An overview of the research topics that are investigated at the Competence Center Biomedical Data Science can be found below.
SIGNAL AND INFORMATION PROCESSING
Usually, biomedical data cannot be interpreted right away. Instead, signal processing methods are required to extract meaningful and interpretable information. The spectrum of signal and information processing techniques is broad: There are linear and non-linear methods, methods in the time domain as well as in the spectral domain, recursive as well as non-recursive algorithms and continuous as well as discrete methods.
MULTIMODAL DATA INTEGRATION
Usually, there are various techniques to investigate a scientific phenomenon. Most of the time, the best results are obtained if several techniques are combined. In the case of neuroscience, for example, EEG data can be combined with fMRI data to obtain a high temporal as well as a high spatial resolution. When integrating data of distinct modalities, several issues must be taken into account. For instance, the data needs to be co-registered with respect to time and space. In addition, there are manifold ways to merge and integrate multimodal data. The development of such methods is one of the main research topics of the center.
COMPLEX NETWORKS IN BIOMEDICAL SYSTEMS
The complex networks theory is a universal approach that can be applied to different fields starting from social sciences to brain research. The center uses such methods for the analysis of functional and structural neural networks. Furthermore, it successfully applied graph theory methods to model the interactions and movements of individuals or groups in urban environments.
In neurosciences, one major goal is to explain mechanisms of cortical information transmission and storage. These include activity patterns in neural networks which often are associated with oscillations, waves and synchronization. Previous studies have shown that synaptic plasticity is one the main concepts for imitating human information processing. The approach has been formalized into the modified Stochastic Synaptic Model (MSSM) which is successfully evaluated. Practical application was realized with parallelized implementation using OpenCL and FPGAs. The concept has also been successfully utilized in industrial applications.
Due to the development of wearables as well the minituarisation of IT and communication devices, mobile health applications have become part of our daily life. Still, the acquisition of psycho-physiological and biological data outside of the lab poses several new challenges to data analysis methods. Moreover, new infrastructure solutions are required for such continuous long-term recordings. The competence center develops new methods, algorithms and software to make mHealth technologies available for diagnostic applications, therapy, training, prevention and self-management.
Brain connectivity is a neuroscientific research field that focuses on the structural and functional connections within the brain. Structural connectivity can be investigated with diffusion-weighted MRI. The data obtained with this neuroimaging technique provides insight into the structural neuronal wiring, nerve fiber bundles and crossing fiber pathways. The term functional connectivity describes several distinct kinds of connectivity. For instance, brain regions that are simultaneously more active than others during a certain task are believed to be functionally connected. Similarly, brain regions whose neural activity is synchronized are expected to exchange information. This broad definition of functional connectivity has led to an enormous amount of connectivity measures. The competence center applies various of them and has even developed some additional measures.
DEVELOPMENT OF NEW METHODS AND ALGORITHMS
Over the last hundred years, neuroscientific acquisition techniques have improved tremendously. By now, even several mobile EEG devices and bidirectional brain-computer interfaces are available. New acquisition techniques usually lead to new kinds of data that require the development of innovative processing methods and algorithms. At the same time, new methods of analysis can help to extract additional information from former data. Thus, the development of new methods and algorithms is crucial for scientific progress. In neuroscience, particular attention is paid to online methods that enable an immediate signal analysis and provide an instant feedback for the user. Usually, methods that are developed within the neuroscientific field are universal and find various applications in other disciplines like physics, biology or geology.
BDSS, MV, NIS
As the name suggest, the discipline medical visualization focuses on the comprehensive and complete visualization of all kinds of biological and medical data. With respect to neuroimaging, the visual integration of several image modalities, the interactive visual exploration of image data, the extraction of relevant nerve fiber bundles from diffusion-weighted data and the effective use of modern graphics hardware for an effective visual presentation of complex three-dimensional phenomena within the brain is of particular interest. At the competence center, this work is carried out in close cooperation with neuroscientists or clinical physicians to ensure a high usability.
