Projects

Deep learning has emerged as a key technology in biomedical image analysis, but is difficult to handle for non-experts due to the high demands on computing power and data management. This project will develop a platform to facilitate large-scale statistical analysis of multimodal biomedical imaging and patient data using state-of-the-art deep learning methods. The proposed infrastructure will provide access to state-of-the-art algorithms and define a standardized data infrastructure that can be easily deployed in heterogeneous environments. Our prototype will provide an effective mechanism for sharing pre-trained algorithms and advanced analytical tools. The platform is aimed at the biomedical research community and will provide scientists with novel, powerful and validated tools to address challenges such as image-based disease phenotyping and predictive modeling. State-of-the-art analysis pipelines will be implemented and packaged into user-friendly toolboxes that can be directly used in clinical workflows and enable the extraction of imaging biomarkers and quantitative measurements. Our approach is based on three basic principles: Data linkage (across systems), data stewardship (patient privacy and legal/ethical compliance) and data interoperability (use of public APls and open standards). To achieve this, we will build on an existing model: Data will be kept in a secure environment, using AI algorithms to train with sensitive patient data within our clinic's firewall. Two approaches will be supported: a secure learning orchestration server that will provide learning coordination for secure data enclaves at our partner hospital, the University Hospital Erlangen (UKER), and secure sandboxes that will enable model development in a secure environment hosted by the university at FAU. As in a federated learning paradigm, most of the models will move through our infrastructure, not the data. We will set up (and support) the infrastructure in these environments with support from the Department of Artificial Intelligence in Biomedical Engineering (AIBE) at FAU, the Radiology Department at UKER and the Regional Computing Center Erlangen (RRZE) to provide the required capabilities. Our proposed solution is highly interoperable and scalable for other clinics and enables integration e.g. with the Medical Informatics Initiative. This approach will provide secure and compliant access to the clinics' PACS and electronic patient records, and enable reproducible research.

Virtual twins have the potential to be used as prognostic tools in precision medicine to support individualized disease management. However, training these models requires large volumes of data from various sources, which is challenging due to privacy regulations like the General Data Protection Regulation (GDPR). Recently, privacy-preserving computational methods, such as federated learning, have emerged, offering a way to utilize extensive data effectively while protecting sensitive patient information.

In dAIbetes, our main medical objective is to provide personalized predictions of treatment outcomes for type 2 diabetes, a condition affecting 1 in 10 adults globally and leading to annual costs of approximately 893 billion EUR. Although healthcare professionals are improving at addressing diabetes risk factors like diet and exercise, there are currently no guidelines for predicting treatment outcomes tailored to individual patients.

dAIbetes brings together advaned expertise in federated learning, artificial intelligence, cybersecurity, diabetes data standardization, clinical validation, as well as in legal and ethical evaluation of applying advanced federated machine learning to personalized medicine. 13 Partners from 13 European countries and the US will jointly implement the project which is structured into 9 Work Packages (WP1-WP9). At FAU, we are working on WP3, i.e., the development of virtual twin apps for training of virtual twin models that will use data from type 2 diabetes patients.

In FLabNet, we will harness the potential of algorithmic network biology and distributed machine learning to address two exemplary unmet needs in paediatric oncology: prediction ofchemotherapy side effects like neutropenic fever and early-stage detection of rare malignantdiseases such as myeloproliferative neoplasms. Based on >54 million laboratory test resultsfrom >500,000 patients from the Core Dataset of the German Medical Informatics Initiative (MII),we will create personalised networks, where nodes represent individual laboratory measurementsand edges encode patient-specific relationships. We hypothesise the emerging personal graph representations to capture the unique spectra and dependencies of the individual patients’ health anddisease characteristics. The networks will be used as signatures for label-efficient graph-based pre-dictors such as graph kernels; and we will provide privacy-preserving federated implementationsof our predictors that are fully interoperable with MII standards. To achieve its objectives, ourconsortium combines expertise in algorithmic systems biology (FAU), paediatric oncology (UKER),quantitative analysis of laboratory data (UKER), federated learning for biomedicine (Bitspark GmbH& FAU), and professional software development (Bitspark GmbH). These synergistic skill sets willenable us to combine laboratory diagnostics, computational systems medicine, and privacy-preserving machine learning, advancing the state of the art in quantitative analysis of laboratory data for precision medicine in paediatric oncology and beyond. 

Entzündung beginnt vor allem in Geweben, und ihre Ausprägung und ihr Ausmaß werden weitgehend von diesen kontrolliert. Infiltrierende Immunzellen sind hochgradig plastisch und integrieren sowohl molekulare als auch biophysikalische Informationen aus ihrer Umgebung, die allesamt ihre Effektor-Funktionen sowie ihr weiteres Schicksal bestimmen. Mit dem Fortschritt hochdimensionaler OMICs-Ansätze haben wir begonnen, die zelluläre Heterogenität komplexer entzündlicher Infiltrate besser zu verstehen. Während jüngste analytische Ansätze zelluläre Interaktionen auf der Grundlage von Transkriptionsprofilen vorhersagen können, bleiben ihre tatsächlichen funktionellen Interaktionen und die Einwirkung ihrer mikroanatomischen Umgebung jedoch weiterhin spekulativ. Um die Beziehung zwischen Gewebestruktur, Architektur der Mikroumgebung und Zellfunktionen verstehen zu können, müssen multidimensionale Ex vivo-Zellanalysen durch quantitative In-situ-Bildgebung ergänzt werden. Neueste 2D-Hyperplex-Mikroskopieverfahren erlauben mittlerweile mehrere Parameter mit räumlichem Kontext zu erfassen. Mit diesen Techniken lässt sich jedoch die komplexe 3D-Architektur von strukturell vielschichtigen und kompartimentierten Geweben wie dem Darm oder dem Gehirn nicht adäquat abbilden. Gleichzeitig erfordert die Fülle von Daten, die durch Bildgebungs- und OMICs-Techniken erzeugt werden, nachfolgende integrative Analyseschritte, die bei einer manuellen Auswertung in der Regel nicht zu bewältigen sind. Um diese Lücken zu schließen, werden wir neue Mikroskopieanwendungen und quantitative, durch maschinelles Lernen unterstützte Analysemethoden für Gewebe entlang der Darm-Hirn-Achse mit hoher Granularität sowohl in 2D als auch in 3D entwickeln. Dieses explorative Vorhaben dient der Etablierung innovativer Methoden und wird dazu beitragen, räumliche Daten mit zellulären Profilen zu integrieren, um Organisationsschemata zu entschlüsseln, die Gesundheit und Krankheit zugrunde liegen.

The goal of this project is the development of an application for supporting the physical rehabilitation therapy of prostatectomy and incontinence patients in planning and execution. An AI-driven algorithm for automatic planning will be developed and extended by a machine learning approach for live exercise execution feedback. The developed application will be clinically evaluated regarding effectiveness and therapy benefit. 

Rapid advances in single-cell RNA sequencing (scRNA-seq) technology are leading to ever-increasing dimensions of the generated molecular data, which complicates data analyses. In NetMap, new scalable and robust dimensionality reduction approaches for scRNA-seq data will be developed. To this end, dimensionality reduction will be integrated into a central task of the systems medicine analysis of scRNA-seq data: inference of gene regulatory networks (GRNs) and driver transcription factors based on cell expression profiles. Each resulting dimension will correspond to a driver GRN, and the coordinate of a cell in this low-dimensional representation will quantify the extent to which the particular driver GRN explains the cell's gene expression profile. These new methods will be implemented as a user-friendly software platform for exploratory expert-in-the-loop analysis and in silico prediction of drug repurposing candidates.

As a case study, we will investigate CD4 helper T cell exhaustion, a potential limiting factor in immunotherapy. NetMap's strategy consists of (1) analyzing phenotypic heterogeneity of depleted CD4 T cells, (2) identifying transcriptional mechanisms that control this heterogeneity, (3) amplifying/eliminating specific subsets and testing their functional impact. This will allow the development of an atlas of the gene regulatory landscape of depleted CD4 T cells, while the in vivo testing of key regulatory transcription factors will help demonstrate the power of the developed methods and allow evaluation and improvement of predictions. 

The Cooperative Health Research in South Tyrol (CHRIS) study offers a comprehensive overview of the health state of >13,000 adults in the middle and upper Val Venosta. It is the largest population-based molecular study in Italy with a longitudinal lookout to investigate the genetic and molecular basis of age-related common chronic conditions and their interaction with lifestyle and environment in the general population. In CHRIS, the combination of molecular profiling data, such as genomics and metabolomics, together with important baseline clinical and lifestyle data offers vast opportunities for understanding physiological changes that could lead to clinical complications or indicate the prevalence or early onset of diseases together with their molecular underpinnings. 

Where disease-focused studies often have a clear hypothesis that dictates the necessary statistical analyses, population-based cohorts such as CHRIS are more versatile and allow both testing existing hypotheses as well as generating new hypotheses that arise from statistically significant associations of the available data. Ideally, this type of explorative analysis is open to biomedical researchers that do not necessarily have experience with data analysis or machine learning. Network-based approaches are ideally suited for studying heterogeneous biomedical data, giving rise to the field of network medicine. However, network medicine techniques have so far mainly been used in the context of studies focusing on individual diseases. Network-based platforms for the explorative analysis of population-based cohort data do not exist.

In DyHealthNet, we will close this gap and develop a network-based data analysis platform, which will allow to integrate heterogeneous data and support explorative data analytics over dynamically generated subsets of the CHRIS study data. To fully leverage the potential of the available multi-level data, the DyHealthNet platform combines (1) data integration using standardized medical information models (HL7 FHIR), (2) innovative index structures for scalable dynamic analysis, (3) machine learning, and (4) visual analytics. DyHealthNet will render the CHRIS population cohort data accessible for state-of-the-art privacy-preserving, network-based data analysis. DyHealthNet will hence enable mining of context-specific pathomechanisms for precision medicine, and will serve as a blueprint for dynamic explorative analysis of multi-level cohort data worldwide. To achieve these objectives, the DyHeathNet consortium combines expertise in population-based cohort studies (Fuchsberger) and in the development of complex algorithms for the analysis of molecular networks (Blumenthal), applied biomedical AI and software systems (List), and customized index structures for scalable data management (Gamper).

