QUAITOM – The Infrastructure for QUantitative AI-based TOMography
3D imaging based on µCT X-rays and neutrons allows seeing structures inside materials. By visualizing internal structures we can understand mechanisms important for developing new medicine and diagnostic tools or for inventing materials for a sustainable future. Quantitative image analysis is necessary for explaining the imaged structures. Often µCT images are complex and large, and therefore difficult and time-consuming to analyze. Here, AI-based deep learning can be the solution. But new analysis techniques are necessary to minimize the time spend on annotating training data.
The challenge with 3D imaging data is that the individual images are large and contain structures unique to that single dataset. Therefore, it is necessary to provide labels specific to the data at hand. The question is how to do that in the most efficient way such that the minimal number of images needs to be labeled to train a sufficiently good model.
QUAITOM will facilitate that research in AI is joined with imaging science by providing data and challenges that reflect the analysis problems in 3D imaging. Based on this data, the AI community will develop new algorithms to be used by imaging scientists. To make this accessible, we will create a platform where analysis methods, software tools for evaluation, and data are available.
The expectation is that QUAITOM will lead to new principles for the deep learning-based analysis of 3D imaging data. These new deep learning-based analysis methods will enable valuable knowledge from µCT imaging to be extracted and used for a more detailed understanding of our world. QUAITOM links with the large-scale science facilities MAX IV and ESS and will provide a platform for translating advanced science to solutions for the benefit of society.
Technical University of Denmark has published an article on the project (external link in English): DTU will make it easier to analyze unique 3D images – DTU Compute.