Projects

Aortic valve dysfunction account for severe cardiac malfunction associated with patient disabilities. Transcatheteral aortic valve implantation (TAVI) is a minimal-invasive procedure for aortic valve dysfunction that requires careful planning with non-invasive imaging, which is typically performed with contrast-enhanced computed tomography (CT). This project aims to implement interpretability techniques to explain model predictions, aiding clinicians in understanding key factors affecting TAVI outcomes for better decision-making.

Predicting patient-specific liver deformations for radiofrequency ablation procedures is essential for accurate treatment planning and intraoperative guidance. In this project we address the challenges by developing a novel AI-driven framework that combines the best approaches to accurately model patient-specific liver deformations. Moreover, to overcome deep learning “black boxes” the PhD should also explore new approaches towards quantitative assessment of such solutions, such as failure mode analysis, model architecture analysis and reduction, as well as continual learning aspects.

Despite successful reperfusion therapy, approximately 50% of acute myocardial infarction (AMI) patients experience incomplete myocardial recovery due to microvascular injury, substantially increasing the risk of major adverse cardiac events. This project aims to enable early and accessible detection of microvascular injuries using standard electrocardiograms (ECGs). The PhD candidate should develop e.g., a deep learning model that leverages state-of-the-art multimodal contrastive learning methods but also novel generative data augmentation techniques and integration of large pre-trained models to enrich ECG encoders with other data modalities (e.g., cardiac MR). This research will lead to scalable, cost-effective tools for early risk stratification and timely intervention in AMI patients.

Modern RT in oncologic patients is fundamentally linked to medical imaging. This PhD project aims to develop a robust dosiomics framework that enhances outcome prediction in radiotherapy by incorporating spatial dose distribution characteristics into conventional radiomics pipelines. Dosiomics, a novel subfield of radiomics, includes detailed dosimetric information—such as dose-volume metrics and Gamma Index maps—to identify predictive patterns beyond anatomical imaging. The project aims to establish lasting collaborations with UKT Tübingen AI Cluster, Brainlab AG, Cyted Health (UK), using both academic and industrial expertise in AI, radiotherapy planning, and clinical translation.

In medical imaging, Conditional Diffusion Models hold great promise for improving image reconstruction, resolution enhancement, data completion, and segmentation of anatomical structures or abnormalities. However, open challenges remain in effectively integrating conditional information (e.g., clinical metadata) and ensuring robustness across diverse imaging protocols and patient populations. This project should explore conditional diffusion models to address key challenges in medical imaging, focusing on reconstruction, segmentation, and analysis tasks. Building on prior successful collaborations among the co-PIs, this project will develop innovative tools to enhance clinical workflows.

Recent advances in photon-counting detectors allow spectrally resolved CT imaging, one of the main imaging modalities for many clinical indications. Initially, reconstruction methods focused on K-edge-based substance identification, while newer methods now enable broader medical applications. We aim to extend this to multispectral CT and automated stroke prediction from photon-counting CT (PCCT). This project will combine material decomposition and noise reduction, leveraging both modeling and machine learning. By developing efficient problem adopted learning strategies for diffusion models applied to spectrally resolved CT data, we will advance clinical PCCT applications.

Carotid plaques are one of the main risk factors for stroke. Accurate assessment of carotid plaque composition is essential for evaluating plaque stability and detecting calcified components, as this information can significantly impact therapeutic decision-making. PCCT offers a significant advancement by improving both temporal and spatial resolution while enabling multi-energy imaging and advanced texture analysis techniques such as radiomics. However, the application of PCCT to carotid plaques has not been extensively studied. The PhD should train a deep learning model on an existing retrospective dataset, with external validation conducted on a prospective patient cohort.

This PhD project aims to advance our understanding of cerebral metabolism and tissue reorganization following traumatic brain injury (TBI) through the use of multinuclear quantitative MRI. By combining ³¹P MR spectroscopy, ²³Na MR imaging and quantitative 1H MRI, the DC will investigate alterations in energy metabolism, sodium homeostasis and microstructural alterations across acute, subacute, and chronic stages of TBI, identify novel biomarkers of secondary injury and develop prognostic models linking early metabolic disturbances to longterm clinical outcomes in TBI patients. Ultimately, this research aims to improve TBI stratification & therapeutic monitoring, paving the way for precision neurology in critical care.

In this project, the PhD student will develop an adaptive optics-based nonlinear microscopy platform and explore applications for deep tissue imaging. This project will build on recent imaging developments achieved in a joint project between the Institutes of Biomedical Physics and Physiology, which included the acquisition of a powerful pulsed laser system for high-end imaging tasks designed for deep tissue penetration. The student will implement and evaluate a novel dynamic optical element to counteract light scattering from thick tissue layers. This tool will improve the imaging depth beyond the current state of the art.

In this project, the DC will investigate novel approaches for Magnetorelaxometry Imaging (MRXI) of magnetic nanoparticle (MNP). MRXI signals do not only contain information on the amount of nanoparticles in a certain volume, but also on their surrounding biological environment, e.g. temperature, binding states, and viscosity. The PhD will work on experimental investigations including the development of measurement procedures and suitable phantoms and on reconstruction approaches based on e.g. deep learning and physics-informed neural networks With these findings, we aim at advancing magnetic nanoparticle-based treatments, possibly providing clinicians with a novel imaging modality in the future.

The development of targeted ligands for oncological applications in nuclear medicine presents a rapidly growing area of research. Radiolabelled ligands applicable for theranostic or multi-modal approaches bear an unequivocal strength to improve personalized patient care. These compounds combine the possibility of radiolabelling for imaging and therapy and through the introduction of additional functionalities can be used for synergistic multimodal imaging/therapy as well as multireceptor targeting. In this project innovative conjugation and radiochemistry strategies will be investigated allowing to introduce different functionalities in the same molecule. Preclinical testing overseen by a radiopharmacist will include radiolabelling experiments, as well as in vitro / in vivo studies to characterise
receptor specific targeting and therapeutic efficacy. Clinical advice of a radiologist will support the selection process of most appropriate combinations of imaging/therapy concepts and receptor targeting functionalization.

In this project, the PhD candidate will develop a novel optical coherence tomography (OCT) prototype for high-resolution human tissue imaging. Unlike conventional OCT imaging, the system will be equipped with dynamic contrast based on the nano-scale motion patterns of cellular structures. Its applications aim at demonstrating the fast and non-destructive assessment of tumor characteristics as well as the longitudinal tracking of a versatile in-vitro tumor model, generating a seminal OCT image dataset for the training of clinically relevant foundational models aiming to support intra-surgical imaging applications such as the enhanced definition of surgical margins.