The majority of diagnostic medicalimaging pipelines follow the same principles: raw measurement data is acquiredby scanner hardware, processed by image reconstruction algorithms, and thenevaluated for pathology by human radiology experts. Under this paradigm, every stephas traditionally been optimized to generate images that are visually pleasingand easy to interpret for human experts. However, raw sensor information thatcould maximize patient-specific diagnostic information may get lost in thisprocess. This problem is amplified by recent developments in machine
learning for medical imaging. Machine learning has been used successfully inall steps of the diagnostic imaging pipeline: from the design of dataacquisition to image reconstruction, to computer-aided diagnosis. So far, thesedevelopments have been disjointed from each other. In this project, we willfuse machine learning for image reconstruction and for image-based diseaselocalization, thus providing an end-to-end learnable image reconstruction andjoint pathology detection approach that operates directly on raw measurementdata. Our hypothesis is that this combination can maximize diagnostic accuracywhile providing optimal images for both human experts and diagnostic machinelearning models.

The goal of this project is the development of an application for supporting physical rehabilitation therapy in planning and execution. An AI-driven algorithm for automatic planning will be developed and extended by a machine learning approach for live exercise execution feedback. The developed application will be clinically evaluated regarding effectiveness and therapy benefit.

X02 nutzt die in EBM erzeugten Bilddaten und mechanischen Messungen, um Deep Learning-Methoden zu entwickeln, die Wissen über Spezies hinweg transferieren. In silico und in vitro Analysen werden deutlich spezifischere Daten liefern als in vivo Experimente, insbesondere für menschliches Gewebe. Um hier Erkenntnisse aus datenreichen Experimenten zu nutzen, werden wir Transfer Learning-Algorithmen für heterogene Daten entwickeln. So kann maschinelles Lernen auch unter stark datenlimitierten Bedingungen nutzbar gemacht werden. Ziel ist es, ein holistisches Verständnis von Bilddaten und mechanischen Messungen über Artgrenzen hinweg zu ermöglichen.

As one of the most important aspects of diagnosis, treatment planning, treatment delivery, and follow-up, medical imaging provides an unmatched ability to identify disease with high accuracy. As a result of its success, referrals for imaging examinations have increased significantly. However, medical imaging depends on interpretation by highly specialised clinical experts and is thus rarely available at the front-line-of-care, for patient triage, or for frequent follow-ups. Very often, excluding certain conditions or confirming physiological normality would be essential at many stages of the patient journey, to streamline referrals and relieve pressure on human experts who have limited capacity. Hence, there is a strong need for increased imaging with automated diagnostic support for clinicians, healthcare professionals, and caregivers.

Machine learning is expected to be an algorithmic panacea for diagnostic automation. However, despite significant advances such as Deep Learning with notable impact on real-world applications, robust confirmation of normality is still an unsolved problem, which cannot be addressed with established approaches.

Like clinical experts, machines should also be able to verify the absence of pathology by contrasting new images with their knowledge about healthy anatomy and expected physiological variability. Thus, the aim of this proposal is to develop normative representation learning as a new machine learning paradigm for medical imaging, providing patient-specific computational tools for robust confirmation of normality, image quality control, health screening, and prevention of disease before onset. We will do this by developing novel Deep Learning approaches that can learn without manual labels from healthy patient data only, applicable to cross-sectional, sequential, and multi-modal data. Resulting models will be able to extract clinically useful and actionable information as early and frequent as possible during patient journeys.

Recent legislative development, such as the European Health Data Space, expand access to anonymizied health data for various entities. While these advances offer opportunities for medical research and innovation, they also increase the risk of compromising individuals' privacy.

This project addresses the critical tension between the growing utility of health data and the need to protect individual privacy through organizational, infrastructural, and technical approaches. A key component of the technical solutions is privacy-enhancing technologies (PETs), such as secure multi-party computation and (local) differential privacy, which safeguard individuals' privacy while enabling the statistical analysis of aggregate data.

Dieses Projekt ist Teil der Forschungsgruppe (FOR) "Schnelle Kartierung von quantitativen MR bio-Signaturen bei ultrahohen Magnetfeldstärken". Es konzentriert sich auf die Erweiterung, Beschleunigung und Verbesserung der Diffusions- und quantitativen Suszeptibilitäts-Magnetresonanztomographie. Das Arbeitsprogramm ist in zwei Teile gegliedert. Im ersten Teil wird ein beschleunigtes Protokoll für die klinischen Projekte der FOR vorbereitet. Im zweiten Teil sollen eine weitere Beschleunigung sowie Qualitätsverbesserungen erreicht werden. Konkret werden wir eine lokal niedrigrangig regularisierte echoplanare Bildgebungssequenz für die diffusionsgewichtete Bildgebung implementieren. Sie nutzt Datenredundanzen bei Akquisitionen mit mehreren Diffusionskodierungen, um das Signal-Rausch-Verhältnis effektiv zu erhöhen und damit den Akquisitionsprozess zu beschleunigen. Die Sequenz wird im Wesentlichen beliebige Diffusionskodierungsmöglichkeiten ermöglichen (z.B. b-Tensor-Kodierung). In einem zweiten Schritt werden wir eine verschachtelte Mehrschuss-Version dieser Sequenz entwickeln, um Bildverzerrungen zu reduzieren, die bei der 7-Tesla echoplanaren Bildgebung störend sind. Für die quantitative Suszeptibilitätskartierung (QSM) werden wir eine Sequenz mit einer Stack-of-Stars-Aufnahmetrajektorie implementieren. Da die Magnitudenbilder von Gradientenechosequenzen, die zu unterschiedlichen Echozeiten akquiriert werden, Datenredundanzen aufweisen, die mit denen von diffusionskodierten Bildern vergleichbar sind, werden wir bei der Bildrekonstruktion ebenfalls eine lokale Regularisierung niedrigen Ranges verwenden. Die radialen Trajektorien dieser Sequenz sollten für eine unterabgetastete und damit beschleunigte Bildrekonstruktion gut geeignet sein. In einem zweiten Schritt werden wir die Fähigkeiten unserer Sequenz durch eine quasi-kontinuierliche Echozeit-Abtastung erweitern, bei dem jede Speiche ihre eigene optimierte Echozeit hat. Dies wird eine verbesserte Qualität der QSM ermöglichen, wenn Fett im Bild vorhanden ist, wie es häufig bei Muskeluntersuchungen und in der Brustbildgebung der Fall ist. Bezüglich der QSM-Rekonstruktion werden wir Verfahren des tiefen Lernens entwickeln, um eine qualitativ hochwertige Rekonstruktion mit einer geringeren Menge an Bilddaten als bei herkömmlichen Rekonstruktionsansätzen zu ermöglichen. Wir werden bestehende neuronale Netzwerke von niedrigeren Feldstärken auf 7 T anpassen und deren Fähigkeiten so erweitern, dass wir auch atemzyklusabhängige Feldkarten. Dieses Projekt wird parallele Sendemethoden (pTx) vom pTx-Projekt der FOR erhalten. Wir werden die entwickelten Sequenzen nach dem ersten Jahr an die klinischen Projekte der FOR liefern. Darüber hinaus werden wir wesentliche Auswerte- und Bildrekonstruktionsmethoden an die anderen Projekte der FOR transferieren.und quasi-kontinuierliche Echozeiten in die Rekonstruktion integrieren können.

In Germany alone, the number of patients with chronic wound healing disorders is estimated at around 2.7 million. According to projections, the treatment of chronic wounds accounts for € 23 - 36 billion per year. Of the treatment costs for chronic wounds, 4.6 to 7.2 billion € alone are accounted for by the associated cost-intensive dressing materials. The aim of the SWODDYS project is to research the fundamentals for a new type of intelligent wound dressing for the treatment of acute and chronic wounds, which can monitor the energy-metabolic tissue and wound healing status individually for each patient and online by integrating fluorescent dye-based oxygen, pH and CO2 sensors. 

The EU project TEF-Health aims to test and validate innovative artificial intelligence (AI) and robotics solutions for the healthcare sector and accelerate their path to market. It is led by Prof. Petra Ritter, who heads the Brain Simulation Section at the Berlin Institute of Health at Charité (BIH) and at the Department of Neurology and Experimental Neurology of Charité – Universitätsmedizin Berlin. The MaD Lab of the FAU is one of the 51 participating project partners from nine European countries.

Diabetes is an overwhelming disease, directly influencing more than 422 million people worldwide who are living with this disease. Type 1 diabetes is the most severe form of the disease. The management of type 1 diabetes is especially difficult for young children and adolescents. Additionally, the most feared complication of type 1 diabetes – hypoglycemia – might occur after several hours, for example, during the night. 

The DIAmond project will address the personalized and better management of type 1 diabetes using data science and machine learning to gain insights into the problem of hypoglycemia. Data from the DIAcamp study is used to advance personalized treatment recommendations. In the DIAcamp study, children participated for one week. They were equipped with a continuous glucose sensor and a wearable device for monitoring heart rate, accelerometry, and further physiological parameters during their participation. Physicians and carers from the DIAcamp study documented insulin doses, carbohydrate intake, and time and type of activity. Within the DIAmond project, novel machine learning algorithms will determine the probability of hypoglycemia. Exploratory analysis of the physiological time series will result in the most predictive features, building the base for personalized treatment recommendations. 

This project is a joint project with the Department of Computer Science, ETH Zurich, Switzerland.

Biomechanical modeling and simulation are performed to analyze and understand human motion and performance. One objective is to reconstruct human motion from measurement data e.g. to assess the individual performance of athletes and customers. Another objective is to synthesize realistic human motion to study human-production interaction. The reconstruction (a) and synthesis of human motion (b) will be addressed in this  research position. New algorithms using biomechanical simulation of musculoskeletal models will be developed to enable innovative applications and services for Adidas. Moreover, predictive biomechanical simulation will be combined with wearable sensor technology to build a product recommendation application.

University education in psychology, medical technology and computer science currently focuses on teaching basic methods and knowledge with little involvement of other disciplines. Increasing digitalization and the ever more rapid spread of digital technologies, such as wearable sensors, smartphone apps, and artificial intelligence, also in the health sector, offer a wide range of opportunities to address psychological issues from new and interdisciplinary perspectives. However, this requires close cooperation between the disciplines of psychology and technical disciplines such as medical technology and computer science to enable the necessary knowledge transfer. Especially in these disciplines, there is a considerable need for innovative and interdisciplinary teaching concepts and research projects that teach the adequate use of digital technologies and explore the application of these technologies to relevant issues in order to enable better care in the treatment of people with mental disorders.

