clean-IT: Towards Sustainable Digital Technologies

High-resolution medical image data (e.g., gigapixel microscopy images, three-dimensional MRI or CT scans) require computationally efficient AI algorithms. Current methods typically process these data exhaustively in small patches, although the object of interest, e.g., a tumor, may make up less than 1% of the overall image. At HPI scale-on-demand AI algorithms are developed that process medical images of any size and resolution, by processing down-sampled images and focusing expensive high-resolution computations on those parts in the image that likely contain an object of interest. Such an approach can save orders of magnitude in processing time, memory and energy while improving analytic performance. For example, for the detection of a local tumor in a microscope slide, this results in energy savings of around 99% compared to the state-of-the-art. More information...

Benjamin Bergner holds a Bachelor's degree in Business and Engineering and a Master's degree in Digital Engineering. He is currently doing his PhD in Machine Learning & Digital Health at the Hasso Plattner Institute at the chair of Prof. Dr. Christoph Lippert. His work revolves around the development of novel deep learning architectures in computer vision and medical imaging. In particular, application aspects in medical imaging are of his interest. For example, methods that allow interpretation of a neural network's classification decision and the development of efficient, large-scale deep learning pipelines are key research aspects that need to be addressed to support medical professionals in the future. In addition to his research, Benjamin leads workshops on artificial intelligence for small and medium-sized enterprises at the Competence Center Mittelstand 4.0 in Berlin.