Machine Learning Methods for Application in Bio-EMC

Bio-electromagnetic (BioEM) engineering, an emerging inter-discipline area with many high-end technologies involved, draws tremendous interest of researchers and scientists. It is foreseeable that human implants will be developed as highly integrated and intelligent devices of ultra-small size, low-power consumption but full abilities of sensing, controlling, data processing, wireless data and power transmission. Among them, one of the most challenging designs is the human brain implant, which is expected to change patients’ life as well as promote normal human beings’ life.

Since brain implants are in a very close interaction with a human body and complicated brain tissues, it is essential to learn and to control their EM effects on a human brain environment. This requires using highly accurate, efficient and reliable modelling, simulation and measurement tools and techniques to characterize their hierarchical, anisotropic, dispersive features. Furthermore, crosstalk and interference between implants and on-/off-body electronics causes more issues in realistic scenarios.

Considering all above, this project aims to develop brain implants and human head models, improve simulation methods and facilitate measurement and validation ways to get closer to real state in this principal criterion. Especially, machine learning (ML) methods, which has witnessed a rapid development in solving complex electromagnetic (EM) problems, will be adapted to solve Bio-EMC problems for brain implants, and three work packages are planned as following:

  • Getting realistic implant model with wireless power transfer and realistic human head model are the main primary step to get in to the problem. Hence, in this work package, the focus is mainly providing accurate simulation models for both of them.
  •  Near Field full-wave simulation provides the most accurate simulation results, but it is obviously time and memory consuming due to huge numerical calculations. Adaptive sampling and ML-based clustering and data acquisition of Huygens surface is expected to be a replacement for full wave simulations. Hence, in this work package, an adaptive sampling method will be used to detect brain implant EM field on human head models and errors in comparison to full wave simulation will be investigated. ML-based Source reconstruction and EM inverse scattering is another interest in this work package which will be used in other parts for optimization as well.
  • Measurements using Near-Field scanner for simulation result validation will be done. Elaborating the problem to DUT-to-DUT correlation of human head model using several brain implants as well as applying Machine Learning Methods for EMC analysis regarding various parameters, such as the position, the distance, the orientation and the frequency is the main interest in this work package.

Workflow of creating digital twin of brain implants for Bio-EMC applications. Machine learning methods will be integrated during each steps, such as near-field sampling, Huygens’ source generation and emission prediction.

Funding: Freie und Hansestadt Hamburg
Contact: Hamideh Esmaeili, M. Sc.
Start date: 01.11.2021

Prediction of the Electromagnetic Biocompatibility of Human Brain Implants

Morten Schierholz, M. Sc., Dr. Cheng Yang. Hamburg Electronics Lab for Integrated Optoelectronical Systems (HELIOS) 01.12.2019 – 30.09.2021

Modelling and simulating implants before they are used in the human body is an important aspect to ensure the safety and quality of the device. Multiple organizations as the IEEE or ICNIRP provide guidelines with regard to safety constraints of electrical devices near on inside the human body distinguishing between different body parts. Staying within the given boundaries is important to not harm the body tissue. One common figure of merit is the specific absorption ratio (SAR) which is the power emitted by the implant into a certain mass of body tissue.
Simulating the brain implant with the different tissue layers of the human brain is a challenging task due to different electromagnetic properties of the tissues and a strong frequency dependency. Furthermore the resulting large wavelength compared with the dimension of the implant from the frequencies up to 100 MHz introduces an additional challenge due to the interest of the accurate near field behavior.
This project focusses on the verified simulation of the brain implant in the human body. Therefore a cross validation approach is pursued with a combination of analytical methods as well as time domain and frequency domain electromagnetic field solvers.

Assumed position of the implant inside the human head. The rectangular loop is a first simple approximation of brain implant to validate the simulation for the frequency range and the geometrical size of the time domain solver. The size of the loop is 1 by 1 mm.

Approximate model of an implant with a printed circuit board (PCB) and chip on top.  The Dimension of the chip is 2.5 by 3 mm. On the bottom of the PCB is the loop for power and data transmission (not shown). Due to the small heights of the chip layers (less than 1µm) the thickness of the metal layers is increased for better visibility.

Extension of CONCEPT-II for the Efficient Modeling of Various Types of Coils

Heinz-D. Brüns, Angela Freiberg, Dennis Zappe. 01.01.2017 – 31.03.2017

Nowadays there is a growing demand for the numerical analysis of the electromagnetic behavior of various types of coils. An example is the technology of wire power transfer where an effective coupling across an air cap is required. Further examples refer to various types of medical imaging such as Magnetic Particle Imaging (MPI) and Magnetic Resonance Imaging (MRI), where coils are used for generating the desired H field distributions.
The in-house tool CONCEPT-II, a program package for solving electromagnetic radiation and scattering problems on the basis of the Method of Moments (MoM), can be used to compute the currents, losses and fields of wire or surface structures. To date the modeling of coils with the available CAD tools was a rather tedious, time intensive task. The aim of this project was to provide a tool in CONCEPT-II enabling the quick discretisation of various types of coils frequently encountered in the mentioned or other technical applications.