The goal of this project is to develop experimental approaches and simulation methods for biomechanical assessment of big wave surfing. This goal will be addressed in collaboration with Sebastian Steudtner and Siemens Healthineers. The methods include, but are not limited to, biomechanical movement analysis, musculoskeletal simulation, and sensor fusion.

The focus of the research activities will be centered on:

• Development of a measurement approach for biomechanical assessment of big wave surfing
• Development of efficient and accurate data processing combining inputs from several sensor systems
• Design of a biomechanical simulation model that reflects the situation during surfing
• Analysis of biomechanical measurements and simulation outcomes to provice advice for big wave surfer to improve performance. 

The main goal of this project is to improve the detector manufacturing for computer tomography (CT). Therefore, data is gathered during the production of a CT-detector. This data is analysed and used to develop and train a machine learning system which should find the best composition of a CT-detector. In the future, the system will be integrated into the process of CT-detector manufacturing which, in result, should further improve the image quality and the production process of CT-devices. Especially, the warehouse utilization and the first-pass-yield should be enhaced. The project is realized in cooperation with Siemens Healthineers Frochheim.

Ziel dieses Teilprojekts ist es, datengetriebene Vorhersagen über Absatz und Haltbarkeit der Produkte unserer Industriepartner zu treffen, um Nahrungsmittelverluste zu reduzieren, während der Umsatz erhöht wird. Eine besondere Herausforderung von frischen Lebensmitteln stellt ihre kurze Haltbarkeit da, welche eine besonders präzise Absatzprognose notwendig macht. Wir adressieren dieses Problem mit einem auf Bayes‘schen maschinellen Lernen basierenden Ansatz. Bayes‘sche Methoden liefern zusätzlich zu einer punktuellen Vorhersage (z. B. über den Absatz) auch ein Maß über die Unsicherheit dieser Vorhersage.

The project aims to identify areas where advanced data analysis and processing methods can be applied to aspects of computer tomography (CT) technology. Furthermore included is the implementation and validation of said methods.

In this project, we analyze machine and customer data sent by thousands of high-end medical devices every day. 

Since potentially relevant Information is often presented in different modalities, the optimal application of fusion techniques is a key factor when extracting insights. 

Over the last years, large amounts of molecular profiling data (also called “omics data”) have become available. This has raised hopes to identify so-called disease modules, i.e., sets of functionally related molecules constituting candidate disease mechanisms. However, omics data tend to be overdetermined and noisy; and modules identified via purely statistical means are hence often unstable and functionally uninformative. Hence, network-based disease module mining methods (DMMMs) project omics data onto biological networks such as protein-protein interaction (PPI) networks, gene regulatory networks (GRNs), or microbial interaction networks (MINs). Subsequently, network algorithms are used to identify disease modules consisting of small subnetworks. This dramatically decreases the size of the search space and prioritizes disease modules consisting of functionally related molecules, positively affecting both stability and functional relevance of the discovered modules.

However, to the best of our knowledge, all existing DMMMs are subject to at least one of the following two limitations: Firstly, existing DMMMs are typically supervised, in the sense that they try to find subnetworks explaining differences in the omics data between predefined case and control patients or pre-defined disease subtypes. This is potentially problematic, because it implies that existing DMMMs are biased by our current disease ontologies, which are mostly symptom- or organ-based and therefore often too coarse-grained. For instance, around 95 % of all patients with hypertension are diagnosed with so-called “essential hypertension” (code BA00.Z in the ICD-11 disease ontology), meaning that the cause of the hypertension is unknown. In fact, there are probably several disjoint molecular mechanisms causing “essential hypertension”, and the same holds true for many other complex diseases such as Alzheimer’s disease, multiple sclerosis, and Crohn’s disease. Supervised DMMMs which take existing disease definitions for granted hence risk overlooking the molecular mechanisms causing mechanistically distinct subtypes.

Secondly, most existing DMMMs are designed for static omics data and do not support longitudinal data where the patients’ molecular profiles are observed over time. Existing analysis frameworks for longitudinal omics data largely use purely statistical means. Consequently, network medicine approaches for time series data are needed.

To the best of our knowledge, there are only three DMMMs which, in part, overcome these limitations: BiCoN and GrandForest allow unsupervised disease module mining but do not support longitudinal omics data. TiCoNE supports longitudinal data but requires predefined case vs. control or subtype annotations as input. There is hence an unmet need for unsupervised DMMMs for longitudinal omics data. Developing such methods is the main objective of the proposed project.

Parkinson's disease after Alzheimer's is the second most common neurodegenerative disease which mainly affects the patient's mobility and produces gait insecurity and impairment. As patients experience various, asymmetrical and heterogeneous gait characteristics, personalized medication should be at the center of attention in controlling motor complications in Parkinson's patients. potentially, inertial measurement units (IMUs) can be utilized for long-term observation of the disease progress and estimating gait parameters. This project is dedicated to detecting and possibly predicting the motor symptoms of Parkinson's disease such as Bradykinesia, Dyskinesia, and the freeze of gait. This also includes the improvement of the existing gait analysis algorithms to fit the parkinsonian gait more accurately, which is the basis of symptom detection.

The digitization of the sports sector leads to the individualization of products and services for everyday athletes. To be able to ensure this, artificial intelligence approaches are needed to know how to create personalized value for the athlete/consumer from large heterogeneous data sets.

An application example for this is the home training sector, which is gaining importance especially due to the effects of the Corona pandemic. Commercial platforms offer initial approaches to the use of immersive media but fail to generate individualization of content by analyzing heterogeneous data sources for the user.

The project will therefore investigate mechanisms for user engagement and motivation. Based on this, a comprehensive adaptive AI system for predicting individual goal achievement will be developed and fused with additional data sources. Based on the predictions, a framework for designing stimulus-driven real-time systems for individualizing immersive user interfaces will be defined. The integration of the resulting subsystems into a high-fidelity prototype enables the transfer to further application domains.

Biopsychology is a field of psychology that analyzes how biological processes interact with behaviour, emotion, cognition, and other mental processes. Biopsychology covers, among others, the topics of sensation and perception, emotion regulation, movement (and control of such), sleep and biological rhythms, as well as acute and chronic stress.

While some software packages exist that allow for the analysis of single data modalities, such as electrophysiological data, or sleep, activity, and movement data, no packages are available for the analysis of other modalities, such as neuroendocrine and inflammatory biomarkers, and self-reports. In order to fill this gap, and, simultaneously, combine all required tools for analyzing biopsychological data from beginning to end into one single Python package, we developed BioPsyKit.

The aim of the project is to develop a rapid automated diagnosis (using artificial intelligence) of a fusion disorder in cases of e.g. concussion by measuring the oculomotor function (eye-tracking) for three-dimensional stimuli.

Musculoskeletal (MSK) models represent the dynamics of the human body and can output many different variables i.e. joint angles, joint moments and muscle force. Personalised movement predictions provide accurate outcome variables than a generic prediction. Therefore, we would like to develop a digital twin of the MSK system, which can then be used for personalised movement predictions. Image-based personalisation is the state-of-the-art. Anthropometric variables, such as bone geometry and muscle attachment points can be derived from magnetic resonance imaging (MRI). Muscle parameters require diffusion tensor imaging (DTI) to visualise the alignment of fibres, which is important for the derivation of the muscle size as well as the fibre length. The goal of this project is to develop a personalised digital twin of the MSK system using DTI measurements, and investigate if such a digital twin can improve accuracy of movement predictions.  The aim is to also investigate to what extent image processing can be automated. Furthermore, identification of groups using the personalised models, e.g. to detect MSK diseases, such as rheumatoid arthritis will be investigated. 

The proposed CRC “Empathokinaesthetic Sensor Technology” (EmpkinS) will investigate novel radar, wireless, depth camera, and photonics based sensor technologies as well as body function models and algorithms. The primary objective of EmpkinS is to capture human motion parameters remotely with wave-based sensors to enable the identification and analysis of physiological and behavioural states and body functions. To this end, EmpkinS aims to develop sensor technologies and facilitate the collection of motion data for the human body. Based on this data of hitherto unknown quantity and quality, EmpkinS will lead to unprecedented new insights regarding biomechanical, medical, and psychophysiological body function models and mechanisms of action as well as their interdependencies.The main focus of EmpkinS is on capturing human motion parameters at the macroscopic level (the human body or segments thereof and the cardiopulmonary function) and at the microscopic level (facial expressions and fasciculations). The acquired data are captured remotely in a minimally disturbing and non-invasive manner and with very high resolution. The physiological and behavioural states underlying the motion pattern are then reconstructed algorithmically from this data, using biomechanical, neuromotor, and psychomotor body function models. The sensors, body function models, and the inversion of mechanisms of action establish a link between the internal biomedical body layers and the outer biomedical technology layers. Research into this link is highly innovative, extraordinarily complex, and many of its facets have not been investigated so far.To address the numerous and multifaceted research challenges, the EmpkinS CRC is designed as an interdisciplinary research programme. The research programme is coherently aligned along the sensor chain from the primary sensor technology (Research Area A) over signal and data processing (Research Areas B and C) and the associated modelling of the internal body functions and processes (Research Areas C and D) to the psychological and medical interpretation of the sensor data (Research Area D). Ethics research (Research Area E) is an integral part of the research programme to ensure responsible research and ethical use of EmpkinS technology.The proposed twelve-year EmpkinS research programme will develop novel methodologies and technologies that will generate cutting-edge knowledge to link biomedical processes inside the human body with the information captured outside the body by wireless and microwave sensor technology. With this quantum leap in medical technology, EmpkinS will pave the way for completely new "digital", patient-centred diagnosis and therapeutic options in medicine and psychology.Medical technology is a research focus with flagship character in the greater Erlangen-Nürnberg area. This outstanding background along with the extensive preparatory work of the involved researchers form the basis and backbone of EmpkinS.

The aim of the D02 project is the establishment of empathokinesthetic sensor technology and methods of machine learning as a means for the automatic detection and modification of depression-associated facial expressions, posture, and movement. The aim is to clarify to what extent, with the help of kinesthetic-related modifications influence depressogenic information processing and/or depressive symptoms. First, we will record facial expressions, body posture, and movement relevant to depression with the help of currently available technologies (e.g., RGB and depth cameras, wired EMG, established emotion recognition software) and use them as input parameters for new machine learning models to automatically detect depression-associated affect expressions. Secondly, a fully automated biofeedback paradigm is to be implemented and validated using the project results available up to that point. More ways of real-time feedback of depression-relevant kinaesthesia are investigated. Thirdly, we will research possibilities of mobile use of the biofeedback approach developed up to then.

