Designing soft magneto-active microactuators: AI-driven patient-specific computational framework for precise intrabody treatment

13^th Global Summit on Artificial Intelligence and Neural Networks

March 14, 2025 | Webinar

Alina Popova

Unversity of Galway, Ireland

Posters & Accepted Abstracts: J Comput Sci Syst Biol

Abstract :

Background: Nowadays medical surgery has made significant advancements, enhancing accessibility to healthcare for all individuals. However, certain treatments require robotic intervention due to the small size of human organs and vessels. Soft materials that can undergo rapid and large deformation through the remote and wireless action of external stimuli. These adaptable soft materials have a huge potential for biomedical and healthcare applications, in the form of microactuators that can be guided remotely by external magnetic fields3. One promising approach to significantly improve their magnetomechanical performance and decrease the necessary magnetic field is the development of soft magnetoactive composites with structured microarchitectures. Regarding microstructure design, prevailing theoretical approaches primarily concentrate on periodic structures.2 Designing microstructured magnetoactive elements (MAEs) is challenging due to the need to determine the optimal microstructure to achieve desired properties. This task involves an inverse, nonlinear, and multiphysics coupling problem. Among other approaches, Generative Adversarial Networks (GANs) offer a solution to these challenges. More specifically, GANs can generate microstructures with desired properties by learning from a dataset. This makes them well-suited for solving MAE design problems. In this work, it is proposed to develop a computational framework based on a generative adversarial network. This ambitious action will be conducted at University of Galway under the supervision of Prof. Stephan Rudykh, an expert in soft magneto active materials Aims and objectives of the project: Previous research of magnetoactive materials showed great potential of their application in remotely controlled medical devices such as targeted drug delivery, microsurgery, and minimally invasive medical procedures. This approach involves detailed investigation and optimization of material structures, to obtain specific physical properties, such as a given frequency response, saturation magnetization and coercivity1. The project aims to develop a computational framework based on Generative Adversarial Networks (GANs) to advance the state-of-the-art in material discovery for soft magnetoactive materials (MAEs) and their application in biomedical engineering and healthcare. The first objective is to estimate the effective properties of a material based on an arbitrary representative volume of the microstructure â?? e.g., by using computer simulations and homogenization techniques. I will apply the finite element method for the solution of the nonlinear magnetoâ??mechanical boundary value problem. The tool for generating a dataset of microstructured MAEs with different architectures featuring diverse properties will be created within COMSOL. Each unit cell will be presented as a graph for a Representative Volume Element (RVE) with 27 vertices. To validate the computational tool and dataset I will print the obtained RVE on one of our 3D printers and test the samples in the laboratory. This objective involves the development of a new theoretical and experimental material modelling framework that will be used throughout the project. By carrying out this objective I will acquire knowledge in material science, mechanical engineering, and applied mechanics. I will use software and framework resources that are easily accessible: Matlab and Comsol. A second objective is to train a GAN at predicting the effective properties, using a prepared dataset obtained on the first objective. Recent studies of modeling magnetoactive materials are based on experimental data5 and theoretical calculations6. However, such approaches have disadvantages such as the necessity of microscale simulations for the calibration of the macroscopic model which are still quite costly, especially in the 3D case5. Furthermore, theoretical calculations involve an inverse, nonlinear, and multiphysics coupling problem, which requires a lot of computation resources. To achieve this objective, GANs, a recently emerged category of machine learning techniques, will be used (Figure 2). GANs can learn the relationships between different material properties and microstructural features even in systems with non-linear relationships, which make them suitable for solving material design challenges. By carrying out this objective I will acquire knowledge in neural networks, machine learning and applied mathematics. I will use accessible software and framework resources: Python programming language and PyTorch framework. The third objective is to develop an optimization procedure based on the GAN, targeted at reducing the magnetic field needed to induce a given functionality. To validate samples of the generated microstructures I will implement 3D modeling and fabricating samples, thus, I will evaluate performance of generated structures in laboratory experiments. This objective involves the fabrication of new generated microstructures with optimal parameters. I will use software and framework resources that are easily accessible: Python and PyTorch. For the experimental part I will use SolidWorks for modeling 3D elements and printers Stratasys j35 and Allevi. By carrying out this objective I will acquire knowledge in 3D modeling, material science and 3D printing. This research will involve close collaboration with University of Galway and SoftMatterLab supervised under Prof. Rudykh over a period of 4 years. By carrying out this research I will acquire knowledge in biomedical engineering, material science, mechanical engineering, and applied mechanics. Keywords: Soft magneto-active devices, Artificial Neural Network (ANN), Multiphysics material modeling.