Background
In a data-driven approach for detecting osteoarthritis in human knees, the individual shape and structure of each sample
can be a major source of training problems. Biological samples show strong variability in size, local tissue distribution,
and boundary geometry, and this variability can dominate the learning signal if not treated consistently. As a result,
models may learn sample-specific geometric artifacts instead of pathology-relevant patterns, which reduces robustness
and generalization across large datasets.

Objective 
The objective is to develop an automated pipeline that processes CT images of human knee cartilage samples and prepares
them for simulation-driven data generation. The pipeline should generate an individual geometry for every sample in
a consistent and reproducible way, so that differences in size, shape, and bone-to-cartilage proportion are represented
correctly in the downstream model. It should also identify relevant tissue regions, support robust finite-element model
preparation, and provide stable processing quality across all samples.

Requirements
• A math oriented background like Technomathematics, Mechanical Engineering, Data Science, or a comparable field.
• Programming skills in Python or MATLAB are a plus.
• Knowledge in image detection or image analysis and FEM is advantageous.
What We Offer
• Weekly meetings with structured feedback.
• Close technical support during implementation and validation.
• Active mentoring throughout the full project period.

Contact
Malte Brand
Research Associate
malte.brand@tuhh.de
Room M2.536