Modern imaging methods allow resolutions below 1mm and, thus, provide detailed structural information on the human brain. To evaluate this information, a huge amount of calculations and algorithms are needed. One sub class of these algorithms is the class of tractography algorithms. They allow finding global structures as well as tracking and bundling them. Another very important part of the evaluation chain is the visualization of these data. In a project of the research group Medical Visualization, existing methods are analyzed, compared and performance-optimized on a GPU-basis. Furthermore, the project aims to develop new clustering algorithms and to optimize the visualization of tractography data. The tight cooperation between neuroscientists of the Max-Planck-Institutes for Cognitive and Brain Sciences and Neurological Research provides a tight coupling to up-to-date research in this area and ensures demand-oriented results.
MEACHINE LEARNING & ARTIFICIAL INTELLIGENCE
Artificial Intelligence (AI) is a huge field of research dealing with diverse facets of acceleration and improvement of decision processes by means of automation. Advanced AI-problems can be formulated in terms of Machine Learning (ML). The task of ML is, to automatically assign chosen properties to any sort of data. The scientific research in this field aims to provide suitable strategies to find an input-output-function accomplishing such a task sufficiently well. The pathway to finding such a function in a given case is often divided into three main steps, each of them representing a huge research area in itself: feature extraction, feature selection and classification/property assignment.
Brain-computer interfaces (or brain-machine interfaces) are utilized in the treatment of various neurological diseases. For instance, BCIs are used to promote the motor rehabilitation of stroke patients. In addition, they are applied to substitute functions that the user is no longer able to perform without help. Thus, severely paralyzed persons such as ALS patients utilize BCIs for communication or to control devices like robotic arms, wheelchairs, orthoses, prostheses, virtual typewriters or web browsers. Using stimulating electrodes, even a bidirectional communication between the brain and the computer can be achieved (bidirectional cortical communication interface, BCCI). To minimize both the training time and the training effort of the patient, machine learning techniques may be applied.
ANALYSIS OF MOBILE DATA
Mobile sensor devices become more and more important in research and medical applications. For the first time in history, researchers get the opportunity to record biomedical data outside of the lab under real-life conditions. Usually, even long term recordings of several hours or days are possible. However, this kind of mobile data is not easy to analyze. For instance, the lack of specific experimental conditions implies that blind source separation methods need to be applied that do not depend on the existence of external time markers. If the mobile data is transferred via Bluetooth (which is the case for the majority of the mobile sensor devices), researchers also have to deal with Bluetooth delays, Bluetooth buffers and connectivity problems. In addition, motion artefacts are to be expected. To still obtain data of sufficient quality, specific preprocessing techniques need to be developed and applied.
Neuroimaging is a neuroscientific discipline that aims to image the structure and function of the central nervous system. Popular neuroimaging techniques are for example magnetic resonance imaging, functional magnetic resonance imaging, computer tomography, magnetoencephalography, positron emission tomography or diffusion weighted imaging. In medical practice, neuroimaging techniques are usually applied for diagnostic purposes. For instance, they are utilized to detect intracranial or metabolic diseases. In addition, neuroimaging techniques are widely used in research applications. Neuroimaging data supports, for example, the construction of brain-computer-interfaces, the understanding of neurological disorders and brain mapping.
The competence center designs and performs its own psychological and neuroscientific experiments. This could be single experiments or complex multimodal studies. Usually, studies with patient are carried out in collaboration with cooperating partners from medicine or psychology. Furthermore, members of the center were already involved in studies in collaboration with robotics or neuroprosthetics research groups.
DIGITALIZATION IN HEALTHCARE
As in all areas of life, digitalization also becomes more and more important in healthcare and medicine. In 2015, the electronic health card was introduced in Germany to facilitate billing, administration and access to healthcare. Prospectively, the medical infrastructure all over the country will be changed from a paper-based system to storage of all medical data of a patient in an electronic patient file. This provides enormous opportunities for applications in digital infrastructure and patient-centered medicine.
BDSS, MV, NIS
The research field eHealth covers all electronic processes in healthcare and medicine. It ranges from electronic applications to management of patient data, prescriptions or lab results to clinical decision support systems, and telemedicine services. Furthermore, a promising branch is given by mHealth technologies that use mobile devices to collect patient data, provide healthcare information or directly support treatment and therapy. This research field is constantly growing with increasing technologization, digitalization, and miniaturization.
BDSS, MV, NIS