| This project page is currently under construction and new projects will be added continuously! |
Current Projects:
This page regards the project for the module "DSA - Domain-Specific Aspects of Data Science" from the master's program Data Science.
Please note, that according to to curriculum, you have to be finished with at least 3 of the following lectures before you can start with the project:
The DSA module itself consists of three lectures:
On this page, you find descriptions of potential projects that you can do at the group for Computational Biomechanics at the ILSB.
The Domain-Specific Lecture in Data Science that fits for these projects is 202.064 Computational Biomaterials and Biomechanics, which is offered at the ILSB in the winter term.
You either have to have finished this lecture before starting with the project but it is also an option to do both at the same time.
Content
For the creation of finite element (FE) models, the outer boundary of the bones has to be found in clinical CT images. Several tools exists for this task, but typically require some manual cleaning of the data afterwards.
In this project, an automated approach using neural networks should be tested.
Figure: Automated masking of bones in CT images
In a first step, ex vivo scans of femora are used, where the background is mostly air. When a model is established, also in vivo scans can be used, which also contain soft tissue and other bones.
Requirements
Dataset
Objective
Next Steps
Content
Bone consists of two major parts: a dense outer layer (cortical bone) and a porous inner layer (trabecular bone). In many applications, it is required to label these two volumes separately, for example to only evaluate trabecular or cortical properties.
Currently, image processing algorithms are used, that try to find the boundary between the two volumes. These algorithms are relatively slow, especially for large bones or very small voxel sizes.
One idea would be to use artificial intelligence to label the two volumes in the bone. Different AI methods can be employed and tuned.
Figure: Segmentation of cortical and trabecular bone from a binary image into a labeled image.
Requirements
Dataset
Objective
Next Steps
Content
Bone contains a microstructure, called trabecular bone. Using CT-based morphometric analysis, it is possible to quantify this microstructure for example in terms of density, orientation, spacing, or shape coefficients.
In the past, typically only single region of interest (ROI) were evaluated, however, modern tools such as Holistic Morphometric Analysis (HMA) can be used to visualize the microstructure over the whole volume.
HMA results in several thousand datapoints over the whole bone volume. These datapoints are not only highly autocorrelated but also the morphometric indices itself are correlated to each other.
In this project, visual data science methods should be applied for the explorative analysis of CT-based morphometry, based on HMA.
Requirements
Dataset
Objective
Next Steps
Content
Bone contains a microstructure, called trabecular bone. For previous projects, 5mm cubes of trabecular bone were digitally cut from micro-CT (µCT) images. Morphometric indices, such as density, orientation, spacing, or shape coefficient can be measured on those cubes.
To better understand how these morphometric indices are influenced by the local geometry of the bone, a visual data analysis should be performed to visualize both the 3D-rendering of the cubes as well as the morphometric indices. The indices itself are either scalar, vector, or (2nd and 4th order) tensorial quantities.
Figure: Morphometric indices measured on two different bone cubes
Requirements
Dataset
Objective
Next Steps