Multimodal data fusion and analysis

This subject area is intended to research methods of artificial intelligence and machine learning for image generation as well as image and measurement data analysis. The aim is to support the sensor technology developed at Fraunhofer MEOS with regard to image pre-processing and image generation, image analysis and measurement data analysis using machine learning. This is intended to create added value and an adaptation to an application that goes beyond pure hardware development. Main focus is on applications in biomedicine.

In addition, there are a wide range of possible applications for the developed processes in other areas, e.g. in manufacturing and quality monitoring in the production of products with micro- and nanostructures, the quality inspection of optical surfaces or the control of very fast sorting processes using event-based cameras. Techniques for segmenting and classifying 2D image data as well as image enhancement and denoising are generally used in a wide range of areas beyond biomedicine. The topic combines all activities for the software-supported evaluation and analysis of sensor and image data.

Denoising methods for quantum-Inspired imaging and microscopy

Quantum imaging that looks like a happy face
© Fraunhofer IOF
Quantum imaging that looks like a happy face

Quantum imaging is an innovative technique with the potential to significantly improve biomedical imaging. However, since it is affected by an extremely low signal-to-noise ratio, subsequent image enhancement is essential.

Therefore, noise reduction algorithms are being developed at Fraunhofer MEOS using the example of biomedical imaging. This occurs using prior knowledge about the sample.  The combination of quantum imaging with AI-supported denoising for medical technology applications is a unique selling point. The denoising method can also be used in traditional imaging, such as microscopy. 

Event-based cameras for high-throughput cell analysis

In contrast to conventional cameras, in which all pixels are recorded simultaneously, in event-based cameras only individual pixels react to changes in brightness. That’s why only changes in the environment, i.e. moving objects, become visible.

MEOS will be the first to use event-based cameras in the field of life sciences to take technical advantage such as high dynamic range, no motion blur, latency in the microsecond range and a minimized amount of data. For this purpose, analysis programs and algorithms are developed in applications such as image-based flow cytometry or real-time monitoring of biological processes (e.g. cell secretion profiles) in automated ATMP production and lab-on-a-chip applications.