Roman Rosipal UM-SAS SAS

Home

Research

Papers

CV

Data and Codes

rrLab
Research interests
  • Research in the area of applied statistics, machine learning, computational and cognitive neuroscience
  • Multivariate data analysis; latent variable regression, classification and dimensionality reduction methods;
    Dynamic Bayesian Networks for data fusion; nonlinear kernel learning and support vector machines
  • Electrophysiological data analysis (EEG, EOG, EMG); event-related potentials (ERPs); sleep process modelling;
    study of cognitive fatigue; brain-computer interfaces (BCIs); vigilance, drowsiness and fatigue monitoring
Current Research projects
  • Enhancing cognition and motor rehabilitation using mixed reality (TInVR) (2022-2026).
    We aim to study specific forms of social interaction using state-of-the-art technology - virtual reality (VR) which is motivated by its known benefits. The project has two main parts, human–robot interaction (HRI) and therapist– patient interaction (TPI). The interactions are enabled using head-mounted displays and controllers allowing the human to act in VR. We propose two research avenues going beyond the state-of-the-art in respective contexts. In HRI, we will develop scenarios allowing the humanoid robot to learn, understand and imitate human motor actions using flexible feedback. Next, we develop scenarios for testing and validating human trust in robot behavior based on multimodal signals. We will also investigate physical interaction with a humanoid robot NICO. In TPI with stroke patients, we develop a series of VR-based occupational therapy procedures for motor and cognitive impairment neurorehabilitation using an active and passive brain-computer interface, and we will validate these procedures. We expect observations from HRI experiments to be exploited in TPI. The proposed project is highly multidisciplinary, combining knowledge and research methods from psychology, social cognition, robotics, machine learning and neuroscience. We expect to identify features and mechanisms leading to trustworthy processes with a human in the loop, as a precondition of success, be it a collaborative task or treatment in VR.
  • Towards an ecologically valid symbiosis of BCI and head-mounted VR displays: focus on collaborative post-stroke neurorehabilitation (ReHaB) (2022-2025).
    A growing body of evidence suggests that integrated technologies of brain-computer interfaces (BCI) and virtual reality (VR) environments provide a flexible platform for a series of neurorehabilitation therapies, including significant post-stroke motor recovery and cognitive-behavioural therapy. When immersed in such an environment, the subject's perceptual level of social interaction is often impaired due to the sub-optimal quality of the interface lacking the social aspect of human interactions. We propose a user-friendly wearable low-power smart BCI system with an ecologically valid VR environment in which both the patient and therapist collaboratively interact via their person-specific avatar representations. On the one hand, the patient voluntarily, and in a self-paced manner, manages their activity in the environment and interacts with the therapist via a BCI-driven mental imagery process. This process is computed and rendered in real-time on an energy efficient wearable device. On the other hand, the therapist's unlimited motor and communication skills allow him to fully control the environment. Thus, the VR environment may be flexibly modified by the therapist allowing for different occupational therapy scenarios to be created and selected following the patient's recovery needs, mental states, and instantaneous responses. Careful attention will be paid to balance known neurophysiological evidence of the process with artificial intelligence (AI) within the active BCI protocols to avoid running into conceptual pitfalls. Computed features of EEG signals will serve to monitor the patient's engagement, cognitive workload, or mental fatigue in real-time. These indicators will be combined with observable patient’s performance and behaviours to improve the accuracy of mental state estimation. Exceeding critical mental state levels will signal the therapist to activate appropriate countermeasures in the form of environmental and task changes. To challenge and overcome existing technologies, commercially available head-mounted VR displays (HMD) combined with miniaturized energy- efficient microcontroller units will be employed for EEG signal processing, BCI discrimination and on-board classification implementation, and a full-duplex communication with the HMD controllers. Advanced dry EEG sensors suitable to operate and be placed on the scalp without interfering with the HMD will be developed and tested. A novel patient-to-therapist multimodal collaborative environment augmented through VR immersion and by AI monitored patient’s brain activity will be created. By combining these pieces, a low-power wearable BCI-HMD system will be constructed. A series of clinical studies will validate the system.
  • Smart deep brain stimulation as a treatment strategy in treatment-resistant depression (VEGA1) (2022-2025).
    Impaired connectivity between different brain areas underlines pathophysiology of multiple brain disorders. It is possible that impaired connectivity between prefrontal cortex and ventral pallidum is involved in depression. Smart deep brain simulation, combining real-time detection of the neuronal activity in in prefrontal cortex with the stimulation of ventral tegmental area, might be thus an effective in depression. We aim to examine the cortico-tegmental connectivity and to test the antidepressant-like effectiveness of the smart deep brain stimulation in an animal model of depression.
  • Causal analysis of measured signals and time series (VEGA2) (2022-2025).
    The project is focused on the causal analysis of measured time series and signals. It builds on previous results of the project regarding the generalization of the Granger test and proposals for new tests in the reconstructed state spaces. The aim is to develop new methods and algorithms for bivariate and multidimensional causal analysis. We will understand the investigated time series and signals as one-dimensional manifestations of more complex systems or subsystems. We will extend the detection of causality between the two systems to multivariate cases - dynamic networks with nodes characterized by time series. Such complex networks are widespread in the real world. Biomedical applications are among the best known. Brain activity, determined by multichannel electroencephalographic signals, is a crucial example. We will show that causality research is currently at a stage that allows for ambitious goals in studying effective connectivity (i.e., directed interactions, not structural or functional interconnections) in the brain.
