Artificial Intelligence

AI- and Radiomics-Based Analyses of Whole-Body MRIs in Multiple Myeloma

In multiple myeloma, a large number of pathologies are often distributed throughout the entire skeletal system, which can be fully captured using modern whole-body MRI examinations. However, analyzing these large image datasets with highly complex patterns poses a major challenge for radiologists in their daily clinical practice, as the detailed image analysis of such datasets is very time-consuming and not all details can be analyzed visually with the highest accuracy. 

Therefore, since 2019, our interdisciplinary group of radiologists, computer scientists, and hematologists has been developing and researching state-of-the-art image analysis methods in the fields of artificial intelligence and radiomics to enable automatic, detailed, and objective analyses of the entire bone marrow.

Developments and publications to date: 

• Automatic AI-based detection of focal lesions:

Our group has developed an initial AI algorithm capable of automatically detecting focal myeloma lesions from MRIs, enabling an automatic estimation of tumor burden from MRI images.

• Automated AI-based bone marrow segmentation:

Our group has developed several AI algorithms capable of automatically segmenting bone marrow. This enables the isolation of key image regions from whole-body MRI scans for downstream analyses.

• Radiomics models for predicting biopsy results or treatment response:

Based on the automatically segmented bone marrow, radiomics analyses can be used to automatically predict biopsy results or treatment response from MRI.

The goal of our research group is to establish AI- and radiomics-based whole-body MRI analyses as a reliable basis for decision-making in oncological imaging in the long term.

 

Artificial Intelligence in Urological Radiology

Artificial intelligence (AI) is becoming increasingly important in radiological diagnostics, particularly in prostate MRI. Our research shows that AI-based methods can significantly support the interpretation of MRI images and thus the diagnostic process.

Modern AI algorithms automatically analyze MRI images and detect patterns indicative of prostate cancer. Large international studies have demonstrated that such systems can make significant contributions to the detection of clinically relevant tumors. At the same time, they can help standardize diagnostics and reduce reliance on individual experience.

Another important approach is the combination of AI with established clinical grading systems (e.g., PI-RADS). Studies from Heidelberg show that this can improve risk assessment and reduce unnecessary prostate biopsies—by up to about half in model analyses, without overlooking relevant tumors.

The goal of these developments is to make diagnostics safer, more precise, and less invasive for patients. AI serves not as a replacement, but as a supportive tool for radiologists to enable well-informed and individualized treatment decisions.

 

AI- and simulation-supported translational spin-lock MRI

Spin-lock MRI is a specialized MRI technique that characterizes tissue at a particularly sensitive physical level. It responds to molecular mobility, exchange processes, and interactions with macromolecular tissue components. This creates additional image contrasts that can provide clues about tumor structure, necrosis, fibrosis, inflammation, and early treatment effects.

Clinically, the method opens up new possibilities in oncological imaging. It complements established MRI protocols with a functional tissue perspective and supports the characterization of lesions, the assessment of treatment response, and the classification of complex tumor changes.

Our research focuses on AI- and simulation-based spin-lock MRI. We combine physical modeling, sequence simulation, phantom experiments, image registration, and data-driven analysis through to application in healthy volunteers and patients. This creates a translational platform bridging radiology, MR physics, computer science, and AI-based image analysis. The goal is to extract robust information from complex MRI signals for diagnostic and therapeutic monitoring purposes.