News Category: Press Release

“With ‘AISA’, students from a wide range of disciplines are given the excellent opportunity to train to become top AI software experts. The talents gained in this way can work on future-oriented and practical solutions in the development of AI software and thus further advance the international appeal of Cyber Valley,” says Baden-Württemberg’s Science Minister Theresia Bauer. “Through the AI Software Academy, we are accelerating our international competitiveness in terms of digitalization and artificial intelligence across all disciplines by closing a previously unaddressed gap in education,” continues Prof. Wolfram Ressel, Rector of the University of Stuttgart. “In this way, AISA makes a twofold contribution to our vision of intelligent systems for a sustainable society.” Prof. Stefan Wagner, Head of the AISA Board of Directors and Director at the Institute of Software Engineering at the University of Stuttgart confirms, “Through our innovative education and research concept in combination with the powerful AISA infrastructure, we will significantly advance artificial intelligence and its use by means of powerful software.”

AISA is dedicated to the challenges in business and science

Rapidly advancing digitization and the extensive use of AI are influencing our everyday lives, as well as business and science. This results in many new types of challenges:

• How is software for AI applications developed efficiently?
• How do you make AI software robust and flexible?
• How do you apply AI software in business and academia?
• And how do you build didactics that is able to transfer knowledge about the development, use and maintenance of AI software into many different disciplines and thus shape the AI experts of tomorrow?

In AISA Research, projects from a wide variety of fields are funded

In AISA Research, interdisciplinary research projects around the topic of AI are funded. Applications (X), software development (SE) and their didactics are linked with AI to the following: AI+SE+X. The AISA Board of Directors selected the first eleven Ph.D. projects for AISA Research in August 2021. The integration of the different fields can be seen in the funded projects: Researchers from architecture, biology, chemistry, didactics, computer science, mechanical engineering and environmental engineering cooperate. Projects closely related to the Clusters of Excellence SimTech and IntCDC are also represented.

The doctoral researchers from AISA Research offer targeted teaching content for students in the master's area and for Cyber Valley doctoral students by means of AISA Training. In addition, project work, seminars and hackathons are organized and supervised here. The additional qualification is structured by an innovative, multi-level certificate program, which allows for a precise match to individual qualification needs, from thematically focused microcredentials to the Diploma of Advanced Studies.

AISA consists of three components: Research, Training and Infrastructure

The structure of AISA is based on the three elements of research (AISA Research), training (AISA Training) and a powerful infrastructure (AISA Infrastructure), which form the AISA Triangle. Both AISA Research and AISA Training rely on the AISA Infrastructure, which provides AI-specific high-performance computing. Good availability and easy access allow methods and AI pipelines to be developed efficiently. AISA is managed by a board of directors consisting of Prof. Stefan Wagner (chair), Tenure-Track Prof. Maria Wirzberger, Prof. Steffen Becker, Prof. Steffen Staab, and Prof. Felix Fritzen, who in turn represent the different disciplines in AISA.

About the funding from the state of Baden-Württemberg

AISA is another project from the “Zukunftsland BW – Stärker aus der Krise” package of measures. As part of this package, the state government is making targeted investments in important fields of the future, thereby sustainably strengthening the state as a location for business and innovation. The program lays the foundation for Baden-Württemberg to emerge stronger from the corona crisis.

Related links:

• Cluster of Excellence Data-Integrated Simulation Science (SimTech)
• Stuttgart Center for Simulation Science (SC SimTech)
• Cluster of Excellence Integrative Computational Design and Construction for Architecture (IntCDC)
• Mission statement and strategic goals of the University of Stuttgart

“We are very thrilled to welcome the four new units to the ELLIS network,” said Dr. Nuria Oliver, co-founder and vice-president of the ELLIS Society as well as Chief Scientific Advisor at the Vodafone Institute, Berlin. “We have now reached a critical mass at which the units are also effectively starting to collaborate with one another. Instead of speaking of stand-alone units, we’re delighted to see the network of units thrive.”

