Datum / čas
Date(s) - 24.10.
14:00 - 16:00
Kategorie ne Kategorie
Elizaveta Saifutdinova „Advanced Analysis of Brain Activity“
Analysis of recorded brain activity is one of the main investigation methods in modern sleep medicine and research. Long duration and complex nature of the data make it difficult for manual investigation. Moreover, high inter-subject variability could cause problems in automatic processing. In our projects, we focus on the problem of automatic EEG pattern detection. We concentrate on artifacts and sleep spindles as two typical patterns in sleep EEG. We review the methodologies and strategies used in real sleep research practice. Moreover, we investigate and test the performance of state-of-the-art approaches for the tasks. We propose enhancement methodologies and use expert’s strategies for automatic method development. The proposed methods utilize recent advances in EEG pattern detection. They are adaptive and fully unsupervised. Testing is performed on the data collected from subjects suffering from a sleep disorder which increases inter-subject variability of the data. We analyze obtained results in aspects of formal statistical measures and using visual inspection of the data providing more details about data nature.
Elizaveta Saifutdinova is a PhD student of the CogSys dept. She has focused on advanced methods for detection of specific patterns in long-term EEG recordings. She is cooperating tightly with the National Institute of Mental Health.
George Manis „Bubble Entropy: An Entropy Definition in m-Dimensional Space with Minimal Dependence on Parameter Estimation“
Computational methods based on entropy estimation of a physiological system are today the most popular and promising non-liner methods for Heart Rate Variability (HRV) analysis. Entropy estimation in m-dimensional space is used in most HRV research articles and is also introduced gradually in clinical practice. The two most popular of those methods are certainly Approximate Entropy and Sample Entropy.
Both Approximate and Sample Entropy requires the estimation of two parameters: the embedding space m and the threshold distance r. The selection of these two parameters is always an issue and should be given attention. It affects the classification capability and may also be dependent even on the data set. In practice, we overcome this problem by setting standard values for m and r (i.e., m=2, r=0.2), practically accepting that the problem is difficult to be solved. Reducing the significance of the parameters or even eliminating them results in decoupling the method from the data and the application and also removes subjective factors.
Bubble Entropy has been introduced to alleviate this problem. The target is to propose an entropy definition in which there will be no r parameter involved. This parameter is the most difficult one to be estimated, since m is an integer number ranging only in a small set of meaningful values. Experimental results with both real and synthetic HRV signals showed that Bubble Entropy presents remarkable stability and exhibits increased descriptive and discriminating power compared to other definitions of entropy, including the most popular ones.
George Manis is an Associate Professor of Biomedical Engineering in the School of Engineering, Dept. of Computer Science and Engineering, University of Ioannina, Greece. His research interests include Biomedical Engineering and Computing Systems. He received his PhD from the National Technical University of Athens (1997), his MSc from Queen Mary University in London (1993) and his Diploma in Electrical and Computer Engineering (5-years degree) from the National Technical University of Athens (1992). He is affiliated with the University of Ioannina since 2000, while he has also served in University of Crete (1 year), University of Patras (1 year) and the Hellenic Open University (20 years as adjacent stuff). He has been a visiting researcher in University of Amsterdam (2006), Politechnico di Milano (2009) and Universita degli Studi di Milano (2015). He has published 1 book, supervised the Greek edition of 2 more and has published more that 100 articles in scientific journals and conferences.