A novel postural control model of walking is explored to characterise the components of dynamic balance control. For this purpose, clinically annotated gait movements are used as input data and muscle actuated multi-body models are extended by a sensorimotor level. Neuromotor and control model parameters of (patho-)physiological movement are identified with the help of machine learning methods. Technical and clinical validation of the models will be performed. New EmpkinS measurement techniques are to be transferred to the developed models as soon as possible.

Even though sport shoe models can differ in properties (fit, cushioning, bending stiffness etc.) the products come out of mass production and are made for generic groups of athletes. Humans are highly individual and research has shown that athletes are responding individually to certain shoe characteristics. Goal of this research is to create predictions for individual athletes to benefit from the right choice of shoes - be it through increased comfort, better performance or injury prevention.

The integrated Research Training Group (iRTG) offers all young researchers a structured framework programme and supports them in their scientific profile and competence development. The diverse measures provided enable the young researchers to work on their respective academic qualifications in a structured and targeted manner. Particular attention is paid to their networking and their ability to communicate intensively and to take responsibility for their own scientific work. The doctoral researchers are supervised by two project leaders.

We propose a novel way of reconstructing medical images rooted in deep learning and computer vision that models the process how human radiologists are using years of experience from reading thousands of cases to recognize anatomical structures, pathologies and image artifacts. Our approach is based on the novel idea of a variational network, which embeds a generalized compressed sensing concept within a deep learning framework. We propose to learn a complete reconstruction procedure, including filter kernels and penalty functions to separate between true image content and artifacts, all parameters that normally have to be tuned manually as well as the associated numerical algorithm described by this variational network. The training step is decoupled from the time critical image reconstruction step, which can then be performed in near-real-time without interruption of clinical workflow. Our preliminary patient data from accelerated magnetic resonance imaging (MRI) acquisitions suggest that our learning approach outperforms the state-of-the-art of currently existing image reconstruction methods and is robust with respect to the variations that arise in a daily clinical imaging situation. In our first aim, we will test the hypothesis that learning can be performed such that it is robust against changes in data acquisition. In the second aim, we will answer the question if it is possible to learn a single reconstruction procedure for multiple MR imaging applications. Finally, we will perform a clinical reader study for 300 patients undergoing imaging for internal derangement of the knee. We will compare our proposed approach to a clinical standard reconstruction. Our hypothesis is that our approach will lead to the same clinical diagnosis and patient management decisions when using a 5min exam. The immediate benefit of the project is to bring accelerated imaging to an application with wide public-health impact, thereby improving clinical outcomes and reducing health-care costs. Additionally, the insights gained from the developments in this project will answer the currently most important open questions in the emerging field of machine learning for medical image reconstruction. Finally, given the recent increase of activities in this field, there is a significant demand for a publicly available data repository for raw k-space data that can be used for training and validation. Since all data that will be acquired in this project will be made available to the research community, this project will be a first step to meet this demand.

Public Health Relevance

The overarching goal of the proposal is to develop a novel machine learning-based image reconstruction approach and validate it for accelerated magnetic resonance imaging (MRI). The approach is able to learn the characteristic appearance of clinical imaging datasets, as well as suppression of artifacts that arise during data acquisition. We will test the hypotheses that learning can be performed such that it is robust against changes in data acquisition, answer the question if it is possible to learn a single reconstruction procedure for multiple MR imaging applications, and validate our approach in a clinical reader study for 300 patients undergoing imaging for internal derangement of the knee.

The extent to which a neural network can be used to effectively personalise gait simulations using motion data is explored. We first investigate the influence of body parameters on gait simulation. An initial version of the personalisation is trained with simulated motion data, since ground truth data is known for this purpose. We then explore gradient-free methods to fit the network for experimental motion data. The resulting network is validated with magnetic resonance imaging, electromyography and intra-body variables.

Repeated exposure to acute psychosocial stress and the associated stimulation of biological stress pathways over a period of time can promote the transition from acute to chronic stress. Unfortunately, established laboratory stress protocols are limited for repeated use due to high personnel and resource demand, creating the need for novel approaches that can be conducted at a larger scale, and, possibly, remotely.

Therefore, this project aims to develop and test novel methods for inducing acute stress without requiring extensive personnel and resource demands. The project will explore the use of digital technologies, such as virtual reality and mobile apps, to create stress-inducing scenarios that can be experienced (remotely) by study participants. The project will also investigate the use of physiological and behavioral measures to validate the effectiveness of the stress induction methods.

Human movement is a complex process that depends on many factors such as body constitution, health condition, but also external factors. Joint angles, joint moments and muscle forces are variables quantifying the movement to give valuable insights about these factors. Simulation of musculoskeletal models can be used to perform detailed movement analysis to obtain these variables. The application of simulation is two-fold: Reconstruction of measured motion and prediction of new motion. Motion reconstruction can give valuable insights for example for sports analysis in marathon runners or medical gait assessment of Parkinson’s patients. Simulation can predict changes in human motion in response to environmental changes. This is beneficial to, for instance, support virtual product design of footwear or below-knee prostheses.

However, for accurate and detailed simulations, the personalization of musculoskeletal models is crucial. Precise scaling of segment and muscle parameters can be achieved using magnetic resonance imaging (MRI) which requires time and cost consuming measurements additionally to the movement acquisition and expert knowledge. State-of-the-art methods relying only on movement recordings scale segment parameters and muscle attachment points. But they do not scale muscle parameters like maximum isometric forces.

We will combine optimal control simulation with the application of advanced machine learning methods to personalize segment as well as muscle parameters based on marker and ground reaction force. The goal is to make personalized simulations feasible in healthcare, sports science, and industrial practice. To this end, we aim at developing an approach with three key improvements: First, it can be applied without additional and time-consuming measurements using expensive modalities; Secondly, it can be used without expert knowledge but operates automatically; Thirdly, it is feasible with limited computational re-sources, i.e., computational power and time.

There is a wide range of medications for RA patients, Clinical trials and real-time experience demonstrate that sometimes these treatments have adverse effects, for better benefits and later minimizing the damage, we should predict the response for each person.

This project aims to collect medical data on rheumatology arthritis, select the best factors and identify important clinical features associated with remission and then create a model to predict remissions in patients and prediction of treatment response and course of activity for each patient using machine learning methods. This project could help in preventing wrong prescriptions and time-wasting before disease progression.

We want to reach the aim by using medical data collected and recorded by rheumatologists from patient characteristics, disease courses, laboratory data, and medication data. Our partners from the medicine side are helping to collect and access existing data. The partners in MaD-Lab carry out Machine learning and data analytics approaches on them to find a remission or development of the prognostic model.

In D04, innovative, non-contact EmpkinS sensor technology using machine learning algorithms and multimodal reference diagnostics is evaluated using the example of Parkinson’s-associated sleep disorder patterns. For this purpose, body function parameters of sleep are technically validated with wearable sensor technology and non-contact EmpkinS sensor technology in comparison to classical poly-somnography and correlated to clinical scales. In an algorithmic approach, multiparametric sleep parameters and sleep patterns are then evalulated in correlation to movement, cardiovascular and sleep phase regulation disorders.

The broad mission of our Center for Advanced Imaging Innovation and Research (CAI2R) is to bring together collaborative translational research teams for the development of high-impact biomedical imaging technologies, with the ultimate goal of changing day-to-day clinical practice. Technology Research and Development (TR&D) Project 1 aims to replace traditional complex and inefficient imaging protocols with simple, comprehensive acquisitions that also yield quantitative parameters sensitive to specific disease processes. In the first funding period of this P41 Center, our project team led the way in establishing rapid, continuous, comprehensive imaging methods, which are now available on a growing number of commercial magnetic resonance imaging (MRI) scanners worldwide. This foundation will allow us, in the proposed research plan for the next period, to enrich our data streams, to advance the extraction of actionable information from those data streams, and to feed the resulting information back into the design of our acquisition software and hardware. Thanks to developments during our first funding period, we are now in a position to question long-established assumptions about scanner design, originating from the classical imaging pipeline of human radiologists interpreting multiple series of qualitative images. We will reimagine the process of MR scanning, leveraging our core expertise in pulse-sequence design, parallel imaging, compressed sensing, model-based image reconstruction and machine learning. We will also extend our methods to complex multifaceted data streams, arising not only from MRI but also from Positron Emission Tomography (PET) and other imaging modalities, as well as from diverse arrays of complementary sensors.

Onlinebetrug verursachte 2019 weltweit ca. 56 Mrd. $ Schaden. Die Methoden dafür werden stetig komplexer und damit schwerer aufzudecken. Im Vorhaben sollen digitale Betrugsidentitäten und Betrugsformen durch KI in Echtzeit erkannt werden.

Digitization is advancing at a high pace in all areas of our lives. The development of algorithms based on artificial intelligence and more extensive and better data sets are creating new opportunities in many areas of science.

Nevertheless, it is common for many biologists, veterinarians, and animal caretakers to observe the animals manually, which is very time- and labor-expensive and comes with severe limitations. The development of an automated camera-based system seems to be self-evident.

The automated analysis of video footage from surveillance cameras is a possibility to evaluate activity/inactivity and stereotypy of the animals as well as their enclosure use over long periods of time and thus allows an objective assessment. In the project's next phase, further behavior analysis will be added (for example, secondary behavior such as feeding, playful behavior, or interaction). The most challenging step for such a system is the re-identification (reID) of individual animals in every camera perspective

Electronic health records (EHR) are commonly institution-specific, provided by hospitals, insurance companies, or other institutions to fulfill their own objectives, thus, causing stored health information to be isolated, fragmented and duplicated across providers. Consequently, patients may lack complete access to their medical histories. As a solution, countries such as Denmark and Israel have long adopted nationwide EHR for their health care systems^a, in which the health information is well managed and digitally connected to avoid duplicate records and improve the quality and co-effectiveness of medical care as well as patient safety. In Germany, the “Appointment Service and Supply Act” adopted on 14th March 2019, requires the German statutory health insurance to provide EHR for all insured persons from 1st January 2021 onwards^b. As specified by German Health Care Information Technology Infrastructure in accordance with section 291a SGB V, the new EHR should store complete medical histories of patients such as previous diagnoses, therapeutic decisions, treatment reports and self-measurement values. Among other benefits, patients should have power to select freely between providers, hold data sovereignty for their EHR, and withdraw access rights at any time.