Recent Research projects
  • Enhancing cognition and motor rehabilitation using mixed reality (ECoReMiR) (2017-2021).
    Technological advancements based on mixed reality (MR) offer various challenges for research and medical treatment. The project focuses on two objectives related to healthy subjects and hemiparetic patients after stroke. First, we will test the hypothesis whether cognitive training using appropriately designed MR environment will enhance perceptual and cognitive performance in healthy subjects. This will be tested by computerized psychological experiments as well as by measuring event-related potentials or ERPs. Second, we will test the hypothesis whether experience with training in MR (in combination with motor-imagery based brain-computer interface developed by us) will enhance oscillatory sensory-motor rhythms. This will be tested by measuring subject's EEG activity before and after each training session, clinical testing, as well as by the questionnaires aiming to learn about human factors including mental fatigue, motivation, irritation or sleepiness due to training. In both objectives, we will design and implement a set of testing procedures, carry out a battery of dedicated experiments, and critically evaluate the results with the goal to validate MR designs.
  • Effects of sleep disturbances on day-time neurocognitive performance in patients with stroke (SleepCog) (2013-2016).
    Sleep deprivation, whether from disorder or lifestyle, whether acute or chronic, poses a significant risk in day-time cognitive performance, excessive somnolence, impaired attention or decreased level of motor abilities. Sleep deprivation is closely related to sleep fragmentation often associated with short several second long arousals. Although limited studies of partial sleep restriction and sleep fragmentation have revealed important sleep indices leading to cognitive deficits, a challenging question how a typical, good quality, structure of sleep should look like remains open. To improve these results the project will investigate and evaluate a novel probabilistic sleep model. In a preliminary series of tests on healthy subjects it has been shown that the model contains significantly more objective information about external measures of the sleep quality than the traditional sleep staging. Patients with specific cerebral lesions will be studied in the project. It is well-known that targeted patients are strongly vulnerable to sleep disturbances that often lead to deficits in their day-time cognitive and attentional performance. Experimental studies with such patients and with focus on relating their sleep patterns and disturbances with a day-time performance are limited and so far have not been carried out in Slovakia. It is expected that the project will deliver not only new academic research results but also important clinical knowledge.
  • Brain-computer interface with robot-assisted training for rehabilitation (BCI-RAS) (2013-2017).
    We will apply advanced tools and methods of applied informatics for the design and development of an intelligent system allowing the users to go through the process of self-controlled training of impaired motor pathways. We will combine the brain-computer interface (BCI) technology with a robotic arm system into a compact system that can be used as a robot-assisted neurorehabilitation tool. The BCI directly uses the signal of the brain electrical activity to allow users to operate the environment without any muscular activation. However, several critical issues need to be addressed before using BCI in neurorehabilitation, namely, issues ranging from signal acquisition and selection of the proper BCI paradigm to the evaluation of the affective state, cognitive load and system acceptability of the users. The project will address these issues by using new signal processing and machine learning algorithms, training protocols and intelligent methods for the users' physiological state changes detection and monitoring, recently developed by the researchers participating in the project. Novel BCI training protocols, including the neurofeedback training based on multi-way analysis of EEG data, will be used and validated. Finally, the system will be tested in clinical practice on selected patients with motor impairments caused by stroke, as well as on healthy volunteers.
  • The Neurosensory Optimization of Information Transfer (NOIT) (2011-2014).
    The NOIT project aims to show that we can counteract the effects of sleep deprivation and fatigue on cognition with the aid of an automated EEG biofeedback (EBF) system that enables us to continually manage processes in our left and right brain hemispheres.
Research developments
  • APECS stands for Advanced Physiological Estimation of Cognitive Status, a system of signal processing and machine learning algorithms for mental state estimation. APECSgui is a MATLAB toolbox that allows non-experts to build EEG-based models for mental state estimation.
  • EEtrac is a computer system for simultaneous real-time processing of EEG and Eye-tracking signals. It allows a laptop computer to display information which is dually contingent on the mental state and gaze of the user.
  • PSM stands for Probabilistic Sleep Model, a model of the sleep process with a higher temporal and spatial resolution based on information fused from a set of different sensors, The model is based on so-called Gaussian mixture models that cluster spectral characteristics of the signals into a number of sleep states with a high temporal resolution and without a prior definition of how many and which states are reached during a night of sleep. Thus, the model frees itself from some of the limits of classical sleep signal analysis, namely that stages are defined by what an expert can identify visually in the signal and by the arbitrary rough division into 30 second pieces, historically still rooted in the use of paper EEG. The proof that this new way of describing sleep correlates better with how a patient feels and performs in the morning points to the clinical validity of the approach which could lead to new ways of analyzing sleep in medicine and beyond.