Research

Our aim is to advance and translate methods and applications for processing and analysis of medical imaging data using artificial intelligence (AI) and machine learning (ML) methods.

Specifically, we focus on:
1. Developments for dynamic and multi-parametric MRI
2. Development of AI methods for medical imaging
3. Translation of AI to clinical applications

with the interest of imaging applications and computing technology to provide explainable and human-interpretable methods that enable the integration of AI solutions into clinical practice and that provide patient-centered workflows for the comprehensive diagnosis and management of neurological, cardiovascular and oncological patients.


Research focuses

Overview of data processing steps and information exchange

I. Developments for dynamic and multi-parametric MRI

The development of efficient acquisition strategies for multi-parametric and dynamic imaging is essential for non-invasive tissue and metabolism characterization. Imaging in conjunction with appropriate reconstruction techniques, the handling of motion and other sources of imaging artifacts can improve the obtained image quality. The aim is to provide an improved workflow with automatic generation of high resolution imaging data and to automatically derive clinical biomarkers that can be used in diagnosis.

II. Development of AI methods for medical imaging

The application of AI methods in acquisition, reconstruction, post-processing or analysis are investigated. The aim is to provide reliable, robust, specific and sensitive methods. The inclusion of AI into medical data processing can help to improve performance by (but not limited to) increasing precision, boosting quality of service, easing processing and reducing computational times. ML-based medical image analysis still suffers from limited robustness and generalizability to out-of-distribution data. Furthermore, In the context of large epidemiological studies, manual image analysis is often not feasible due to the overwhelming amount of data. The integration with non-imaging-data and expert knowledge can provide a better understanding of the causal generative process and can thus contribute to building ML models for medical imaging that capture the relevant information and allow for reliable predictions in a clinical setting.

III. Translation of AI to clinical applications

We aim to implement state-of-the-art AI methods for clinical applications to support the clinicians in their daily work. These projects enable automated processing, simplified workflows and patient-phenotypic processing and analysis. Imaging and non-imaging data are coherently processed to guide and monitor patients with neurological, cardiovascular and oncological diseases. One example use-case is the automatic analysis of whole-body imaging data such as PET-CT. In this context we launched an ML challenge: autoPET.

Research projects

For current ongoing research projects, please refer to our publictions, codes and documentation or open student projects.