In line with those guidelines, OnePatient^c is a patient-centered EHR system that stores data locally under the sovereignty of individual device owners, thereby enabling patients to take control of their health information, provide offline access to medical data, ensure privacy management and to avoid a single point of failure. The OnePatient EHR system can be provisioned on any of the patients’ devices; therefore, patients technically own their medical data while the device and software manage it. On the one hand, these developments simplify the technical and organizational challenges to implement data regulations such as the General Data Protection Regulation (GDPR) of the European Union. On the other hand, the data will not only be in isolated, heterogeneous and distributed environments but also pose a new challenge to the conventional data transaction procedures employed in machine learning (ML) today [4]. The traditional procedures for acquiring big data in ML involve several parties from collecting the data, transferring it to a central data repository and fusing it to build a model, whereas the data owners may be unclear about these procedures and the model future use cases, for that reason, may violate laws such as GDPR.

Therefore, to address these challenges, federated learning (FL) approaches can be leveraged to build ML models that can be sent to train locally–where the data is located. In this manner, only the model updates that contain anonymous results which cannot be reverse-engineered are returned to the central data repository. Leveraging FL and the account of the FL existing studies [1; 2; 3], although not focusing on the emerging EHR systems’ architecture like OnePatient, we aim to attain four objectives. The first is to investigate and design novel FL frameworks that enable local systems to collaboratively train a ML model that patients can benefit from without divulging their medical information to a central entity; moreover, medical practitioners will be able to access the training process of the FL frameworks to adjust the diagnostic criteria of the model, and therefore increase trust and accuracy of the model outcome. Secondly, we aim to investigate and compare the accuracy and performance of the model trained in a centralized way and the FL frameworks that will be proposed. Thirdly, to protect the data during training from potentially malicious models and participants, we aim to use countermeasures such as differential privacy and multi-party computation to ensure privacy guarantees. Finally, for proof of concept, we aim to demonstrate the effectiveness of FL frameworks using existing databases and suitable ML tasks with the data.

1. Brisimi, T. S., Chen, R., Mela, T., Olshevsky, A., Paschalidis, I. C., & Shi, W. Federated learning of predictive models from federated electronic health records. Int.J.Med.Inf. 2018; 112: 59-67. 

2. Roy, A. G., Siddiqui, S., Pölsterl, S., Navab, N., & Wachinger, C. Braintorrent: A peer-to-peer environment for decentralized federated learning. arXiv preprint arXiv:1905.06731 2019; 

3. Xu, J., & Wang, F. Federated learning for healthcare informatics. arXiv preprint arXiv:1911.06270 2019;

4. Yang, Q., Liu, Y., Chen, T., & Tong, Y. Federated machine learning: Concept and applications. ACM Transactions on Intelligent Systems and Technology (TIST) 2019; 10: 1-19. 

a. https://www.gesundheitsindustrie-bw.de/en/article/news/ehr-and-phr-digital-records-in-the-german-healthcare-system
b. https://www.gematik.de/anwendungen/e-patientenakte/#
c. https://refinio.net/software.html

The hearing aid of the future will be more than just an amplifying device. It may be used as fitness tracker to capture the user’s movements and activity level. Furthermore, it may be used as home monitoring device assessing the user’s vital parameters, tracking the user’s activity status or detecting falls. Hearing aids are becoming more complex and most modern hearing aids are already equipped with additional sensors such as inertial sensors. Acceleration signals are analyzed with signal processing algorithms to enhance speech intelligibility and audio quality. Furthermore, inertial sensors may be used to analyze the user’s movements and physical activity. Hearing aid amplification settings may be adapted according to the current activity. Moreover, given the user's explicit consent, activity recognition enables a long-term tracking of the user’s daily activity status.

The objective of this project is to investigate automatic activity recognition based on inertial sensor data. Therefore, data of different activities will be recorded using the IMU sensor integrated in the hearing aids. The hearing aids are provided by the cooperation partner WS Audiology. Machine learning algorithms will be developed to automatically classify different activity patterns. 

The complex biathlon performance consists of the components running speed, shooting stand stay (shooting stand and shooting time), and the shooting result together. Competition analyses of the last years show that the hit rate continuously increased. At the 2014 and 2018 Winter Olympics, all Placed on the podium with hit rates of over 95%. This trend continues unabated. The shooting performance of German athletes in high performance and follow-up training shows in this comparison reserves, especially with regard to the consistency of biathlon shooting in competitions, and the to the season's highlight.

Within the scope of this project, a mobile and affordable sensor system for the detection of muzzle movements can be developed. This could enable all biathletes in the future to be DSV's high-performance sports system that can benefit from technical support for dry training. The measurement system will consist of an inertial measurement unit (IMU) and a smartphone app. The IMU is attached to the rifle barrel by a special holder during training and measures the movement of the muzzle of the barrel. In the course of the training, the data collected in this process are transmitted wirelessly and in real-time to the smartphone and evaluated there. 

The system is designed to provide feedback on the current action directly during practice as well as feedback on the muzzle stability at the end of training in the period shortly before the shot is fired. These values should not only be recorded in dry training, but also in training and competition mode, in order to be able to compare the values in dry training with the complex biathlon performance. By analyzing the values in the longitudinal section, individual training effects can be identified and training recommendations can be derived accordingly. 

Breast cancer is one of the leading causes of death in the field of oncology in Germany. For the successful care and treatment of patients with breast cancer, a high level of information for those affected is essential in order to achieve a high level of compliance with the established structures and therapies. On the one hand, the digitalisation of medicine offers the opportunity to develop new technologies that increase the efficiency of medical care. On the other hand, it can also strengthen patient compliance by improving information and patient integration through electronic health applications. Thus, a reduction in mortality and an improvement in quality of life can be achieved. Within the framework of this project, digital health programmes are going to be created that support and complement health care. The project aims to provide better and faster access to new diagnostic and therapeutic procedures in mainstream oncology care, to implement eHealth models for more efficient and effective cancer care, and to improve capacity for patients in oncologcal therapy in times of crisis (such as the SARS-CoV-2 pandemic). The Chair of Health Management is conducting the health economic evaluation and analysing the extent to which digitalisation can contribute to a reduction in the costs of treatment and care as well as to an improvement in the quality of life of breast cancer patients.

This new course is designed to familiarize engineering students with the basic procedures and tools of deep learning. A high level of application relevance is ensured by concrete examples from the respective engineering disciplines, such as mechanical engineering, materials science, production engineering, electrical engineering, process engineering or medical technology. The new course offering is designed to enable students to efficiently apply more complex methods of machine learning as a tool for the analysis of large amounts of data. To this end, the course is divided into several blocks, each of which consists of theoretical and practical modules or units. Our first VHB course ML4Engineers I introduces the basics of machine learning, while the current course will cover more advanced topics.

Mobility in atypical parkinsonism: effects of physiotherapyWider research context/theoretical frameworkParkinsonian gait disorders and reduced mobility are pivotal symptoms of Parkinson´s disease (PD) and of atypical parkinsonian disorders (APD), including Multiple System Atrophy (MSA) and Progressive Supranuclear Palsy (PSP). Their onset signs the transition towards disability and increased mortality. While several randomized controlled trials investigated efficacy of exercise-based interventions for PD, for APD this area remains thus far widely unexplored. Results of our pilot study investigating efficacy of a physiotherapy program in APD showed improvements of gait parameters as reflected by instrumented gait analysis in lab. Current advances in development of wearable sensor-based technologies now reach clinical applicability also under remote unsupervised conditions (home-monitoring).Hypotheses/research questions /objectivesIn this randomized controlled trial we expect to detect greater improvement of gait performance and mobility in PD and APD patients by gait-focused versus standard physiotherapy and home-training. In particular, we aim to investigate whether:1) Gait-focused versus standard physiotherapy and home-based exercise improve lab and home-based gait parameters, physical activity and clinical rating scales in PD and APD patients.2) PD, MSA and PSP patients differ in their response to gait-focused versus standard physiotherapy and home-based exercise as detected by lab and home-based gait parameters, physical activity and clinical rating scales.Approach/methodsThe trial will be multicentric, randomized, double-blinded and controlled. The intervention to test consists of an inpatient gait-focused physiotherapy followed by an unsupervised gait-focused home-based training program. The control group receives a standard physiotherapy/home-based training, which addresses parkinsonian features without focusing on gait. Lab gait analysis is performed by shoe-insole-sensors to retrieve objective gait parameters (e.g. gait velocity, stride length etc.). Home-monitoring is performed using similar shoe-sensors and one sensor at low-back position allowing for the analysis of routine activities such as sitting, lying, standing and walking.Level of originality/innovationImproving gait disorders in PD and APD patients by gait focused PT and home-based exercise increases patients´ independence and forestalls the risk of falls representing a great achievement for patients. This study lays the foundation for the development of a telemedical approach by which patient groups can be included in clinical trials remote from expert centers.Primary researchers involvedGregor Wenning (lead-PI), Cecilia Raccagni (study coordinator), Jochen Klucken (Erlangen-PI), David Benninger (Lausanne-PI) and Bas Bloem (Nijmegen-PI).Björn Eskofier (Erlangen) and Kamiar Aminian (Lausanne) will be responsible for the technical part concerning the sensor system.

Sensorische Anwendungen finden heutzutage durch moderne Technologien (v.a. Smartphone/Smart-Watch vermittelt) vielfach Einzug in den Alltag. In diesem Zuge stellt sich die Frage, inwieweit auch sensorische Messungen der regulären Schwangeren-Vorsorge (Herzfrequenz, Blutdruck, Sonografie und Kardiotokografie), die dem Standard nach in der Hand des Arztes oder der Ärztin liegen, in den Smart-Home Bereich transferiert werden und valide Ergebnisse liefern, sowie zukünftig die Klinik-besuche schwangerer Frauen reduzieren bzw. spezifizieren können. Im Fokus der Fragestellung dieses Projekts steht die klinische Usability, die gesellschaftliche Akzeptanz, die Compliance durch die betroffenen Akteure und die Weiterentwicklung dieser sensorischen Techniken im häuslichen Bereich sowie damit assoziierte ethisch/medizinrechtliche Themen.