The ELLIS unit in Jena, Germany, involves three research institutions: the Friedrich Schiller University Jena, the Max Planck Institute for Biogeochemistry, and the Institute for Data Science of the German Aerospace Center. At the interface between climate and environmental science as well as machine learning (ML) their emphasis lies on scientific knowledge integration in and generation from machine learning approaches. “The ELLIS Unit Jena now being part of the European network offers access to excellent AI expertise in Europe, leading to fantastic new avenues of interdisciplinary collaboration. AI is extremely relevant for earth, environmental and climate science, which are the main themes of our unit,” said Prof. Markus Reichstein, one of the two directors of the ELLIS Unit Jena and director at the Max-Planck-Institute for Biogeochemistry as well as at the Michael-Stifel-Center Jena for Data-driven and Simulation Science Jena. “This unit manifests the successful collaboration of the involved parties over the past ten years. We see the ELLIS Unit Jena as central node in the network, with the potential to foster the scientific breakthroughs which, through linkage to policy and business, eventually contributing to a sustainable development of our planet,” added Prof. Joachim Denzler, director of the Institute for Data Science of the German Aerospace Center and Chair of the Computer Vision Group at the University of Jena. 

In Stuttgart, Germany, the newly established ELLIS unit is situated at the University of Stuttgart. Here the research focus is on interactive intelligent systems, natural and programming language processing, learning theory, and robot learning. One of the two unit’s founding directors and head of the Institute for Stochastics and Applications at the University of Stuttgart, Prof. Ingo Steinwart, said, “The ELLIS selection process is led by top scientists from different countries and is highly competitive and highly selective. The fact that there is now the ELLIS unit in Stuttgart not only underlines the excellence of the location with its research foci, but also in particular that of the researchers involved.”

“Being accepted into the ELLIS network is a great success for our location in these important AI future areas, which will significantly increase our visibility within Cyber Valley, nationally and internationally,” added Prof. Andreas Bulling, Professor of Human-Computer Interaction and Cognitive Systems from the Institute of Visualization and Interactive Systems at the University of Stuttgart, and second founding director of the ELLIS unit in Stuttgart. “The chances of success for future research proposals as well as for attracting young researchers at all levels of the scientific career ladder increase.”

The ELLIS unit in Milan, Italy, brings together excellent researchers from four institutions: Bocconi University, Politecnico di Milano, University of Milan, and University of Milan-Bicocca. The unit's main research areas are interactive learning, statistical learning and non-convex optimization, health and computational biology, as well as natural language processing and computational social sciences. Additional areas they intend to investigate include sequential decision-making and algorithmic game theory, as well as neural networks in connection with classical AI and neuroscience. “The ELLIS Milan unit is the result of a joint effort, combining the research groups of four major universities that have complementary skills and strong interdisciplinary components. Thanks to this recognition, Milan is now firmly on the map of European AI excellence,” said Prof. Nicolò Cesa-Bianchi, director of the ELLIS Unit Milan and Professor of Computer Science at the Department of Computer Science of the University of Milan. “The decision to join forces under the ELLIS framework is also a very timely decision, given that the Italian AI strategy is beginning to unfold. The goal is to fuel collaborations between the research groups, taking advantage of their complementary strengths.”

The new UK ELLIS unit is located at the University of Manchester. It brings together experts in AI fundamentals with experts in other fields. Their research focus will be probabilistic modelling and Bayesian inference, AI technologies that work better with people, ML for digital health and medicine, and privacy-preserving ML. “The University of Manchester continues to grow as a center of excellence for AI research and the new ELLIS unit will further strengthen this activity,” said Prof. Magnus Rattray, Director of the ELLIS unit Manchester and Professor of Computational & Systems Biology and Director of the Institute for Data Science and AI at The University of Manchester. “Through the ELLIS Unit Manchester we will be able to better link ML researchers across Europe with impactful applications in biomedical and health research.” The University of Manchester has established a strategic partnership in a shared professorship with the director of ELLIS unit Helsinki, Prof. Samuel Kaski from the Aalto University, Finland. This northern link will be utilized in the future to set up the ELLIS unit Manchester and the ELLIS unit Helsinki as a twin unit, with tight collaboration already under way through joint research and exchange, which will be opened up to the rest of the ELLIS network.

About ELLIS

ELLIS (European Laboratory for Learning and Intelligent Systems) is a European AI network of excellence initiated in April 2018. Its central goal is to foster European research excellence in machine learning and related fields by offering scientists outstanding opportunities to carry out their research in Europe, and to train the next generation of young European researchers in this field of strategic importance. ELLIS has a three-pillar strategy to maintain and foster European technological sovereignty in this competitive field. The three pillars focus on fellows, students and locations. The aim is to invest in existing talent, foster new talent, and build a network of outstanding European innovation ecosystems. The ELLIS units are selected on the basis of scientific excellence by a committee of leading scientists from several different countries. At the moment, ELLIS consists of 34 units in 14 European countries and Israel. More information about the units and their scope can be found here: https://ellis.eu/units.

