Currently, I am a research scientist at Fraunhofer MEVIS. As part of the KI-FDZ project, we would like to make machine learning more accessible to the researchers. I am part of a team that are developing a workbench to help experts, who do not have a lot of experience in machine learning and data analysis, use machine learning and data science tools in their research. In this project, we are making standardized machine learning pipelines available, so no programming is required, yet give enough freedom to ML experts to implement their own algorithms. In this project, we are using various languages and tools, such as Python, R, SQL, MLflow, AutoML, etc.
In another project, in collaboration with Uni-Klinikum Freiburg, I use deep learning to segment structures in MR images. In the next step, I implemented algorithms to automatically extract the knee's shape and kinematic properties of the patients, with recurrent patellar dislocation for better understanding of diseases and risk factors.
Before joining Fraunhofer MEVIS, I was a postdoctoral fellow in the University of Bern, Institute for Surgical Technology and Biomechanics. We used statistical models, machine learning and image processing techniques to segment the bones and also improve the prediction of bone biomechanical properties.
My PhD thesis, was focused on building patient-specific biomechanical model of total knee replacement with a special focus on patella. One of the most important features for estimating bone strength is the anisotropic information of the bone. This information however is not visible in clinical CT scans. We proposed to use image registration to estimate this information for patients. Solving FE models is a time consuming process, which makes this technique less appealing to the clinicians. We proposed to use machine learning techniques to estimate the outcome of FE calculations in real-time.
I finished my Master of science study in Sharif University of Technology, in Iran, in the field of pattern recognition and computer vision. My master thesis is entitled "Face Recognition in Subspace Domain Using Kernel Methods", during which I learned a lot about nonlinear recognition methods
I received my B.Sc from Islamic Azad University of Mashhad, in computer engineering.
Elham Taghizadeh
Senior Research Scientist
Fraunhofer MEVIS