Ziel des Projektes ist, die Vorsorge für schwangere Frauen zu optimieren und zu vereinfachen, indem sowohl bewährte als auch innovative Sensorik in die Heim-Versorgung überführt und mit künstlicher Intelligenz und maschinellem Lernen analysiert wird. In diesem Projekt werden direkte Anwendungsmöglichkeiten zur Implementierung der Smart-Sensorik geschaffen, welche die optimierte Gesundheitsbetreuung durch die Ärztin oder den Arzt, aber auch die eigene Kontrolle und Optimierung der metabolischen Aktivität durch die schwangeren Frauen ermöglicht. Als Zielgruppe sind schwangere Frauen und deren Partner/innen angesprochen, die offen sind für die gesundheitsbezogene Anwendung moderner, digitaler Medien (Smartphone, Smart-Watch etc.).

The HOOP project aims to:

1)       Develop of a universal platform to train PD patients. Parkinson’s disease (PD) patients

repeatedly report great disparities in access to, and quality of, the full range of

Parkinson’s disease services. This represents a social risk for them in terms of access to

medical specialists, caregivers or medical therapy. The value of therapy services in PD

has outlined the importance of having access to these interventions from an early

stage of the disease. A solution which improves the delivery of training to PD patients,

and represents an opportunity to enroll and follow-up, personalized physical therapy

programs at home are important for promoting patient’s long-term engagement and

self-management of their own health.

2)       Improve the quality of life of PD patients through the platform. In addition to the

benefits provided to Parkinson’s patients and according to recent studies showing that

in 2016 a 39,3% of the population in Europe is over 50 years old, which means that

there will be an increase in the mobility and fragility issues due to the aging. In this

sense, it is very important to promote systems and solutions that encourage healthy

living habits with the aim of improving the quality of life.

3)       Provide a final product available and ready for market u p-take. During 2017, HOOP is

being developed from a research laboratory prototype to a semi-commercial product.

Various exploitation plans are being investigated and several potential target markets

related to Parkinson’s disease have shown interest in our product. This means that

HOOP is a mature project, which is close to achieving commercialisation. However, in

order to achieve the latest and most important goals of the project, a commercial

version of the prototype must be finalised, tested and validated. Moreover, the

possibility to extend the target markets would increase the added value of the

product.

The patient-centred healthcare ecosystem (PHE) is a future digital healthcare model where all healthcare stakeholders will eventually transition from an isolated approach to a collaborative approach around the patient. Among other benefits, the PHE will enable individuals to take control of their health information in a confidential and secure environment. Currently evolving PHEs either use a centralized database or blockchain technology for storing medical records.

On the one hand, healthcare industries that store data in a centralized database experience more data breaches than any other sector, as revealed in the latest report by the Office of the Australian Information Commissioner on data breaches. Moreover, the Protenus Breach Barometer in the US reported 369 health data breaches in the third quarter of 2018, affecting 8 million patients. On the other hand, the immutable nature of data storage in blockchain makes it impossible for users to erase their stored information, which goes in contrary to the European regulation on data protection.

The OnePatient PHE by Refinio ONE (a German-based health technology startup) is based on peer-to-peer technology; an alternative to centralized database and blockchain technology. Although blockchain also uses peer-to-peer technology, the data is inherently shared publicly, and in case the blockchain encryption gets broken, all the data becomes public. In contrast, when the OnePatient PHE encryption gets broken, only one user is affected at a time. However, storing medical records in peer-to-peer technology still requires research in terms of security and education and awareness to users about data security and privacy.

In our work, we aim to i) investigate the possible and inherent security and privacy issues for future PHEs like OnePatient, ii) design security models such as Firewall, Trust Reputation System, etc. to provide additional security, and iii) finally evaluate the effectiveness of our security models. 

Parkinson’s disease (PD) is a chronic movement disorder characterized by progressive gait impairment, leading to reduced mobility, poor quality of life and frequent falls. moveIT improves healthcare for patients via wearable gait&fall sensors enabled as healthcare products by an innovative digital health pathway (DHP). The DHP defines the clinical application of these new technologies for multidisciplinary healthcare using stratified patient cohorts and care networks targeting gait&falls in PD.

Drivetrain technology companies are presented to the emerging challenge to meet the comprehensive demands of modern systems by consolidating the competencies of the latest developments in the field of machine learning. The efforts for Schaeffler to establish those competencies are easier to accomplish by including a strong research partner. This partner for the given project will be the Machine Learning and Data Analytics Lab of the FAU. As a result of this project Schaeffler and the business unit obtain methodological competences in general as well as new approaches of deep learning algorithms for signal analysis in particular.

The goal of this project is the investigation of multimodal methods for the evaluation of interventional workflows in the operation room. This topic will be researched in an international project context with partners in Germany and in Brazil (UNISINOS in Porto Alegre). Methods will be developed to analyze the processes in an OR based on signals from body-worn sensors, cameras and other modalities like X-ray images recorded during the surgeries. For data analysis, techniques from the field of computer vision, machine learning and pattern recognition will be applied. The system will be integrated in a way that body-worn sensors developed by Portabiles as well as angiography systems produced by Siemens Healthcare can be included alongside.

The aim of the project is to design and implement a system that is able to holistically assess athletes. Therefore, novel individualised digital services, products and tools to enable a holistic mental and physiological performance optimisation will be developed. This is an improvement compared to current state of the art methods, since these approaches only consider physiological data for optimising performance.

This project investigated methods and techniques of machine learning in the field of localization. Deep neural networks have been used to model nonlinear signal propagation in radio-based localization to enable position determination even in difficult metallic environments. Furthermore, it was investigated to what extent the temporal consideration of radio signals by means of recurrent neural networks can provide an added value directly in the localization algorithm and to what extent these can be efficiently combined with classical methods (e.g. Kalman filters).

In addition, we obtained first results regarding a fusion with camera data. Radio-based localization systems have advantages over optical localization technologies when it comes to occlusions. On the other hand, radio-based systems have problems with metallic structures/surfaces, since the radio waves are reflected on metallic surfaces and are thus received via several paths at the receiving antennas. A fusion filter has been developed, which compensates the mutual weaknesses of the systems and allows an exact but at the same time robust tracking.

Stress is a hidden epidemic – the World Health Organization estimates that mental diseases, including stress-related disorders, will be the second leading cause of disabilities by the year 2020. Since the negative economic impact of stress is substantial, there is an interest in detecting stress-related diseases as early as possible for early intervention, such as precision medicine approaches.

Stress can be differentiated into chronic and acute stress. As an example, social interactions with others can trigger acute stress. This response is characterized by strong biological reactions that affect the whole body through widely spread autonomic innervation and the secretion of stress hormones. Whereas adequate stress responses are a crucial and healthy physiological reaction, defective stress responses have been linked to DNA damage, over-expression of inflammatory genes, and declines in cognitive functioning, which are well known markers of physiological and biological age.

Therefore, the goal of this work is to use machine learning methods for the classification of those stress response patterns. Compared to classification by a trained professional this approach has the potential to reduce the required time, as well as increase the objectivity of the grouping.

The main goal of this project is to improve the overall system quality and customer satisfaction.

In this project, we analyze IoT data (machine logs and sensory data) sent by thousands of high-end medical devices every day. The extracted information can include physical parameters and additional extracted event patterns. This data can be used to predict the failure of specific components and correlate malfunction to machine usage. As a consequence, system stability can be improved and procedures for system testing can be recommended.

Furthermore, information from customer service data (e.g. tickets) is processed and fused with machine data to predict customer sentiment. With that, customer satisfaction can be improved via proactive service.

Methods designed and used include:

 

• Time Series Analysis (esp. mixed-typed and irregularly sampled)
• Deep Learning
• Process Mining
• Data Fusion
• Text Mining

Teamworkand communication between group members get increasingly important in the taskof solving complex problems [1]. To measure the social interactionand the dynamic of groups, members of the MIT Media Lab have developed an opensource sensor platform called “Open Badges”. The badges are usedto determine several parameters like vocal activity, proximity to other membersand location during a meeting, a workshop, a discussion or an event. Afteranalysis of these parameters real-time feedback can be given to improve thebehavior of members within a group. 

Inthis joint MIT-FAU project, the existing sensor platform is expanded andoptimized for modularity and maintainability. Further measurement capabilitiesare included in the platform and analysis and energy-optimization algorithmsare implemented.

The ability to walk defines human nature and limits the mobility, independence, and quality of life. Falls are the leading cause of both fatal and nonfatal injuries among older adults, causing severe injuries such as hip fractures, head trauma, and death. Increased fall risk is a key symptom in Parkinson’s disease (PD), limiting the independency and mobility of patients.

So far, no validated technical solutions exist to identify the individual’s rising fall risk before the first fall occurs. Therefore, we will investigate algorithms, that are able to predict the fall risk based on specific gait patterns, captured by shoe integrated inertial sensors. The data for the evaluation of fall risk associated gait patterns will be acquired by means of a continuous long-term monitoring system.

To ensure a successful progress of this project we will combine three strategies in the research and development phase:

1. Usage of distinct sensors that enable gait assessment with high biomechanical resolution
2. Development and evaluation of machine learning based gait pattern algorithms
3. Digital biobanking of clinical distinct gait patterns to individualize fall risk monitoring.

The overall goal of the project is the investigation of novel machine learning based algorithms that enable the determination of PD patients’ fall risk using continuous gait data. Since existing algorithms and test procedures in related clinical research are typically limited to one-time assessments, we will investigate new algorithms for a continuous gait monitoring system, that will identify disease specific changes of gait with a high reliability.

At the same time, we will generate clinical understanding and validated data for individualized applicability that is required for medical product licensing, as well as economic effect sizes of technology application in healthcare strategies.