Media contact

Daniela Diaconu
c/o Max Planck Institute for Intelligent Systems
Max-Planck-Ring 4
72076 Tübingen, Germany
Email: daniela.diaconu@ellis.eu
Phone: +49 7071 6011541

While young surgeons’ training will always require the guidance and experience of a senior surgeon, artificial intelligence (AI) assisted training methods may one day provide valuable support. By drawing on data, such technologies could enhance training by complementing instruction with information on how a tumor could best be removed from healthy tissue.

Until now, surgical education has been challenging at multiple levels: not only do young surgeons have limited time to practice, they must be supervised by senior surgeons whose schedules are often tight. At the same time, human tissue is hard to come by, and using animals to train students raises ethical concerns. Practicing on patients is also out of the question, as the risk of something going wrong or not meeting medical standards is too high.

Development of artificial organs as training environment

In response to these challenges, researchers at the University of Stuttgart at the Institute of Physical Chemistry have created educational tools that could one day be used to train surgeons, much like flight simulators train pilots. The team, which comprises engineers and medical doctors, have developed a range of artificial organs – among them a liver, a kidney, and a bladder. With these artificial organs, the scientists aim to create a training platform that provides high-fidelity surgical scenes and a quantitative and objective assessment of a trainee’s skills in real time. Ultimately, the scientists want to develop organs that are easily reproducible and standardized, as this would enable their use in the training of surgeons around the world.

The artificial organs are known as cyber-physical organ twins, which are physical systems combined with virtual features. The consistency of the tissue, the anatomy, the haptic perception, even the look of the tiny blood vessels are very similar to natural organs. Additionally, sensors measure a range of parameters, including the surgeon’s fine motor skills when moving an endoscope through the organ, whether a needle has punctured correctly at the required location, or the time needed to localize all the tumors. An open-source algorithm that is also applied in self-driving cars to distinguish between objects has been re-trained to detect the lesions, such as carcinomas or kidney stones. Once the endoscope reaches the realistic target, the trainee can simulate a biopsy of the protrusion and try to improve the accuracy and efficiency with each session. All the while, data is collected that would not otherwise be available, even in a real surgical procedure on a patient.

Training environment makes evaluation of operations possible

“If we look at computer vision or natural language processing, the data is very rich and structured. But in the area of biomedicine, especially in surgeries, data collection is very difficult. How can the quality of a surgical procedure be quantified? A real-life surgery happens in an environment in which collecting real-time data is almost impossible,” says Tian Qiu, leader of the Cyber Valley Biomedical Microsystems research group at the University of Stuttgart. The group, which focuses on integrating hardware with AI, develops tools that collect large amounts of data in biomedicine in order to learn from it and improve minimally-invasive surgical tools and procedures. The team collaborates closely with physicians from the university hospitals in Freiburg and Tübingen.

“As we aim to enhance the field of biomedicine with artificial intelligence, we combine physical systems with virtual features. This enables us to collect data on surgical skills. Experienced surgeons lay the foundation and then we compare their skills with those of future surgeons. In each training session, we visualize a trainee’s data, assess how the surgery was performed, and evaluate which skills need to be improved,” says Tian Qiu.

The team recently published the results of its work in three journal papers and one conference paper. In one of these papers, they present a soft liver phantom that can assess how well an endoscopic inspection and tumor biopsy is performed inside the hollow biliary ducts that weave through the liver. Creating a realistic soft liver replica is a challenging task. “Our organ phantoms offer much more objective evaluations and feedback, much better than virtual reality simulators can, for instance,” says Xiangzhou Tan, who is a physician at the University Hospital Tübingen and brings medical expertise to the Biomedical Microsystems group.

Manipulating an endoscope, a long and thin instrument, equipped with a miniature camera at the tip, operated outside the patient’s body, is another important part of a surgeon’s training. “One prototype we developed can measure specific endoscopic controlling skills,” explains Do Yeon Kim, a Ph.D. student and engineer in the group. “When a trainee navigates through an organ, many factors are considered. For instance, they must assess how long it takes them to screen all areas inside the organ and make sure they haven’t missed anything and that they have detected all the tumors in a reasonable amount of time.” To evaluate manipulation skills, Do Yeon Kim trained the algorithm with data that experienced doctors provided. A trainee’s performance is then measured against these thresholds. “One could say we have created an advanced flight simulator for surgeons.”

