Project Details
Layman's description
In the realm of bio-imaging, understanding the function of living organisms at the cellular and sub-cellular level has long been a sought-after goal. While existing medical imaging technologies like MR and CT are able to offer valuable high-level insights about our health, they fall short when it comes to visualizing the complicated structure of for instance brain neurons or bone trabeculae. At these scales, the inner workings of the human body are still a great mystery. However, thanks to new fourth-generation synchrotron facilities like MAX IV and ESRF, it is now possible to perform non-destructive high-resolution x-ray microscopy of these tissue structures at a very large scale, enabling visualization of all cells within centimeter-sized tissue samples.
Using an array of x-ray lenses, the x-ray beam from the synchrotron can be focused on either a smaller or larger area of the sample, enabling us to visualize the same tissue in multiple resolutions. This opportunity allows us to study the hierarchical nature of biological structures across multiple scales from millimeters to micrometers.
However, with this approach comes with multiple challenges. Firstly, multiple local tomographic images over the same sample are expected to require impractical amounts of data resources. Secondly, prolonged exposure of such high-intensity x-rays eventually destroys the sample.
To resolve these issues, this PhD project employs the help of Artificial Intelligence, specifically learning-based super-resolution. The idea is that if we can learn how the micro-structure of how these samples look in high-resolution, we may be able to reconstruct these structures from low-resolution scans obtained by widening the x-ray beam over a larger area of the sample. This means that even from low-resolution data, we will be capable of producing detailed, high-resolution images, offering insights into cellular structures and their functions without needing to record vast amounts of data or destroying the samples.
The necessary training data to train the super-resolution model will be collected through collaboration with the Danish beamline DanMAX at MAX IV. By combining cutting-edge imaging technology with state-of-the-art AI algorithms, this project aims to shed light on the inner workings of biology, paving the way for subsequent research.
Using an array of x-ray lenses, the x-ray beam from the synchrotron can be focused on either a smaller or larger area of the sample, enabling us to visualize the same tissue in multiple resolutions. This opportunity allows us to study the hierarchical nature of biological structures across multiple scales from millimeters to micrometers.
However, with this approach comes with multiple challenges. Firstly, multiple local tomographic images over the same sample are expected to require impractical amounts of data resources. Secondly, prolonged exposure of such high-intensity x-rays eventually destroys the sample.
To resolve these issues, this PhD project employs the help of Artificial Intelligence, specifically learning-based super-resolution. The idea is that if we can learn how the micro-structure of how these samples look in high-resolution, we may be able to reconstruct these structures from low-resolution scans obtained by widening the x-ray beam over a larger area of the sample. This means that even from low-resolution data, we will be capable of producing detailed, high-resolution images, offering insights into cellular structures and their functions without needing to record vast amounts of data or destroying the samples.
The necessary training data to train the super-resolution model will be collected through collaboration with the Danish beamline DanMAX at MAX IV. By combining cutting-edge imaging technology with state-of-the-art AI algorithms, this project aims to shed light on the inner workings of biology, paving the way for subsequent research.
| Status | Active |
|---|---|
| Effective start/end date | 01/01/2024 → 31/12/2026 |