Vision Impairmentslike amblyopia can be treated by digital therapies using virtual reality systems.These therapies help to avoid or at least reduce the lifelong consequences dueto developmental disorders. Amblyopia is a functional vision impairment of asingle eye based on insufficient development of the visual system in earlychildhood. In Central Europe the prevalence of amblyopia is about 5 – 6%.Amblyopia is the main reason for vision impairments during the first 45 yearsof life. Occlusion therapy as the current gold standard in therapy treats onlythe weak eye at the expense of binocular vision and is associated with disadvantages:amblyopic childs treated by occlusion therapy showed a decrease in visual acuityin the healthy eye by almost half of the increase in visual acuity in theamblyopic eye after the treatment. The here proposed VR-AMBLOYPIE TRAINER isnot intended to show these deficits as it does not utilize occlusion but ratherimproves binocular vision. Virtual Reality is said to be the upcoming keytechnology in medicine. Educational, diagnostic and therapeutic approaches increasinglymake use of this technology.

The aim of the novel therapeuticconcept of the VR-AMBLOYPIE TRAINER is the increase of visual acuity in the weakeye as well as training the binocular vision. The task of the amblyope is to returnballs to a virtual player. The amblyope is playfully interacting with the system.Thereby the task can only be accomplished if both eyes work together andimportant depth information can be extracted to correctly return the ball. Repetitiveperforming of this visual task with increasing level of difficulty helps to improvebinocular vision. 

Autonomous applied logistics, especially in context of Industry 4.0, is an important factor for fully automated systems. These systems involve autonomous operation of loading and unloading processes, safety measures by detecting persons in danger zones and the general optimization of the logistical processes.

Especially, the application of these systems in a harbor environment, where different systems from all over the world interact, increases the complexity of the loading and unloading processes. The aim of this research project is to determine the feasibility of automating the unloading process by segmenting, classifying and fitting laser range data by using Machine Learning techniques.

In this project, we investigate musculoskeletal modeling and simulation to analyze and understand human movement and performance. Our objective is to reconstruct human motion from measurement data for example for medical assessments or to predict human responses for virtual product development.

 

Reconstruction of Human Motion: Biomechanical analysis using wearable systems

Inertial sensor systems provide the possibility of cheap gait analysis in everyday life. One major challenge is to achieve a high quality gait analysis based on noisy sensor measurements. Moreover, inertial sensors can only quantify human joint kinematics and are not able to measure joint kinetics as performed in gait laboratories. Existing systems are based on an integration of the inertial sensor data for estimating human poses. This error-prone integration can be avoided using a computer simulation of a biomechanical model that tracks the measured sensor signals. Furthermore, such a model can give insight into joint kinetics, muscle control and other gait-related parameters such as stride length, stride time and ground-reaction force.

 

Synthesis of Human Motion: Predictive biomechanical simulation for design applications

Sports and medical products such as running shoes, bandages or prostheses should support and improve our movement. But, how to derive optimal design parameters? The conventional process of prototyping and testing is often time-consuming, expensive, hazardous or even not realizable. Our purpose is to avoid prototyping and testing by virtual product development to derive optimal design parameters. We investigate biomechanical simulation to predict the influence of design parameters on human movement and performance.

Die Digitalisierung birgt große Potenziale zur Steigerung der Ressourceneffizienz industrieller Produktionsprozesse. Durch Technologien im Kontext von Industrie 4.0 können produktionsnahe Daten kurzzyklisch erfasst und aggregiert werden. In Anbetracht der dadurch zunehmenden Datenkomplexität und des Datenvolumens stehen Mitarbeiter jedoch vor der Herausforderung, diese Daten zu analysieren und zu interpretieren sowie die Nachhaltigkeit der eingeleiteten Maßnahmen zu bewerten, wobei die kognitiven Fähigkeiten oft an ihre Grenzen stoßen.

Verfahren des Maschinellen Lernens (ML) können hier neue Formen der Arbeitsteilung zwischen Maschinen bzw. Software als Entscheidungsvorbereiter und Mitarbeitern als Problemlöser zu ermöglichen. In der industriellen Praxis werden ML-Verfahren meist situativ und von Experten entwickelt eingesetzt, so dass der Aufwand entsprechend hoch ist. Des Weiteren verfügen kleine und mittlere Unternehmen (kmU) häufig nur über wenig Ressourcen und Expertise, um diese Potenziale zu nutzen.

Ziel dieses Projektes ist es, ein Qualifizierungskonzept zu entwickeln und durchzuführen, um den Kenntnisstand bzgl. ML-Verfahren von Mitarbeitern in Produktions- und Qualitätsbereich sowie von Studierenden mit den genannten Schwerpunkten gezielt zu erweitern. Die Teilnehmer entscheiden sich dabei entweder für die Spezialisierungsrichtung "Produktion" oder "Qualität". Jede Spezialisierungsrichtung besteht aus vier praxisorientierten Anwendungsfällen, in denen die Teilnehmer geeignete ML-Verfahren kennenlernen und in konkreten individuellen Projekten mit ca. 10 Wochen Dauer anwenden. Die Anwendungsphase wird von der wissenschaftlichen Leitung des Projekts individuell gecoacht. Die Anwendungsfälle orientieren sich an bestehenden Geschäftsprozessen und Problemstellungen in der Industrie zum Qualitätsmanagement und zur Optimierung von Produktionsprozessen, wodurch ein einfacher Transfer und eine hohe Akzeptanz auf industrieller Seite sichergestellt werden soll. 

The Innovation Lab for Wearable and Ubiquitous Computing is a project funded by the Center for Digitalization Bavaria. The goal of this project is the implementation of a practical course, where students develop innovative prototypes in the fields of Wearable and Ubiquitous Computing by applying agile development techniques. The project ideas originate from three sources: the students themselves, researcher or external industry partners.

Performance Analysis in team sports is an emerging field in computer science. In Europe's leagues, a large amount of data is recorded during the season. Based on methods of machine learning and signal processing an automated, fast and accurate analysis of matches is possible.

In this project, the performance of a single player and the behavior of the whole team (e.g. tactics) is calculated based on position and inertial sensor data.

Gangstörungen und Mobilitätseinschränkungen treten bei einer Vielzahl chronischer Erkrankungen auf und verursachen in Deutschland ca. 20% der Kosten im Gesundheitswesen. Die Veränderung des Gangbildes ist charakteristisch für den Fortschritt dieser Erkrankungen. Sie wird durch die behandelnden Ärzte zur diagnostischen Bewertung der Erkrankung und zur Unterstützung der therapeutischen Entscheidungen auch heute schon visuell befundet.

Das Parkinson Syndrom (PS) zählt zu den klassischen chronischen Bewegungserkrankungen und ist die häufigste neurodegenerativ bedingte Bewegungserkrankung. Der chronisch fortschreitende Verlauf führt zu einer langsamen und ständigen Zunahme der Symptome und einer fortwährenden Notwendigkeit, die Therapien entsprechend anzupassen. Die charakteristischen Bewegungs- und Gangeinschränkungen sind dabei im klassischen, medizinischen Versorgungsprozess sowohl stationär, als auch ambulant Ziel der Diagnostik, um die Krankheitsentität und das Ausmaß der Bewegungseinschränkung und somit der Reduktion der Lebensqualität zu erfassen.

Ziel dieses Projektes ist die prototypische Umsetzung einer telemedizinischen Ganganalyse zur Unterstützung bei der Therapieeinstellung und zum Nachweis von Therapieeffekten bei Parkinson Patienten. Dafür sollen Ganguntersuchungen mit tragbaren Sensortechnologien aus der klinischen Laborumgebung in ein nicht-supervidiertes Monitoring (d.h. ohne Anwesenheit einer medizinischen Fachkraft) im häuslichen Patientenumfeld überführt werden.
Die auf Basis dieses engmaschigen Alltags-Monitorings erhobenen Gangparameter  und Patientenselbsteinschätzungen sollen dem Therapeuten Aufschluss über die Wirkung seiner verordneten Therapie geben. Damit wird der behandelnde Arzt in die Lage versetzt, seine Therapie (z.B. Medikation) telemedizinisch individuell auf seinen Patienten anzupassen. Der zurzeit notwendige stationäre Aufenthalt zur Medikationseinstellung könnte dem Patienten erspart und eine frühere Einstellung auf eine optimale Therapie unterstützt werden. Dies verbessert  die Lebensqualität des Patienten und reduziert direkte (Krankenhauseinweisungen) und indirekte Kosten (Komorbiditäten) im Gesundheitssystem.

The project ‚Schlafkur mit Begleitanalyse‘ aims to close a major gap in the market and offer citizens with sleep difficulties an accessible, effective and digitally supported solution. According to the Health Report 2017 of the DAK Gesundheit, 80% of the working population in Germany sleep poorly. One in ten workers suffer from sleep disorders, thus affecting over 1.5 million people in Bavaria.

We scientifically evaluate the combination of intelligent sleep intervention technology for sleep support, an intelligent sensor for sleep diagnostics and an individualized, IT-based intervention program for each participant with sleep problems. We plan to offer an efficient, low cost service to improve sleep as part of the digital health care industry in Bavaria.

The target group is mainly those in the 30-65 year-old age range of working persons suffering from non-organic sleep disorders. Because there are many types of non-organic sleep disorders: from severe, long-term insomnia to sleep disorders during menopause to mild, recurring problems with falling asleep or remaining so. Usually, these symptoms are independent of a true sleep disorder. Despite a considerable number of affected people, there is a lack of appropriate (self-) analysis and insight, due, in part, to poor social recognition of sleep disorders as well as a lack of diagnosis of the widespread disease "insomnia". This means that the mild to moderate limitations of the affected people do not have access to the appropriate medical or psychological treatment. This project now addresses this population aiming to improve the quality of life and to prevent chronic sleep disorders in the medium term; we provide the patient with individual and professional coaching to recover a restful and peaceful sleep.

Dieses Vorhaben zielt darauf ab, neben neuen universitären Lehrveranstaltungen fürMasterstudierende, klein- und mittelständische Unternehmen (KMU) dabei zuunterstützen, die im Rahmen der Digitalisierung verfügbaren Daten mittelsmaschinellem Lernen (ML) sinnvoll zu nutzen und damit Entscheidungsprozesse zuunterstützen. Durch die thematische Aufstellung der Projektpartner an der FAUstammen Seminare, Praktika und Vorlesungen aus unterschiedlichen Säulen derInformatik und bieten somit vielfältige Vertiefungsmöglichkeiten für Studierende.Ergänzend wird in den industrieorientierten Hands-On-LABs mit Präsenzzeit dasTheoriematerial der Seminare, Praktika und Vorlesungen für praktischeProblemstellungen der Industrie aufbereitet. Es sollen z.B. Lösungsansätze fürdie in Deutschland relevanten Bereiche Industrie 4.0, Produktion oderAutomotive erarbeitet werden.