References of recent publications on the topic of cyber-physical organ twin:

1. Liver phantom and needle puncture skill assessment
   Tan, X., Li, D., Jeong, M., Yu, T., Ma, Z., Afat, S., Grund, K.-E. & Qiu, T. Soft Liver Phantom with a Hollow Biliary System, Ann. Biomed. Eng. 2021. https://doi.org/10.1007/s10439-021-02726-x
2. Bladder phantom with blood vessels and tumors for endoscopy
   Choi, E., Waldbillig, F., Jeong, M., Li, D., Goyal, R., Weber, P., Miernik, A., Grüne, B., Hein, S., Suarez-Ibarrola, R., Kriegmair, M. C. & Qiu, T. Soft Urinary Bladder Phantom for Endoscopic Training. Ann. Biomed. Eng. 2021. https://doi.org/10.1007/s10439-021-02793-0
3. Endoscopic training and skills assessment
   Tan, X., Kim, D. Y., Jeong, M., Li, D., Miernik, A. & Qiu, T., Computer-Assisted Quantitative Assessment of Endoscopic Skills based on a High-fidelity Endourological Phantom. Under revision.
   Kim, D. Y., Tan, X., Jeong, M., Li, D. & Qiu, T., Computer-Assisted Quantitative Assessment of Endoscopic Skills based on a Urological Phantom. 55th DGBMT Ann. Conf. on Biomed. Eng. Accepted.

Scientists use super-resolution microscopy to study previously undiscovered cellular worlds, revealing nanometer-scale details inside cells. This method revolutionized light microscopy and earned its inventors the 2014 Nobel Prize in Chemistry. In an international collaboration, AI researchers from Tübingen have now developed an algorithm that significantly accelerates this technology.

Single-molecule localization microscopy (SMLM) is a type of super-resolution microscopy. It involves labelling proteins of interest with fluorescent molecules and using light to activate only a few molecules at a time. Using this trick, multiple images of the same sample are acquired. To create a meaningful picture, a computer program unscrambles the data and compiles the complete image. While the technique can be used to locate molecules with high precision, it has one major drawback – it requires scientists to acquire a large number of images, which makes the process very time-consuming.

In an international collaboration, the team of Jakob Macke, Professor for Machine Learning in Science at the University of Tübingen, has developed a new algorithm that overcomes this limitation of SMLM. The joint work with the Ries Group at the European Molecular Biology Laboratory (EMBL) Heidelberg and Dr. Srinivas Turaga’s team at Janelia Research Campus (Virginia, USA) was published in Nature Methods.

Deep learning enables highly accurate single-molecule localisation

The DECODE (DEep COntext DEpendent) algorithm is based on deep learning: It uses a neural network that learns from training data. Instead of using real images, however, the network in this case is trained with synthetic data generated by a numerical simulation. By incorporating information about the microscopic setup and the imaging physics, the researchers achieved simulations that closely matched real-world acquisitions. “The neural network that we trained using simulated data can thus also detect and localize fluorophores in real images,” explains Artur Speiser, who, together with Lucas-Raphael Müller, was the lead author of the paper.

One of the benefits of DECODE is that it accurately detects and localizes fluorophores at higher densities than were previously possible. This means that fewer images are needed per sample. As a result, imaging speeds can be increased up to tenfold with minimal loss of resolution. In addition, DECODE can quantify uncertainties – so the network itself can detect when it is unsure of its localization.

Interdisciplinarity expands the perspectives of research

“This work is indicative of the approach of our Cluster of Excellence ‘Machine Learning: New Perspectives for Science’”, Macke says, whose chair is part of the Tübingen cluster. “We originally developed the ideas underlying the machine learning approach in a very different context, but through collaborating with experts in computational microscopy we were able to turn them into powerful methods for analyzing SMLM data.”

The team has also built a software package which implements the DECODE algorithm. “The software is simple to install and free to use, so we hope it will be useful for many scientists in the future”, adds Dr. Jonas Ries from EMBL.

Contact:

Prof. Dr. Jakob Macke
University of Tübingen
Cluster of Excellence – Machine Learning for Science
jakob.macke@uni-tuebingen.de