I'm a Computer Scientist with doctoral, postdoctoral, and group leadership experience in the division of Medical Image Computing (MIC) at the German Cancer Research Center (DKFZ). I specialized at the intersection of machine learning / deep learning, medical image analysis / computer vision, and research software engineering. My work spans the following domains:
-
🔬 Population-scale medical image analysis, automated body-composition quantification, cardiometabolic and respiratory health: As co-applicant and DKFZ coordinator of a DFG-funded SPP 2177 Radiomics project (project announcement), I worked on trustworthy large-scale imaging analysis using population-scale data from the German National Cohort (NAKO). I reframed medical image segmentation quality control (QC) around practical utility, introducing evaluation strategies that better capture real-world QC performance, together with tuning mechanisms to optimize QC outcomes at scale. I also developed a Python/DL pipeline for reliably detecting and correcting fat–water swaps in population-scale imaging data.
-
🩺 Translational medical AI: As group leader of the Translational Medical AI team at MIC, I contributed to and coordinated collaborative research with medical doctors from the Heidelberg/Mannheim and Freiburg university clinics. Together, we developed a broad range of ML/DL approaches for imaging-based disease research spanning pulmonary, cardiovascular, oncologic, metabolic, and musculoskeletal diseases. Selected work: COPD assessment · pedicle-screw planning · aortic-dissection detection · radiomics workflow standardization · colorectal-neoplasia prediction · multi-disease CT biomarker analysis
-
🧩 Shape-based segmentation using 3D Statistical Shape Models: Explicit geometric modeling of anatomical shapes encodes important structural information from underlying images. During my PhD, I advanced key components of 3D statistical shape models for image-based model fitting and segmentation. I combined more discriminative landmark appearance modeling over larger image context with robust 3D landmark detection, jointly addressing ambiguities in conventional approaches, streamlining model application, and improving model-fitting and segmentation performance beyond handcrafted solutions. IEEE TMI paper describing this work
- I was a core developer and maintainer of the well-known MITK medical imaging interaction toolkit for 6 years (240+ code contributions).
- ADetect — Developed with clinical collaborators at the University Medical Center Mannheim: a robust, extensively validated DL pipeline for high-performance aortic dissection detection in CT, designed for emergency settings.
- Shape-model multi-organ segmentation — Developed a shape-based organ-labeling approach provided as a containerized standard tool across German university hospitals via the Kaapana/JIP ecosystem.