Our research focus is on the application and development of artificial intelligence (AI) tools, such as deep neural networks, in the medical field. One of our main medical domains are head&neck, communication disorders, and neuroscience. We foster new biomedical data acquisition and processing techniques and work on TinyML neural networks. We further develop new datasets together with our clinical collaborators and work on the quantification of health states. As most of our work is clinically motivated, we seek solutions that impact diagnosing, treating, and monitoring patients. Please see also our publications.
Knowledge Domains
Embedded AI and tinyML
We are optimizing AI algorithms to fit on constrained computing devices, such as microcontrollers and Edge TPUs. For example, we use evolutionary algorithms with custom objective functions to propose the best deep neural network architecture for a given environment. We are one of the world’s first teams to create an end-to-end pipeline testing during the optimization the ability of a neural network to run directly on the target hardware.
The main purpose is to create lightweight, highly capable neural networks for wearables or remote devices. For example, we are developing a wearable sensor with our collaborators at McGill, Canada, that is able to detect airway symptoms in real time.
Computer Vision
We are experts in deriving a pixel-wise classification in a given image (semantic segmentation) with a focus but not limited to, biomedical images. We develop and investigate the best and most efficient methodologies using contemporary artificial intelligence methods. We contribute to recent advances in medical imaging and computer vision, by providing fundamental understanding in neural network design. We further use object detection and (markerless) tracking algorithms to compute animal or object trajectories. With this, we are able to uncover behavioral phenotypes of animals, cancer phenotypes, or single-cell neural activity patterns. Next to the data analysis itself, we create custom interfaces to efficiently label and annotate (biomedical) data to lower annotation costs.
XD biomedical signal analysis
We work on multidimensional (XD) biomedical data, such as audio and accelerometer data (1D), images (2D), videos or z-stacks (3D), as well as time-lapse imaging (4D). We explore new signal analysis and processing techniques to unravel feature importance and interpretation strategies.
Generative Artificial Intelligence
Large-language models (LLMs) and text-to-speech (TTS) systems are our specialty. We are using LLMs for medical data and electronic health records analysis, data standardization, and code co-development. We explore further the use of TTS systems for highly dynamic, personalized therapy methods in communication and speech development disorders.
Image acquisition techniques
We have experience in designing, building, operating, and evaluating microscopes, such as DMD-based, light-sheet, and two-photon microscopes, as well as custom behavioral rigs and endoscopy systems. We not only combine and create novel hardware components but also provide graphical user interfaces (GUIs) to interact with our tailored systems.
Industry collaborations
Please get directly in touch with us.
Funding
We are grateful to receive generous funding from
- Bundesministerium für Bildung und Forschung (BMBF)
- Staatsministerium für Wissenschaft und Kunst, Bayern (StMWK)
- Staatsministerium für Gesundheit, Pflege und Prävention (StMGP)
- Bayerische Forschungsallianz (BayFOR)
- Joachim-Herz-Stiftung
- Innovationsfonds Lehre
- KI Campus des Stifterverbands