fidelity of 3D spatial dose measurements, ensuring that the data reflect the true dose distribution without the artifacts introduced by ion migration. Here, multidimensional models able to handle 2D and 3D data were developed and tested against dose distributions numerically evolved in time from 20 to 100 h. The results in terms of various metrics show a significant agreement in both 2D and 3D dose distributions. In particular, the mean square error of the prediction spans the range 1×10-6-1×10-4, while the gamma analysis results in a 90-100% passing rate with 3%/2 mm, depending on the elapsed time, the type of distribution modeled and the dimensionality. This method could expand the applicability of Fricke gel dosimeters to a wider range of measurement tasks, from simple planar dose assessments to intricate volumetric analyses. The proposed technique holds great promise for overcoming the limitations imposed by ion diffusion in Fricke gel dosimeters.

Dataset Information

Predicting 3D soft tissue dynamics from 2D imaging using physics informed neural networks.

Publications

Predicting 3D soft tissue dynamics from 2D imaging using physics informed neural networks.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets