DFKI und Inria auf der Vivatechnology DFKI and Inria at Viva Technology 2023 The 60th anniversary of the sign- ing of the Elysée Treaty occurs in 2023. The entire year is dedicated to Franco-German friendship and understanding. DFKI and Inria are marking the an- niversary year at the Franco-Ger- man Lab (booth D11) at the Viva Technology Tech and Startup Fair, June 14-17, 2023. Together with eight other academic partners and the states of North Rhine-West- phalia and Saarland, they present exhibits of current research topics and partner projects, and introduce selected startups. Inria and DFKI are collaboration partners under the terms of a joint strategic research and innovation agenda in AI for Industrie 4.0, AI in- frastructures, AI and cybersecurity, human-robot collaboration, and AI in healthcare. As a driving force for closer Fran- co-German cooperation in the field of artificial intelligence, Inria and DFKI provide a network for Ger- man and French AI researchers to promote respect for fundamental European values and to advance AI research that benefits humans. Visit us at the Viva Technology Fair! www.vivatechnology.com www.vivatechnology.com www.vivatechnology.com

04 C O N T E N T D F K I N e w s 1 / 2 0 2 3 m o c . e b o d a . k c o t s , l s b a n e l e © 06 Explainable AI in Medicine 16 Ethical and Legal Challenges in the Use of Artificial Intelligence in Clinical Decision Making 17 AI-assisted Therapy Support for Tumor Patients Using the Example of Urology 18 SAIL – DFKI‘s Biotech Transfer Lab 20 Medical Projects From the Research Department Smart Data & Knowledge Services: Analysis of Genomic and Proteomic Data, Medical image analysis, Computer-aided Diagnosis and Medical Forensics m o c . e b o d a . k c o t s , ff o k n e d o r o G ©

06 D F K I N e w s 1 / 2 0 2 3 EXPLAINABLE AI IN MEDICINE The German government wants to be sure that Germany and Europe are seen as global leaders in the field of artificial intelligence (AI) when it comes to the level of AI research and development, including application development. The cornerstones of the AI strategy were published for the first time at the end of July 2018. DFKI research has always focused on the application of AI, and is now asking these questions: What are the implications of AI in healthcare? Where does it have the most potential, and what are the limits? How important is explainability? These questions are being considered by DFKI‘s Interactive Machine Learning (IML) department against the background of the initial situation, fields of action, and transferability to medical care applications. Clinical decision support is currently of great importance. It is expected that AI research will become firmly estab- lished in this area over the next two to three years with the availability of high-performance computing and a fo- cus on special realistic tasks, for example, in dermatology for pattern recognition using more than 100,000 images or more than 80,000 x-ray scans. But first, solutions have to be found for three substantial challenges: predictions that are robust, decision support that covers the range of illness cases, and the interpretability of the treatment de- cision. The quality and availability of suitable training data are key to the first two challenges. AI is highly significant in clinical decision support systems for medical professionals. Computer-aided assistance systems that support the physician or the patient in the early interpretation of illnesses promise much-needed relief for patients, physicians, and the healthcare system as a whole. AI applications, especially for specific onco- logical diseases, are eagerly awaited. Dermatology and its sub-diseases is just one example where access to suf- ficient data and integration in complex medical services is essential for the successful use of AI systems in clinical and non-clinical settings.

08 D F K I N e w s 1 / 2 0 2 3 m o c . e b o d a . k c o t s , t o h S - l e x i P © e r e e D n h o J © DETECTION AND OPTIMAL TREATMENT OF EYE DISEASE Ophthalmo-AI focuses on the two most common causes of blindness in people aged 50 and older: Age-related macula degeneration (AMD) and diabetic retinopathy (DR). An intelligent assistance system is under development to support ophthalmologists using image data and clinical data to reach a correct medical diagnosis and determine the best possible treatment. INTELLIGENT, COOPERATIVE SUPPORT FOR DIAGNOSIS AND THERAPY IN OPHTHALMOLOGY Age-related macula degeneration is a result of the aging process in the cells of the central retina (macula = point of sharpest vision). As every person ages, cellular debris accu- mulates in this area and, sometimes, irregular blood vessels grow from the choroid behind the retina. Fluids and blood can leak from them and cause the pathology of “distortion” and progressive vision loss (“wet” form of “late” AMD). Diabetic retinopathy can occur as a result of diabetes. Similarly, blood components can leak into the retinal tis- sue, causing pathological vascular neoplasms – so-called “proliferations” – to form on the retina. These are very vulnerable vessels that can cause strong hemorrhages in the eye, which may result in retinal detachment. Diabetic retinopathy often leads to edema (= swelling) in the area of the macula, a so-called „diabetic maculopathy,“ which causes vision problems. Optical Coherence Tomography, (= cross-sectional imag- ing of the retina) is able to detect the smallest changes and swellings in the area of the macula, such as the presence of irregular blood vessels. This non-invasive procedure is essential in the diagnosis and monitoring of the treatments of both diseases. New drugs available today suppress the Segmentations relevant to diagnosis.

m o c . e b o d a . k c o t s , o i d u t S C D © 10 D F K I N e w s 1 / 2 0 2 3 AI DEVELOPED FOR CLINICAL ROUTINES PROJECT pAI To broaden the use of AI methods in medicine, networked, digital patient data is required, for example, for use in disease diagnosis, personalized treatment, and faster drug discovery. A collaborative project called pAItient (AI innovation center with integrated, rights-protected environment for the development, testing, and clinical assessment of Al-based applications) is establishing an innovation center for artificial intelligence at the University Hospital Heidelberg to provide a unified research infrastructure for the integration of AI solutions in healthcare. The goal is to automate the development of AI applications from the initial idea to the introduction in everyday clinical practices. Using transfer learning methods, i.e., the transfer of learned knowledge to another area, researchers at DFKI are study- ing automated diagnostic techniques with detection based on image data. The project also looks at ways to integrate other AI methods into the infrastructure. Testing and val- idation are ongoing in two areas: First, the extraction of medically relevant data from unstructured texts, and second, the analysis of ultrasound images of the carotid artery with regard to the risk of a heart attack. The focus is on continuous improvement of the model by verifying the results through the involvement of users in correct- ing the results. A sample use case A class of machine learning model that generates new data based on patterns observed in a training dataset in the field of natural language processing (NLP) is called a generative model. In this context, project pAItient aims to use generative models to produce concise summaries of PROJECT DESCRIPTION Consortium • University Hospital Heidelberg (UKHD) • DFKI • German Cancer Research Center (DKFZ) • MINT Medical Financial Sponsor Federal Ministry of Health (BMG) Volume 2.1 million euros Term 10/1/2020 - 9/30/2023 radiology reports – for example, when radiology detects that a heart attack is imminent. Various techniques are being investigated to improve the comprehensibility of the generated summaries. Comprehensibility is checked by contrasting the generated summary with the original content of medical texts by highlighting the relevant input data, which also serves the purpose of verification. While the summaries generated by generative models achieve high static accuracy, in some cases, they are hard to follow or are not easily understood. This makes them unsuitable for the intended audience. A review by medical profes- sionals is intended to address this issue. Contact Hans-Jürgen Profitlich Research Department Interactive Machine Learning hans-juergen.profitlich@dfki.de +49 681 85775 5265 An example of an automated, comprehensible radiology report summary. During generation, the generative model extracts important segments (in green) from the original text (left) and inserts them into the summary (right).

m o c . e b o d a . k c o t s , e n a m a h k g n a S k a s g n o h P © 12 D F K I N e w s 1 / 2 0 2 3 MEDICAL DATA FOR RESEARCH According to a FORSA survey, more than three quarters (79 percent) of Germans are willing to anonymously provide their medical data for medical research. Nevertheless, an absence of medical data remains one of the main reasons why AI has been slow to enter the field of medicine as opposed to other industries. Easy and secure anonymization, synthetic data, and data trust structures have the potential to advance the research and development of AI in healthcare and cut through the knot of data availability. The media attention around AI-driven innovation and dis- ruptive change has increased with the rise of ChatGPT. But, in most industries and areas of life, the changes are already part of the daily reality. The major areas for potential medical applications are in drug and therapy development, diagnos- tics, and decision support to assist in nursing and adminis- trative activities. However, compared with other sectors, AI application research and development is still in its infancy in the medical field. One of the main reasons for this is the lack of availability of data to build and train AI systems. Al- though large amounts of data are available in health infor- mation systems and a high percentage of the population would anonymously provide their data to medical research, genuine training data are still scarce. This is a showstopper for AI systems research in medicine. Building trust Preparing the data for research requires trustworthy struc- tures that simplify and standardize the process of data ac- m o c . e b o d a . k c o t s , a fi a r g o t o f e n o t © quisition. Data trustees mediate between the providers of the data (the persons affected by the data) and the institu- tions using the data, for example, a research institute, ef- fectively building bridges of trust. A high level of protection must be ensured when sharing personal data. Project DaTreFo is creating a data trust with protected data stores for research using easily understandable organ- izational and technical methods. The aim is to establish a high technical barrier against data misuse and create the trust required for a greater willingness to donate data vol- untarily. In recognition of their right to informational self- determination, the methods used to achieve this include the enforcement of data donors‘ requests for revocation and encrypting available data for storage in a form that the data trustee cannot decrypt. The DaTreFo solution is innova- tive and designed to offer consistent encryption and integrity-guaranteed storage of „dormant“ data and ena- ble recoding for authorized purposes without interim disclosure. The DaTreFo project fully complies with the separation between data storage and data users as laid out in the Data Governance Act of the EU Commission. The result provides a general-purpose data trustee architecture for secure data retention. DFKI Trier is working with the Trier University of Applied Sciences and Dedalus, a manufacturer of hospital information systems, to develop a generic solution for use in hospitals to facilitate access to medical data, especially for AI research. Anonymization and synthetic data „Genuine“ data are a real treasure for AI research in medicine. One way to share data in a trustworthy and legally secure

14 D F K I N e w s 1 / 2 0 2 3 INTERACTIVE AI FOR PREDICTIVE AND FLEXIBLE MANAGEMENT OF DISCHARGE AND TRANSITION OF PATIENTS Follow-up care planning in hospitals is a complex, dynamic, and work-intensive process. Artificial Intelligence can improve the quality of the transition to post-discharge care and is key to reliable and continuous patient aftercare. Hospitals have an obligation to provide ef- fective discharge management to support the transition to aftercare. They must en- sure patients will continue to receive appropriate medical care after discharge, while taking social considerations into account. In some cases, if the planning is not continuously adjusted by the discharge management team, the start of aftercare does not match the time of discharge, and after- care capacities may go unused, or the inpatient stay is pro- longed. For the discharge management team, this means increased demands on documentation and communica- tion with patients, relatives, medical staff, and follow-up treatment providers. An AI-supported system is being developed to assist in the management of patient discharge and transition by DFKI and its partners (Nubedian, Empolis, Furtwangen Uni- versity of Applied Sciences, Bonn University Hospital, and Mannheim University Hospital) in Project KIAFlex (Interactive AI-assistance for predictive and flexible control in discharge and transfer management). The project relies, among other things, on case-based reasoning, a machine learning meth- od, to identify individual aftercare needs at an early stage while assisting clinical staff with planning, documentation, and communication. Adapted from analogy-based clinical reasoning, solutions are found based on similar past cases and then adapted for the current patient. This AI technology is particularly useful in maintaining explainability in processes involving small amounts of data. Both of which apply to the increased re- quirements profile of our application scenario. The challenges in the project include predicting the dis- charge date and the need for aftercare; adapting the after- care planning; facilitating the documentation via recom- menders, speech recognition, or templates; relieving the communication workload, for example, through chatbot technologies; and, the integration of solutions into exist- ing medical systems. CLINICAL CARE PROCESS Primary systems Semantically integrated patient data AI-based continuous prediction of aftercare needs Flexible semantic process control and interactive communication PLANNING COMMUNICATION DOCUMENTATION Discharge and transition management More information www.kiaflex.de Contact Prof. Dr. Ralph Bergmann Head of Experience-Based Learning Systems Topic Field Research Department Smart Data & Knowledge Services ralph.bergmann@dfki.de +49 651 201 3876 Joscha Grüger Topic Field Experience-Based Learning Systems Research Department Smart Data & Knowledge Services joscha.grueger@dfki.de +49 651 201 4166

16 D F K I N e w s 1 / 2 0 2 3 ETHICAL AND LEGAL CHALLENGES ETHICAL AND LEGAL CHALLENGES IN THE USE OF IN THE USE OF ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE IN CLINICAL DECISION MAKING IN CLINICAL DECISION MAKING m o c . e b o d a . k c o t s , y fl r e t t u b n e e r g © Artificial Intelligence is gradually making inroads in the medical field. This trend is accompanied by a whole range of ethical and legal questions. How does the use of AI in diagnostic procedures change the role of physicians and patients? Who is responsible for the consequences of AI-supported processes in medical contexts? These questions are addressed in the vALID project by an interdisciplinary research team funded by the German Federal Ministry of Education and Research Computer-aided, automated, and self-learning systems in a clinical care environment promise to optimize diagnosis and therapy decisions reliably and in a resource-ecient manner. Deep Learning is the key enabling technology. However, such methods often seem to function in a black box: they are bet- ter than humans at recognizing patterns in extensive data, but because they are so complex, they remain opaque for users and, to a certain degree, even to software engineers. after kidney transplantation and has prepared a study to evaluate it. The system is designed to predict potential risks such as graft failure, organ rejection, and infection follow- ing a kidney transplant. To make the AI‘s predictions under- standable, a dashboard was developed that visualizes the risk assessment and provides an explanation of the results. To gain insights into the user reactions to an AI system, the study also included so-called Wizard-of-Oz experiments, in which simulations are used. Recommended uses of AI in healthcare Technical experts from DFKI are supporting a project con- sortium that uses AI-based systems in renal medicine to prepare a normative, legal, and technical analysis of clini- cal, AI-driven decision support systems. The consortium is made up of ethicists around Prof. Peter Dabrock at Friedrich- Alexander University Erlangen-Nürnberg (FAU), legal ex- perts around Prof. Susanne Beck at Leibniz University Hannover (LUH), and physicians around Prof. Klemens Budde at Charité University Hospital Berlin. The team around Prof. Sebastian Möller, Head of Speech and Language Technology research at DFKI Berlin, was responsible for the sub-project tasked with the technical implementation of the AI decision support. The empirical insights gained in this way suggest that the use of an AI system should not exclude established structures of decision-making. Both the user group of healthcare pro- fessionals and patients ex- pressed a desire to control the system. Project results are de- scribed in the German language book “Klinische Entscheidungs- findung mit Künstlicher Intelli- genz,” published by Springer. “To exploit their potential for use in the ethically highly sensitive field of medicine in everyday clini- cal practice, AI systems must encourage participa- tory decision support while meeting the require- ments of trust, transparency, and accountability.” Dr. Aljoscha Burchardt, DFKI-Laboratory Berlin The team is working on the advanced development of an existing clinical decision support system for follow-up care More information www.dfki.de/en/web/research/projects-and- publications/project/valid https://link.springer.com/book/9783662670071 Contact Dr. Aljoscha Burchardt Research Department Speech and Language Technology aljoscha.burchardt@dfki.de +49 30 23895 1838

18 D F K I N e w s 1 / 2 0 2 3 SAIL DFKI’S BIOTECH TRANSFER LAB In 2019, DFKI partnered with Sartorius, a global partner in life science research and the biopharmaceutical industry, to establish a joint research lab called Sartorius-AI-Lab (SAIL) in Kaiserslautern. Today, the lab is a prime example of application-oriented research performed courtesy of transfer labs. Sartorius and DFKI announced the opening of the Sartorius-AI-Lab (SAIL) on December 19, 2019. In a cor- porate press release, Dr. Joachim Kreuzberg, CEO of the Sartorius Group, was quoted at the time as saying, “Modern data analysis methods have only been used to a very limited extent in the biopharma industry, both in drug discovery and in production. This is . This is one of the reasons for the ever-longer ger development times and rising costs of f drugs. The use of AI holds the prom- ise of significant advances by fa- cilitating better data exploitation. Our Data Analytics business seg- ment develops powerful software for specific applications in biotech production, which many custom- ers have been using for the past two and half years. We want to expand our ur development efforts in this area and are d are pleased to have found in DFKI a strong part- ong part ner, with more than 1,000 scientists and home to one of the most important international Centers of Excellence for AI and Deep Learning.” At DFKI, work began with one employee and one Ph.D. student. Currently, nine people at DFKI are working di- rectly or indirectly on issues relating to Sartorius. Patent applications have already been submitted regarding the research results, and products can also be found in the Sartorius product portfolio. Beyond this, the results have motivated DFKI to carry out numerous have mot other independent studies into broader other issues raised by Sartorius. issu O One of the priority aims at SAIL is to develop AI-assisted analysis meth- ods for living cell cultures under light microscope imaging, an essential tool for diagnosing illnesses and t developing drugs. For example, when d drug testing, the drug‘s effects on the dr cell culture are quantified using param- cell eters such as cell count, growth (or pro- eters su liferation rate), and cell shape. Recognizing liferation ra the cell outlines in the often very low-contrast images (see figure) and labeling and tracking the individual cells over a period of time is extremely tedious and time- consuming work for a human expert. m o c . k i p e e r F , r o t c e v o r c a m ©

20 D F K I N e w s 1 / 2 0 2 3 . m o c y z e e t c e v , e i t a j o n o m a r p u n s i w © MEDICAL PROJECTS FROM THE RESEARCH DEPARTMENT SMART DATA & KNOWLEDGE SERVICES ECGPlat – Computer-aided Sequence Analysis of Genomic and Proteomic Data The availability of biological sequence data from DNA, RNA, and peptides – the basic components of any living organism – has increased dramatically in recent years. These data open new windows into the processes between and within cells. Deep analy- sis is helping to decode the functional information hidden in the sequences and better understand the related processes in the body. This research enables us to track the development and progression of various diseases and holds the promise of better control treatments. Typically, the analysis of biological sequence data is performed in special laboratories. Advances in AI and Deep Learning now promise accurate, cost-effective, and rapid analysis as an alternative to the laboratory. DFKI has already developed several bio- logical sequence data analytics to understand how processes in living organisms function and predict diseases such as cancer. Such analysis enables the development of suitable drugs and vaccines to influence the course of the disease. More information https://sds_genetic_analysis.opendfki.de Contact Dr. Muhammad Nabeel Asim muhammad_nabeel.asim@dfki.de +49 631 20575 1010 Forensics DFKI provides its computational tools for DNA, RNA, and protein sequence analysis on ECGPlat (Ecient Computational Genomics and Proteomics Sequence Analysis Platform. Advanced Intelligence for Analyzing Medical Images – exAID Medical professionals must often evaluate MRI, CT, or X-ray images without having the benefit of technological support. Rec- ognizing and interpreting relevant patterns and anomalies using the human eye alone is time-consuming, prone to error, and requires years of experience. AI-based diagnostic systems provide an additional tool for correct diagnosis. Deep learning models are used for this and are often seen as a kind of „black box“ because the transfer of such systems from re- search to practical application is hindered by a lack of understanding of how they work. There is an urgent need to decode and explain AI-based findings to increase acceptance and provide a response to ethical and legal issues. The diagnosis assistance system exAID uses the latest advances in explainable AI to integrate existing AI models into every- day clinical practice. This is intended to make the physician‘s clinical workflows reproducible, more ecient, and more reliable. The exAID system enhances human expertise with diagnostic suggestions, textual explanations, and visual highlighting of rel- evant biomarkers. In addition, it provides practitioners with deeper insights into the use of AI in diagnostics, which is an inte- gral component of the workflow. The exAID system relies on high-performance AI models for diagnosing disease. One of the early applications of the system is for detecting skin cancer in dermoscopic images. Using relevant biomarkers (e.g., in pigmentation), malignant changes can be distinguished from harmless moles. More information https://exaid.kl.dfki.de Contact Adriano Lucieri adriano.lucieri@dfki.de +49 631 20575 5361 The exAID diagnosis wizard suggests a diagnosis of the dermoscopic image while also adding an accompanying explanatory text.

H b m G g n o r T © 22 D F K I N e w s 1 / 2 0 2 3 DFKI IS CORE PARTNER IN ATHERO THROMBOSIS AND INDIVIDUALIZED MEDICINE CLUSTER Cardiovascular disease has been the number one cause of death for many years. A breakthrough in combating this widespread disease is the common goal of DFKI in working together with TRON, Mainz University Medical Center, and other partners from the Rhine-Main region, such as BioNTech. The activities are initially scheduled for three years, and the Federal Ministry of Education and Research (BMBF) is providing funding of up to 15 million euros. Launched in March 2023, the “Cluster for Atherothrombo- sis and Individualized Medicine” (curATime) develops cus- tomized treatment methods and prevention concepts for cardiovascular diseases. This work involves state-of-the- art artificial intelligence methods used in biomarker re- search and in combination with immunological processes. The work in the cluster is being driven forward in four fields of innovation: curAIdent, curATreat, curATech, and curAInnovate, as well as in more than 15 parallel projects. Four of these individual projects are presented here as examples: curAIknow aims at capturing biomedical knowledge and at generating and applying knowledge graphs to support the prediction and validation of AI-generated assumptions. The opportunity to gain complete new insight comes when an existing knowledge graph covering almost all relevant public knowledge is used in combination with extensive data from healthy people and patients.

24 D F K I N e w s 1 / 2 0 2 3 IMPROVING THE QUALITY OF LIFE THROUGH AI PERSONALIZED HEALTH PROGNOSES FOR PEOPLE SUFFERING FROM CANCER AND COPD Machine learning is making great strides in many fields, yet its many potential benefits in medicine remain largely untapped. This is an area with diverse and very promising application potential. In particular, patients with life-threatening or chronic illnesses can be helped by virtual assistance systems and individualized predictions of disease progression. ASCAPE (Artificial Intelligence Support- ing Cancer Patients Across Europe) and RE-SAMPLE (Real-time Data Monitor- ing for Shared, Adaptive, Multi-domain and Personalised Prediction and Deci- sion Making for Long Term Pulmonary Care Ecosystems) are two new research projects within the EU Horizon 2020 framework program. Both aim to apply machine learning methods to predict any deterioration in patient health and threats to the qual- ity of life and to suggest interventions. Both projects draw upon DFKI‘s Cyber-Physical Systems research department. In ASCAPE, the focus is on prostate and breast cancer; RE-SAMPLE studies Chronic Obstructive Pul- monary Disease (COPD). A typical hurdle to machine learning in medicine is the lack of data. The data involved in healthcare is usually sensitive and private in nature, and it must be kept strictly

26 D F K I N e w s 1 / 2 0 2 3 AI AND ROBOTICS AI AND ROBOTICS IN HEALTHCARE IN HEALTHCARE DFKI DESIGNS INTELLIGENT CONTROLS AND CONCEPTS FOR ROBOTIC ARMS IN AN INNOVATIVE NURSING BED SYSTEM m o c . k i p e e r f , l e x i p w a r © About 9.4 percent of the population in Germany is classified as severely disabled by Federal Statistical Oce: approximately 16,000 people are living as paraplegics. For many of them, as well as those with muscular dystrophy, multiple sclerosis, amyotrophic lateral sclerosis, or a similar condition, interaction with their environment is severely limited. These people sit in a wheelchair and require a third party to open doors, grasp objects, or assist with personal hy- giene needs. One approach covered by German health insurance is an electric wheelchair mounted with an assistive robotic arm, operated from the wheelchair with a joystick. However, compared to operating a wheelchair, the ma- nipulation of a robotic arm is much more complex as the arm requires freedom of movement in three-dimension- al space. In current practice, the user can only control two of the seven possible directions in any one mode. If they want to change direction, they have to switch to an- other mode. This limits the potential of the robot arm and significantly delays movement because, in some cases, more than one sixth of the time is spent switching be- tween the various modes. DFKI‘s Cyber-Physical Systems department has joined forces with Frankfurt University of Applied Sciences, Westphalian University of Applied Sciences in Gelsenkirchen, and Munevo in a project called DOF-Adaptiv (Adaptive degree-of-freedom embedding as a cooperative user interface for an assistance robot). The project has the aim of developing a solution that provides a situation-dependent setting of the two controllable di- di- ble d rections. It relies on a camera in the robotic arm gripper rip per m grip and a neural network that extracts the most likely intend- ntend d- nd kely inten ed directions of movement from the camera data. This is ra data. This is is is designed to make the operation of the robot assist signifi- gn bot assist sign gnifi- nd r independ ndence. cantly easier and to promote greater user independence. dence e ce. c amera era rob obo ser i use th pot potential y w wit l p pri The project partners work closely with potential primary rimary t signif- and secondary users to develop five scenarios that signif- s tha hat scenarios th fiv ve s for eat- os f e for e icantly impact everyday life, including scenarios for eat- ding scenario c inc clu lud r e of nt ent of the ing and drinking and opening doors. The movement of the s. The movem doo oo do ors me t o ofof of e th th he scenarios s w robotic arm was recorded while the scenarios were run in were run in ere e ru re both simulations as well as in reality. The control of the robot arm was executed by people who are not motion impaired, following the assumption that their movement of the human arm would yield the preferred movements for the robotic arm. The resulting data sets serve as the basis for the probabilistic output of the neural network. In operation, the system recognizes the situation and ap- plies the expected movement directions to the control options available in the joystick input device. Thus the controls developed in this project are designed to offer the user a range of freedom instead of an autonomous path for some hypothetical purpose. The approach cor- responds to the explicit wish of various test subjects who want some independence in carrying out everyday activ- ities, not an alternative to their caregivers. independen ive t lternative , not an al al tive n alternativ ve to bora prp this ide multi-com dd A lleedledaal eaallecal lllaalaca rrojojo eccect t call otiiccotic cc s c obboto ii tteysstyst ssyyssys mmeemmemtteeteyyst Ano Another collaborative project called AdaMeKoR (Adap- Another colla tive multi-component robotic system for healthcare) e mu ve iv ive m tive ve ve takes t takes this idea one step further and involves the follow- a a akees es ing organizations: Johanniter-Unfall-Hilfe, University of ing or or Oldenburg, Osnabrück University, and two DFKI research Old Old areas, Cyber-Physical Systems and the Robotics Innova- re tion Center. AdaMeKoR designs a nursing bed system that, tio in addition to a patient-controlled robotic arm, relieves in the physical strain on the nursing staff by providing ro- botic support when transferring patients. For the robotic arm, AdaMeKoR is comparing different input modes that include conventional manual controls as well as opera- m o c . k i p e e r f , r o t c e v o r c a m ©

28 D F K I N e w s 1 / 2 0 2 3 m o c . e b o d a . k c o t s , 4 0 8 1 r i i m d a w l © The healthcare sector is facing major challenges because of a shortage of medical professionals. The development of an innovative 6G infrastructure and its availability for use in medical systems offers enormous potential to increase eciency and provides innovative approaches for dealing with the shortage of specialists and relieving the burdens on staff. For example, the use of modern communication concepts combined with artificial intelligence methods as well as data transmission that is fast, secure, and in com- pliance with data protection regulations, makes patient monitoring more ecient. High-performance 6G networks with high data rates and low latency make remote surgical interventions and teleconsultation possible. Sixth generation mobile communication: 6G Platform Germany The German platform for 6G technologies in the state of Rhineland Palatinate is located at the University of Technology Kaiserslautern-Landau (RPTU), where it is developing the fundamental technological components for 6G communications and establishing a network of all relevant stakeholders. The project coordinator is Prof. Hans Dieter Schotten, Chair of Wireless Communication and Navigation at RPTU and Head of DFKI‘s Intelligent Networks research department. The 6G Platform primarily works on the harmonization of international regulations and standards while also creat- ing opportunities for active participation by society and industry. There are 18 subject and focus-related industry projects, projects on resilience, and the European Celtic Next Initiative “AI-NET” involved in the development of 6G technologies, with research mainly being driven at four research hubs. Architecture and system design, new communication concepts like Beyond Cellular, Adaptive Radio Access Networks (RAN), resilience, and security by design are just a few of the many ongoing research areas. Technologies like “Internet of Senses,” “Machine Learning at the Edge,” terahertz communication, and quantum communication open up additional opportu- nities to develop novel and secure solutions for the next generation of communication technology. Innovative applications such as brain-computer interfaces (BCI), holographic communication, and networking of IoT and IIoT (Industrial Internet of Things) are all being explored to fully exploit the potential of future communication technologies and create a seamlessly connected future. More information www.6g-plattform.de

30 D F K I N e w s 1 / 2 0 2 3 Under the CLINNOVA Initiative, DFKI works closely with the Luxembourg Institute of Health (LIH). h t l a e H f o e t u t i t s n I g r u b m e x u L © ARTIFICIAL INTELLIGENCE AND HEALTHCARE TOPICS IN EUROPE AI-supported precision medicine, sometimes called per- sonalized medicine, in secure networks – one of the many areas of healthcare that can benefit from artificial intelli- gence. Scientists at DFKI‘s Agents and Simulated Reality (ASR) department are working with physicians, data ana- lysts, and experts in bioinformatics on overarching re- search topics as well as specific issues. Various partners in Germany and neighboring countries, such as Saarland University Hospital and the Luxembourg Institute of Health (LIH), closely cooperate to develop and transfer the lat- ( est technologies in this crucial area of European society. est t tetetecc In particular, an international, digital, and secure system In ppararrtititicucucc laal rr, a noticeably improves healthcare in border regions like the nonotit ceabababablllyly iiiiimpmpmpmpmpm rov four-country corner of Luxemburg, France, Germany, and four-ccccouuntntryyyy ccccorne Switzerland. This is precisely the focus of the CLINNOVA SwSwSwSwitititittzezerland.d.d.dd TTTThihhh s is p initiative: “Unlocking the Potential of Data Science and inininininitiative: “U“U“U“UUnlnnnnnn ocking t Artificial Intelligence in Healthcare.” The goal is to de- ArAArArA tititificficficiaiaiallll InInInnteteteteteeeeelligenceeeee i vvevelop pepersrsononononalalalaalalaalizizizzizededededed aaaaand velop personalized and precise treatment methods with trtrususususssttwtwtwtworrrttht y AAAIAAA . trustworthy AI. OfOfOfOfOfOO coursrse,e,,,, aa ppprerequisisississiiii Of course, a prerequisite is to establish a data network that, among other things, supports the interoperabil- ththththththatatatatat, amamamammonooo gg other thththhthi ity of health data across national borders. Within the ity oofo healtltltltlthhh daddadadatataata aaaaaac CLINNOVA framework, joint research projects are planned between DFKI and the Luxembourg Institute of Health (LIH). A cooperation agreement is currently being prepared for the purpose of establishing a transfer lab- oratory in Luxembourg, where physicians and research- ers will work together on AI topics like precision drugs, personalized health, and trustworthy AI. From DFKI‘s Agents and Simulated Reality department, Prof. Christian Schorr researches AI in medicine. Since the winter semester of 2022/23, he is also teaching “Computer Vision and Artificial Intelligence” at the Zweibrücken cam- pus of the Kaiserslautern University of Applied Sciences. The importance of collaboration between computer scientists and physicians is evidenced by the upcoming symposium “AI in medicine” (https://ki-symposium- homburg.com). The symposium will discuss various as- pects of artificial intelligence in the field of medicine and is planned for November 2023 at the University Hospital of Saarland in Homburg. Organizers are Saarland University, DFKI, and Kaiserslautern University of Applied Sciences, Zweibrücken campus.

32 D F K I N e w s 1 / 2 0 2 3 IN MEDICINE USING THE TOOLS OF PARTICIPATIVE RESEARCH Interactive textiles and smart materials enable new ways to combine textiles and electronics. Traditional craft techniques of textile production integrate microelectronics to create smart materials that act as sensors and actuators. Prof. Dr. Gesche Joost heads the research department Interactive Textiles (InTex) in the development of advanced, human-centric applications in the fields of healthcare, rehabilitation, and medicine, which are then tested in realistic demonstrators and services. Wearables are subject to special ethical, legal, and social challenges because of their use close to the body and their application in the field of health and medicine. Careless use of AI methods risks violation of standards and inappropriate results such as discrimination against patients, misdiagnosis, and misuse of privacy data. “The development of human-centric artificial intelligence begins with empowering the user.” Prof. Dr. Gesche Joost, Head of the Research Department Interactive Textiles Scientists closely involve users in participatory and user-centered ways throughout the entire development phase, constantly reviewing and improving ideas and proto- types in a requirements-oriented, iterative process – not just for functionality but also with ethical and social aspects in mind. TexaS – Construction kit for multi- functional textile-adapted electronic systems In the TexaS research project, a multidisciplinary team of computer scientists, therapists, and patients developed a pair of pants that recognizes movements and couples them with an avatar in the „Spa- ce Time” gaming app. Integrated motion sensors enable the smart garment to act like a game con- troller. Physical exercises such as squats, kicks, and steps by the player trigger movements of the avatar in the game. LEDs and vi- bration engines provide direct feedback. The gaming approach motivates young people, such as those with gait disorders, to engage in physical activity and physiotherapeutic exercise. User autonomy is the project‘s overriding motivation and is ultimately re- sponsible for high user acceptance of this form of therapy. OrthoSuper – Secure data systems and smart sensors for orthopedic care in the future Traditional orthopedic treatments, for example, knee re- habilitation or the fitting of prostheses, require follow-up visits by doctors to check on the progress of the therapy. Such manual measurements accomplish only selective control of the healing process and often delay the adjust- ments required in physical therapy; the convalescence process is interrupted. In the research project OrthoSuper, InTex designs a wearable that supports rehabilitation measures. The aim is to help conventional treatment with a sensor-based, on-body system as well as a multi-camera computer vision system for motion capture. More information www.dfki.de/intex https://youtu.be/MNf5Dmy07mM (Video TexaS) Contact Friederike Fröbel | Daniela Wittmann Research Department Interactive Textiles friederike.froebel | daniela.wittmann@dfki.de +49 30 23895 5958

34 D F K I N e w s 1 / 2 0 2 3 e g g o R . A , k c ü r b a n s O S H © Project AI-TEST-FIELD researchers check the sensors on a carrier that is pulled across the field by a rope drive. Out of the hall, direct from the lab to the farm: An essential success factor in the development of modern high-tech agricultural equipment. 50 MILLION EUROS FOR AI 50 MILLION EUROS FOR AI IN EUROPEAN AGRICULTURE IN EUROPEAN AGRICULTURE GERMAN CENTER FOR AGRITECH IS BEING BUILT IN OSNABRÜCK Osnabrück has long been considered one of the major players in the digitalization of agricul- ture and food production. This position is now further strengthened by agrifoodTEF, an initiative of three institutions – DFKI, Osnabrück University of Applied Sciences, and Agrotech Valley Forum. Currently, agrifoodTEF is establishing national nodes in three European countries with test environments for AI and agricultural robotics. Companies and research institutions from all over Europe will be able to conduct future experiments and validate AI components in prototypes in the Osnabrück region of the state of Lower Saxony and make them ready for market. The team in Osnabrück specializes in agricultural technologies for arable farming and has up to 10 million euros in EU funding. Fog in the lowlands of northern Germany or steep slopes in the vineyards of Trentino pose different challenges to autonomous helpers in agriculture. Such issues must be resolved before the technology can be used. Now, these questions can be addressed in Germany, Italy, and France at “Test and Experimentation Facilities for the Agri-Food Domain.” The European Union (EU) and participating member states are providing leading institutions with up to 50 million euros over the next five years to operate these facilities under the agrifoodTEF initiative. In addition to these three countries with funding of up to 10 million euros each, Austria, Poland, Sweden, and Belgium are par- ticipating with a funding volume of up to 5 million euros each. In Germany, the project is sponsored by the Federal Ministry of Food and Agriculture (BMEL). Since January 1, 2023, each country has been addressing a different topic in agriculture and technology. They are creating the infra- structure needed to develop various functional products: test centers at institutes and in agricultural operations, data sets, software, hardware, machinery, reference im- plementations for certification, and consulting. 3 2 0 2 , r e The first meeting of all European part- ners took place in Trento, Italy, on February 1, 2023. l s s e K o n u r B e n o i z a d n o F ©

36 D F K I N e w s 1 / 2 0 2 3 PAVING THE WAY TO MOON MISSION SherpaTT and LUVMI-X at the edge of the skylight. EUROPEAN ROBOT TEAM AUTONOMOUSLY EXPLORES LAVA CAVE ON LANZAROTE A permanent station on the moon that can serve as a starting point for further exploration of our solar system is a crucial goal of international space exploration. Opti- mal conditions for such a station are likely to be found in hard-to-reach environments like craters or lava caves, which not only offer protection from radiation, meteor- ites, and temperature fluctuations. Such places are also thought to contain important resources like frozen water beneath the lunar surface. Before humans venture into the hidden depths of Earth‘s companion, promising locations will be investigated by autonomous robots. How does the robot get into the cave? Since March 2021, DFKI‘s Robotics Innovation Center has been coordinating with nine European partners in the CoRob-X (Cooperative Robots for Extreme Environments) project, an ambitious exploration concept based on teamwork among different robotic systems. The project was one of the final projects of the Strategic Research Cluster (SRC) Space Robotic Technologies, which comprises several research projects funded under the European Commission‘s Horizon 2020 program and managed by the PERASPERA Program Sup- port Activity. The aim of CoRob-X is to test and validate the technologies developed in SRC for use in future space missions that will enable robots to work as a team in the lava cave scenario. The scenario involves three autono- mous rover vehicles – SherpaTT and Coyote III from DFKI, and LUVMI-X from the Belgian company Space Applica- tions Services – that will jointly explore the entrance to a lava cave, the so-called skylight, and penetrate into the interior of the cave by lowering a robot down by ropes. To prove the feasibility of this complex mission, 25 project team members traveled to Lanzarote from January 21 to February 11, 2023. With almost two years of intensive re- search and development work behind them and several weeks of integration work last September at DFKI Bremen, their job was to test the functionality of the technologies in comprehensive field tests. The volcanic Canary Island, with its extensive lava tunnels, offered ideal test condi- tions. The project partners set up their tents at a previously selected test site, directly next to a skylight - an approxi- mately five-meter-wide hole in the volcanic soil that pro- vides access to a lava cave about four meters below. The team‘s goal was to put components and systems to the DFKI researchers make technical adjustments to Coyote III after a rappelling test.

38 D F K I N e w s 1 / 2 0 2 3 I H H r e f o h n u a r F © DIGITAL TECHNOLOGY DIGITAL TECHNOLOGY FORUM OPENS NEW BERLIN FORUM OPENS NEW BERLIN SHOWROOM SHOWROOM Close-out event of ForeSight project March 15, 2023, at the Digital Technology Forum. The Forum for Digital Technologies (FDT) is starting 2023 with a welcome change of venue. A new showroom opened at Salzufer 15/16 in Berlin-Charlottenburg in mid-February. Anyone who has previously visited the Forum knows that its showroom exhibits a unique combination of digital technology demonstrators and an attractive, modern event center. The Digital Technologies Forum is operat- ed by the DFKI research department Intelligent Analytics for Massive Data on behalf of the Federal Ministry for Eco- nomic Affairs and Climate Action (BMWK). Project part- ner Fraunhofer HHI manages the showroom floor space. Digital innovation “Made in Germany” The new showroom also serves primarily as an exhibition and conference center for Federal Ministry of Econom- ics and Technology-funded projects as well as for other research projects. Expert workshops, project meetings, and networking events take place almost daily, with other exchange formats exploiting multipliers from the digital community. The aim is to accelerate the transfer of results from research to the German economy and to enable the networking of various players from science, industry, and ForeSight results exhibition in Showroom. politics. The new modular room concept provides space for more than 300 people in a total area of 700 square meters. International guests and delegations are frequent visitors to the Forum. Guided tours allow visitors to experience applications in artificial intelligence, blockchain, 3D print- ing, robotics, e-health, secure digital identities, and smart living, as exhibited by the demonstrators. “ToHyVe” Project Testbed The FDT showroom is also the testbed and living lab for the BMWK-funded research project ToHyVe (Toolbox for Hybrid eVents). As such, it brings together project part- ners from the fields of immersive audio and video tech- nologies, artificial intelligence, streaming, data security, virtual worlds, event platforms, exhibition planning, and usability for the purpose of developing more immersive solutions for a range of use cases. These are tested at fo- rum events and can then be flexibly adapted and further developed. As one of the ToHyVe project partners, DFKI develops a virtual conversational concierge. More information www.forum-digitale-technologien.de Contact Nadja Mesheva Research Department Intelligent Analytics for Massive Data nadja.mesheva@dfki.de +49 30 238 95 1848 Showroom visit address c/o Fraunhofer HHI Salzufer 15/16, Lanolinfabrik/Kanalhaus EG 10587 Berlin

I D F K N e w s 1 / 2 0 2 3 Best Paper Award to DFKI Lübeck Author Team f MICCAI 2022 MAD Workshop – e i r b The international conference on Medical Image Computing and Computer-Assisted Surgery 2022 (MICCAI) was held in Singapore last September. At the Medical Applications with Disentangle- ments (MAD) workshop, Dr. Hristina Uzunova, Dr. Jan Ehrhardt, and Prof.Heinz Handels, three researchers at DFKI‘s Artificial Intelligence in Medical Imaging (AIMedI) department, won the Best Paper Award for their paper titled „Low-rank and Sparse Metamorphic Autoencoders for Unsupervised Pathology Disentanglement.“ The paper focuses on the use of AI-based segmentation of medi- cal image data to differentiate between the shape and appear- ance of healthy versus pathological findings. The new procedural method could be used, for example, as the basis for complex image synthesis. Rapid searches in unstructured data – DFKI researchers win Best Long Paper Award at BTW’23 Researchers/authors Benjamin Hättasch, Jan- Micha Bodensohn, Liane Vogel, Matthias Urban, and Carsten Binnig from the DFKI-Darmstadt Lab won the Best Long Paper award at the 20th BTW Conferen- ce held this year in Dresden. Their submission titled „WannaDB: Ad-hoc SQL Queries over Text Collections“ describes an interactive system that enables rapid se- arching in large, unstructured data sets and is able to filter high-quality structured data from a wide range of text collections. BTW is one of the most important conferences for information systems and databases in the German- speaking region. 40 N E W S I N B R I E F i s w e N New DFKI Research Area: Neuro-Mechanistic Modeling Dr. Verena Wolf, Professor of Computer Science at Saarland University, has been selected to head the new research area, “Neuro-mechanistic Mod- eling,” at DFKI Saarbrücken. The new department focuses on the question of how to combine estab- lished knowledge with data-driven machine learn- ing methods. Such real-world knowledge is often described as so-called mechanistic models, so the new department aims to develop a complemen- tary hybrid approach that combines data-driven, neural-based models with mechanistic models. Prof. Verena Wolf. More information: www.dfki.de/nmm Muhammad Nabeel Asim completes PhD work with distinction DFKI Kaiserslautern researcher graduates “summa cum laude.” Nabeel Asim‘s dissertation, “An E- cient Automated Machine Learning Framework for Genomics and Proteomics Sequence Analysis,” ad- dresses the development of an ecient, automat- ed machine learning system for sequence analysis of genomic and proteomic data. He proposes a framework that enables the auto- mated design of predictive pipelines for genom- ic and proteomic sequence analysis tasks. Using this framework, practical web applications can be developed to perform the tasks of sequence analy- sis for various DNA, RNA, and proteins. More information: https://sds_genetic_analysis.opendfki.de

42 D F K I P R O F I L D F K I N e w s 1 / 2 0 2 3 The German Research Center for Artificial Intelligence (DFKI) was founded in 1988 as a non-profit public-private partnership. It has research facilities in Kaiserslautern, Established 1988, non-profit organization (public-private partnership) Saarbrücken, Bremen and Niedersachsen, laboratories in Berlin and Darmstadt and branch oces in Lübeck and Trier. In the field of inno- vative commercial software technology using Artificial Intelligence, DFKI is the leading research center in Germany. Executive Board Prof. Dr. Antonio Krüger Helmut Ditzer Based on application oriented basic research, DFKI develops product functions, prototypes and patentable solutions in the field of infor- mation and communication technology. Research and development projects are conducted in 28 research departments, nine compe- tence centers and eight living labs. Funding is received from govern- ment agencies like the European Union, the Federal Ministry of Educa- tion and Research (BMBF), the Federal Ministry for Economic Affairs and Climate Action (BMWK), the German Federal States and the German Research Foundation (DFG), as well as from cooperation with indus- trial partners. Twice a year, a committee of internationally renowned experts (Scientific Advisory Board) audits the progress and results of state-funded projects. Apart from the state governments of Rhineland-Palatinate, Saarland and Bremen, numerous renowned German and international high-tech companies from a wide range of industrial sectors are represented on the DFKI supervisory board. The DFKI model of a non-profit public-pri- vate partnership (ppp) is nationally and internationally considered a blueprint for corporate structure in the field of top-level research. DFKI is actively involved in numerous organizations representing and continuously advancing Germany as an excellent location for cut- ting-edge research and technology. Far beyond the country’s borders DFKI enjoys an excellent reputation for its academic training of young scientists. At present, approx. 925 highly qualified researchers, admin- istrators and 630 graduate students from more than 75 countries are contributing to more than 565 DFKI research projects. DFKI serves as a stepping stone to leading positions in industry and successful ca- reers as founders of spin-off companies. Over the years, more than 170 staff members have been appointed professors at universities in Germany and abroad. More information www.dfki.de Contact Reinhard Karger, M.A. Corporate Spokesperson German Research Center for Artificial Intelligence, DFKI Saarland Informatics Campus D3 2, 66123 Saarbrücken, Germany reinhard.karger@dfki.de +49 681 85775 5253 Supervisory Board Dr. Ferri Abolhassan, Deutsche Telekom AG (Chair) Elena Yorgova-Ramanauskas, State Secretary of the Ministry of Economic Affairs, Technology, Digitalization and Energy of the state Saarland (Vice Chair) 22 Shareholders Accenture, Airbus Group, Bilfinger SE, BMW AG, Cerence GmbH, CLAAS KGaA mbH, Daimler AG, Deutsche Börse AG, Deutsche Messe AG, Deutsche Telekom AG, Empolis Information Management GmbH, Fraunhofer Gesellschaft e.V., Google Inc., HARTING AG & Co. KG, Intel Corporation, John Deere GmbH & Co. KG, KIBG GmbH, Microsoft Deutsch- land GmbH, Münchener Rückver- sicherungs-Gesellschaft Aktien- gesellschaft in München, NVIDIA GmbH, RICOH Company Ltd., Robert Bosch GmbH, ROSEN Swiss AG, SAP SE, Sartorius Ventures GmbH, Schwarz-Gruppe,Soft- ware AG, Rheinland-Pfälzische Technische Universität Kaiserslau- tern-Landau, Universität Bremen, Universität des Saarlandes, Volkswagen AG, VSE AG, ZF Friedrichshafen AG International Scientific Advisory Board Bi-annual evaluation of publically funded projects: Prof. Dr. Andreas Butz, Ludwig-Maximilians-Universität, München (Chairman) Key Figures 2022 ca. 82,6 € million

Hybrid AI • Industry 4.0, Digital Twins, Semantic Product Memories • Smart Data – Intelligent Analytics for Massive Data • Wearable Computing and Interactive Textiles • Deep Learning and Machine Learning • Knowledge Management and Document Analysis • Softbots, Digital Assistants, and Chatbots • Educational Technologies • Verification and Evaluation of Safety-critical Applications • Cognitive Social Simulation • Information Extraction and Intelligent Web Retrieval • Multiagent Systems • Experience-based Learning Systems • Visual Computing and Augmented Vision • Mobile and Collaborative Robotic Systems • Multimodal User Interfaces and Autonomous Systems • Shopping Assistance and Intelligent Logistics • Safe and Secure Cognitive Systems and Intelligent Security Solutions • Ambient Intelligence and Assisted Living • Driver Assistance Systems and Autonomous Driving • Cyber-physical Systems • Multilingual Technologies and Language Under- standing • Business Process Management and Smart Services • Affective Computing • AI in Medicine and Healthcare • Systems AI Bremen Site Robert-Hooke-Straße 1 D-28359 Bremen +49 421 17845 0 Niedersachsen Site Hamburger Straße 24 D-49084 Osnabrück +49 541 386050 0 Saarbrücken Site Saarland Informatics Campus D3 2 D-66123 Saarbrücken +49 681 85775 0 Laboratory Darmstadt TU Darmstadt Hochschulstr. 10 D-64289 Darmstadt Kaiserslautern Site Trippstadter Straße 122 D-67663 Kaiserslautern +49 631 20575 0 Laboratory Berlin Alt-Moabit 91c D-10559 Berlin +49 30 23895 0 www.dfki.de info@dfki.de