Biomedical Image Recognition

This junior research group will deal with central challenges of machine learning in the context of medical image and sensor data. These include: (1) Self-supervised learning methods to extract relevant markers from large image and signal data without expensive expert labels, (2) Improving robustness against so-called domain shifts, i.e. changes in data statistics caused by different hardware, pre-processing steps or acquisition parameters, (3) Cross-modal learning methods that combine image data with sensor data or textual information (findings). Image and biosignal data are becoming increasingly important in clinical diagnostics and research. This includes CT, MRI and microscopy images as well as EEG, ECG and long-term monitoring with the help of wearables. In pathology, entire tissue sections are increasingly being scanned. The UMG has a large, well-functioning PACS system and is a pioneer in digitalization in pathology. Various innovative microscopy methods are also used.