In this project wearables for sports or rehabilitation are developed. They can be used for training control or analysis.

The aim of this project is the development of a digital, visual perceptual learning system (D-VPL) with gesture recognition and telemedical link to ophthalmologists for visually impaired elderly people, dementia prevention, and patients with traumatic brain injury. The users will react by using gestures to moving objects that will be presented in virtual reality or on a 3-D display. The combination of D-VPL and gesture control leads to a dual task training, and the telemedical link enables applications in medical institutions, senior residences, and rehabilitation facilities.

The increasing complexity of radiotherapy poses a serious challenge toquality management processes. Proper functioning of devices, precisecontrol of process variables and thus ultimately the safety of thepatient regarding exposure to radiation must be ensured. However, Thecomplexity of the radiotherapy workflow lead to seriouos accidents inthe past. Up to date, the quality of only partial radiotherapy workflowsteps is properly assessed and the workflow as a whole is not assessed.The goal of this project is to develop an integrated an automaticquality management information system for a proactive error prevention.Based on quality measures, the whole workflow will be monitored. DataMining, Benchmarking and machine learning tools will be used to detectpotential faults in advance.

The ability to relax is sometimes challenging to achieve, nevertheless it is extremely important for mental and physical health, particularly to effectively manage stress and anxiety. For this reason, we propose a virtual reality experience that integrates a wearable, low-cost EEG headband and an olfactory necklace that passively promotes relaxation.

Due to the increasing quality and availability of low-cost EEG systems they can be applied in such a scenario to measure the current level of relaxation. Therefore, we introduce new algorithms for real-time mental state recognition based on an entropy-based signal processing approach.

Objective health data about subjects outside of the laboratory is important in order to analyse symptoms that cannot be reproduced in the laboratory. A simple daily life example would be how stride length changes with tiredness or stress. In order to investigate this we must be able to accurately segment a stride from daily living data in order to have an accurate measure of duration and distance. State-of-the-art methods use separate segmentation and classification approaches. This is inaccurate for segmentation of an isolated activity, especially one that is not repeated. This could be solved using a model that is based on the sequence of phases within activities. Such a model is a graphical model. Currently we are working with Conditional Random Fields and Hierarchical Hidden Markov Models on daily living data. The applications will include sports as well as daily living activities.

Digitalization of sports is also taking place in table tennis. This is caused by body-worn Wearables. This project captures, analyzes and processes the motion and movement of players,  ball characteristics and other interesting parameters during table tennis games and exercises. In contrast to common standard camera-based analysis within predefined laboratory environments used for most ball sports, only sensors mounted on the racket are included. All necessary electronics should ideally be hidden inside the racket, that the user does not feel affected. For motion analysis, mainly inertial sensors (accelerometers and gyroscopes) are used, as well as magnetometers for the absolute alignment in space and specific piezoelectric sensors for vibration detection. First, the acceleration, the angular velocity and absolute orientation of the racket are measured to classify the stroke type using pattern recognition algorithms.  It is possible to differentiate between forehand and backhand stroke types and various spin types. In addition, the ball impact event is verified by the vibration sensors. Afterwards, the resulting ball speed and spin are estimated shortly after this impact. Finally, the point of impact on the racket is localized by triangulation methods, similar to epicenter localization during earthquakes. All data is calculated on the embedded microcontroller and transferred to a mobile device, such as an Android smartphone via Bluetooth. There, the data is provided to the player as feedback for training support or statistics.

Die Europäische Metropolregion Nürnberg (EMN) bietet eine weltweit einzigartige Konstellation für die industrienahe Forschung im Bereich eingebetteter Systeme für Sport- und Fitnessanwendungen. Das Lab Fitness@ESI wird breit aufgestellt. Es gilt sowohl Themen des Spitzen- und Breitensports zu bedienen, als auch Wege aufzuzeigen, zentrale Themen des stetig an Bedeutung gewinnenden Marktes für Sportmöglichkeiten behinderter Menschen zu erleichtern. Weiterhin werden Bereiche von der persönlichen Fitness bis hin zu Altersgerechten Anwendungen thematisiert. Damit können wichtige Märkte wie der für Imagegewinn und als Vorreiter wichtige Spitzensport und der finanziell attraktive Team- und Freizeitsport adressiert werden. Der Schwerpunkt soll dabei auf elektronischen Produkten im Fitness-Sektor (ähnlich zu bereits existierenden Produktfamilien wie Fitbit, iHealth, miCoach oder Nike+), Web-Plattformen und Anwendungsbeispielen liegen. Eine große Rolle werden in Zukunft auch Produkte für individualisierte Ratschläge und Trainingsempfehlungen spielen. Gerade der Bereich der personalisierten Fitness-Produkte zur Selbstkontrolle und -überwachung der eigenen Gesundheit und Fitness im Rahmen der „Quantified Self“-Bewegung hat ein enormes Marktpotenzial. Auch eine Erweiterung in den Bereichen Ambient Assisted Living (AAL) und Mobile Health (mHealth) ist denkbar.

Die Kompetenzen der Partner sind ideal komplementär, um wichtige Themen des Labs Fitness@ESI angehen zu können. Mit der Kombination der Expertisen der FAU und des Fraunhofer IIS können die bereits jetzt vorliegenden Fragestellungen aus der Industrie angegangen werden. Diese beschäftigen sich u. a. mit Sensorintegration und Datenauswertung, um wichtige Events im Team- und Freizeitsport (Auswertung von Pässen oder Schüssen im Fußball für den Trainer, Vitalparameteranalyse von Läufern für den Breitensport) adressieren zu können. Des Weiteren sind Auswertungen von Positionsdaten von hohem Interesse für Vereine und Medien. Weitere technologische Inhalte des Projekts bewegen sich im Bereich der drahtlosen, körpernahen Sensornetzwerke. Wichtige Forschungsbereiche sind die sensorische Umgebungserfassung und -analyse, Energy Harvesting (Energiegewinnung) für batterie- und kabellose Energieversorgung, drahtlose Energieübertragung, Indoor- Lokalisierung und das Design von neuartigen, miniaturisierten Sensorknoten auf Niedrigenergiebasis. Ein weiterer wichtiger Punkt ist die unauffällige Integration von Sensorik in Kleidung, Schuhen, Sportgeräten und der Umgebung des Anwenders. Mögliche Anwendungen sind die automatische Erkennung sportspezifischer Bewegungen, automatisierte Trainingsunterstützung und Feedback-Anwendungen.

Aus Sicht der Softwareentwicklung liegen die Herausforderungen unter anderem in der Echtzeitverarbeitung und Fusion von Sensorsignalen, der effizienten drahtlosen übertragung und dem gleichzeitigen ressourcenschonenden Umgang mit Energie und Speicher.

Die nötigen Tests neu zu etablierender Systeme an realem Klientel (Elite- und Amateursportvereine, Freizeitsportler, aber auch ältere, sowie seh- oder gehbehinderte Menschen) sind über die etablierten sehr guten Kontakte zu Industrieunternehmen wie der adidas AG, dem Institut für Sportwissenschaft und Sport der FAU und dem Universitätsklinikum Erlangen sichergestellt.

DailyHeart is a system for vertigo assessment during daily-life activities. It addresses the whole pipeline from data acquisition (via unobtrusive body-worn sensors) over data processing to data visualization and potential feedback (via Android application).

The main idea is to assess the autonomous nervous system (ANS) during the orthostatic reaction, characterized by an increase in heart rate after posture transitions, e.g. from lying or sitting to standing. Whereas the ANS of healthy subjects can easily adapt to these posture changes, people suffering from ANS disorders (e.g. Parkinson’s Disease) experience a considerably heart rate response and a decrease in blood pressure, potentially leading to vertigo and syncope.

Clinical results have shown that multiple measurements throughout the day can already indicate a possible disorder of the ANS. For that reason, DailyHeart aims to transfer the clinical assessment into the home environment by detecting posture changes during daily-life activities and thus trigger an HRV and BP analysis for effectively assessing vertigo and thus provide a recommendation of consulting a medical expert.

The aim of the research project is the development of an embedded system for the determination of the human hydration level with non-invasive sensors. The combination of non-invasive sensors and embedded systems enables the determination of the hydration level in many new situations. For example, the hydration level of athletes could be monitored to maintain their optimal performance or the system could be connected to clinical warning systems in hospitals to monitor the hydration level of patients.

Body Sensor Networks are getting more and more important in sports and health. Currently, various isolated applications exist that use body sensors to assist athletes and monitor elderly people. Systems like the adidas miCoach and Nike+ prove the potential of information and communication engineering technology for manufacturers of sports equipment. The perfect product for a leading position in this market would be a central, flexible and generic wearable computing platform instead of isolated applications. To facilitate such a solution, sensors integration in clothing and sports equipment and data analysis capabilities have to be substantially advanced. Additionally, to succeed on the market, new communication and sensor technologies as well as innovative applications have to be developed.

The goal of the project is to bundle and enhance the expertise of the project partners in the described field to develop innovative products. The existing miCoach platform will be the basis for a comprehensive communication and application platform for body sensor network data called "miLife". This platform will provide flexible sensor connection, data analysis and social networking capabilities for applications in team sports, exercise motivation and health monitoring.

RoboCup is a international initiative to promote research in artificial intelligence and autonomous mobile robots. Each year the RoboCup Foundation hosts international tournaments where top research groups of Universities from the whole world participate. Since 2008, the University of Erlangen-Nuremberg also has its own RoboCup team that participates in the small-size league. This league is one of the smallest and fastest RoboCup leagues. Five wheeled robots per team are playing on a field of about 6m x 4m. The maximum size for each robot is 18cm in diameter and a height of 15 cm. The robots get information about the current game situation from two cameras above the field and an external computer, which communicates with the robots via a wireless link. In Erlangen the team is organized as an interdisciplinary student project at the Technical Department. The main goals of this project are to foster creative ideas and team work among technical students from electrical engineering, mechatronics and computer science. Research topics include topics from pattern recognition, embedded systems and artificial intelligence. In the scope of this project the Pattern Recognition Lab employs statistical estimation techniques and tries to extend them towards automotive applications. To promote the project, a non-profit organisation called "Robotics Erlangen e.V." was founded in 2008. In this organization team members as well as supporters of the group are brought together. The project is funded in part by tuition fees as well as private and industry donations.