A Rogowski coil generated from short thin-wire sections.

The H field distribution of a short cylindrical coil fed at a low frequency.

Characterization of Transmit-Receive Antennas for Magnetic Resonance Imaging via Moment Method and Volume Segmentation

Ph. D. Thesis Christian Findeklee. Completed 12.2012

Today’s availability of computation power and memory enables simulating electromagnetic interaction of heterogeneous objects with complex antenna structures. In this study, segmentation of volume currents was used in combination with a port-based formulation of the Method of Moments. Interesting insights into this approach could be found and the patient coupling of a multi-resonant headcoil for Magnetic Resonance Imaging was studied.

Birdcage headcoil with patient model for volume current simulation. (The cylincrical RF-shield is not shown.)

Circular polarized field inside the patient for quadrature driven head coil: The ellipses show the time dependent transversal magnetic field during a period (about 8ns).

Related Publications:

Christian Findeklee
Analyse von Sende- und Empfangsantennen der Magnetresonanztomographie mit der Momentenmethode und Volumensegmentierung
Dissertation 2012. Shaker Verlag Aachen. 2012 (ISBN 978-3-8440-1659-8)

Christian Findeklee, Hanno Homann, Heinz-Dietrich Brüns, Hermann Singer
SAR-optimized coil design via post processing port based simulations in the moment method (MoM)
28th Annual Meeting of European Society for Magnetic Resonance in Medicine and Biology (ESMRMB); Leipzig, Germany, October 6-8, 2011

Christian Findeklee
Array Noise Matching – Generalization, Proof and Analogy to Power Matching
IEEE Transaction Antennas Propagation, vol. 59, No. 2, pp. 452-459, February 2011.

Christian Findeklee, Randy Duensing, Arne Reykowski
Simulating Array SNR & Effective Noise Figure in Dependance of Noise Coupling
19th Annual Meeting of International Society for Magnetic Resonance in Medicine (ISMRM), 2011.

Christian Findeklee
Improving SNR by Generalizing Noise Matching for Array Coils
17th Annual Meeting of International Society for Magnetic Resonance in Medicine (ISMRM), p. 507, Honululu, HI, USA, April 2009.

Christian Findeklee, Gesa Lilith Franke, Ulrich Katscher
A Probe for Electric Properties of Phantom Liquids
17th Annual Meeting of International Society for Magnetic Resonance in Medicine (ISMRM), p. 2939, Honululu, HI, USA, April 2009.

Christian Findeklee, Ulrich Katscher
3D MRI-based Electric Properties Tomographie Reconstruction using Volume Currents in the Method of Moments
16th Annual Meeting of International Society for Magnetic Resonance in Medicine (ISMRM), p. 1178, Toronto, Ontario, Canada, May 2008.

Christian Findeklee
Optimal Noise Matching for Array Antennas
Technical Report PR-TN 2008/00431, Koninklijke Philips Electronics N.V., 2008.

Christian Findeklee, Jens Eichmann, Peter Vernickel
Decoupling of a Multi Channel Transmit/Receive Coil Array via Impedance Inversion
15th Annual Meeting of International Society for Magnetic Resonance in Medicine (ISMRM), p. 1020, Berlin, Germany, May 2007.

Christian Findeklee, Daniel Wirtz
Noise Coupling in parallel MRI Receive Coils
Technical Report PR-TN 2006/01086, Koninklijke Philips Electronics N.V., 2007.

Christian Findeklee, Christoph Leussler, Michael Morich, Gordon Demeester
Efficient Design of a novel Double Tuned Quadrature Headcoil for Simultaneous ¹H and ³¹P MRI/MRS at 7T
13th Annual Meeting of International Society for Magnetic Resonance in Medicine (ISMRM), p. 891, South Beach, Miami, Florida, USA, May 2005.

Christian Findeklee
Efficient Eigenmode Calculation of Electromagnetic Structures by Extrapolating a Port Model from only a few Frequencies
21th Annual Meeting of European Society for Magnetic Resonance in Medicine and Biology (ESMRMB) , no. 224, p. 386, September 2004.

Christian Findeklee, Heinz-Dietrich Brüns Hermann Singer
Electromagnetic Simulation in Anisotropic and Inhomogeneous Media by Volume Currents in the Moment Method
European Electromagnetics Sy