KI-Morph — Petabytescale Image Analysis
Aims and Objectives
The current rapid development of imaging methods for X-ray tomography has led to an increasing field of applications from all areas of life sciences. Here X-ray tomography is used to create 3D data of tissues, down to individual cells and their compartments and up to tissue groups and organs to whole organisms. This 3D data is subsequently used to create 3D models and analyze the morphology of the involved specimen. While the generation of such 3D data using tomographs is mostly automatic and only takes a few minutes, the image processing and analysis is still mainly manual labor, costing even experienced researchers several orders of magnitude more time. Accordingly, the image analysis step is the major bottleneck in the entire processing pipeline. In our novel research project, KI-Morph, we aim to address this challenge by providing a framework for petabyte-scale image processing and analysis.
We are focusing on three main parts:
- Providing the infrastructure for the processing of petabyte-scale imaging data
- Development of AI-algorithms for the segmentation of large-scale 3D tomography data
- Evaluation of the processing pipeline with data from many areas of life sciences
Aims and Objectives
Cancer is one of the leading causes of disease-related death worldwide. In recent years, rapid increase in the molecular understanding of cancer has unraveled significant additional complexity of the disease. Although large amounts of data on cancer genetics and molecular characteristics are available and accumulating with increasing speed, adequate interpretation of these data still represents a major bottleneck. This is exactly where mathematics can be applied to oncology: Through mathematical modeling of complex biological processes we are able to gain novel, unprecedented medical insights. The fields of application of mathematical models include the analysis of biological concepts and medical hypotheses about cancer evolution, and the prediction of clinical outcomes using existing clinical and molecular information. On the other hand, the medical applications give rise to mathematical challenges, which can lead to new methods and algorithms in various fields of mathematics, like data analysis, mathematical modeling and machine learning. Therefore, applying mathematics in the field of oncology will facilitate data interpretation and improve our understanding of carcinogenic processes.
We implement mathematical modeling on the example of Lynch syndrome (LS). LS is the most common inherited cancer syndrome and predisposes affected individuals to developing cancer in the large bowel (colorectal cancer) and other organs. LS is reflects general principles of tumorigenesis and tumor immunology beyond the hereditary context in an exemplary manner. We are focusing on three main parts:
- Mathematically modeling the evolution of hereditary tumors to improve the existing prevention strategies
- Elevating tumor immunology to a genome-wide level
- Predicting the efficacy of clinical approaches for diagnostics, prevention and treatment
Low-level image analysis on statistical manifoldsMore Information
End-to-End Learned Random Walker for Seeded Image SegmentationMore Information
Besides the projects described above I have also helped colleagues analysing data in the fields of asterophysics, cybersecurity and many more.