Image Registration in Biomedical Applications

The goal is to understand, compare, and enhance classical approaches and approaches based on machine learning techniques to solve medical image registration problems.

Several drugs and diseases alter mechanical behavior of the heart. Being able to accurately capture changes in the mechanics is important in order to know whether a drug is working as expected. Recent technology advancements allows for drug testing of cardiac cells in an isolated environment that is similar to the environment inside the body. Such system as are often referred to as Heart-on-chip systems because cardiac cells are placed inside a chamber referred to as a chip. These systems can be used to test drugs on cardiac cells coming from an individual which opens up the possibility to test drugs on a specific patient's cardiac cells without any risk of doing harm to the patient. The motion of cells can be captured by placing a camera above the cells and record. Computer vision techniques can thereby be used to extract displacement and velocity fields.


We would like to apply image registration techniques to the problem of finding the displacement field that governs the movement of cardiac cells. We also know that the mechanics of the cells can be described as a nonlinear hyperelastic material which could also be incorporated into the modeling of the problem.

Learning outcome

In the scope of the thesis the student will apply physics informed neural networks to solve the task and compare it to classical approaches. The architecture of the neural network will be inspired by classical approaches to solve the problem.


A background in applied mathematics, mechanics, physics, and/or computer science is required. Knowledge of partial differential equations, numerical methods such as finite element methods, and/or high-level programming is a must. Knowledge of mathematical optimization and/or machine learning is a plus.


  • Henrik Nicolay Finsberg
  • Johannes Haubner

Contact person