The aim is an in-depth study of advanced techniques in signal processing, analysis and modelling, applied to a variety of medical diagnostic problems. The course does not deal with details of instrumental techniques but focuses on solving problems arising in the analysis and processing of biomedical data after acquisition. Attention is focused on advanced signal processing topics: subspace based signal processing, blind source separation, nonlinear signal analysis, source localization and estimation, as well as on multimodal data processing where heterogeneous sources of information, such as EEG and fMRI or EMG and ECG, etc., are combined.
Through the development of algorithms to solve problems in biomedical signal processing and the application of these algorithms in many case studies, the students learn how to apply these techniques in many practical biomedical problems thereby improving medical diagnosis and online patient monitoring
At the end of the course, the student should have acquired the following skills:
- a sufficient level of expertise in the following advanced signal processing topics: subspace based signal processing, blind source separation, nonlinear signal analysis, source localization and estimation and multimodal signal processing,
- Ability to correctly apply the taught methods to biomedical problems and critically evaluate their performance.
A thorough knowledge in signals and systems is required, including basics in statistics (normality, bias, variance, maximum likelihood, distributions), filtering, system theory, signal transformations (Fourier transform, wavelets), linear algebra (vector spaces, orthogonality, eigenvalue decomposition, least squares), biomedical engineering, Matlab programming.
K.U.Leuven students must have earned credits for the following basic courses
- H01A4A: Applied Algebra
- H01F7A: System theory
- H01L6A: Digital Signal Processing, part 1
or comparable courses.
In addition, the student must have earned credits for the core course
- H03I2A: Biomedical data Processing
- and/orH05F3A: Digital Signal Processing, part 2.
In the latter case, the student is required to obtain basic knowledge in biomedical signals through self-education of chapters 1-2 of the book: Rangaraj M. Rangayyan, BIOMEDICAL SIGNAL ANALYSIS: A Case-Study Approach'', John Wiley & Sons, I nc., New York, 2002.
Articles and literature
Slides, transparencies, courseware
Toledo / e-platform
Is also included in other courses
Information on the Lectures.
In biomedical data processing, the aim is to extract clinically, biochemically or pharmaceutically relevant information (e.g metabolite concentrations in the brain) in terms of parameters out of low-quality measurements in order to enable an improved medical diagnosis. Typically, biomedical data are affected by large measurement errors, largely due to the noninvasive nature of the measurement process or the severe constraints to keep the input signal as low as possible for safety and bio-ethical reasons. Accurate and automated quantification of this information requires an ingenious combination of the following 4 issues:
1. an adequate pretreatment of the data
2. the design of an appropriate model and model validation
3. a fast and numerically robust model parameter quantification method,
4. an extensive evaluation and performance study, using in-vivo and patient data, up to the embedding of the advanced tools into userfriendly user interfaces to be used by clinicians.
Signal preprocessing implies the use of advanced signal processing tools for achieving the best signal separation (noise elimination, filtering out the relevant information such as one metabolite concentration), such as the Singular Value Decomposition (SVD) (which decomposes a matrix into an orthogonal matrix times a diagonal matrix containing the singular values times another orthogonal matrix) and its variants Principal Component Analysis (PCA) and Canonical Correlation Analysis (CCA), Total Least Squares (TLS) (which provides consistent estimators of the parameters in a linear model in the presence of errors on all data) , tensor decompositions (for extracting the source signals using Independent Component Analysis (ICA) and variants), multichannel adaptive filtering, and subspace algorithms. The next issue of modeling and model validation involves the use of dipole modelling techniques for source localization (e.g. to detect focal activity) or of so-called grey-box model fitting techniques which fully exploit the available prior knowledge (biochemical, patient related, expert based), such as a combination of constrained optimisation with subspace based methods. Model parameter quantification strikes the right compromise between computational speed and quality of the computed estimates (numerical robustness). Finally, the clinical performance needs to be evaluated, using patient data, and implemented in a userfriendly way for use in a clinical environment. During the course, special attention is given to the design of improved models and the development of advanced algorithms, as mentioned above, for processing multichannel biomedical data, possibly acquired using various modalities (e.g. EEG and fMRI).
Continuing on the course ``H03I2A Biomedical Data Processing'', the following advanced topics are discussed in 14 two-hour lectures:
- Introduction (1x2h): including an overview of basic data processing tools such as Singular Value decomposition, Canonical Correlation analysis, principal component analysis, etc.
- Subspace based signal processing (3x2h): including state space identification and harmonic retrieval. SVD based algorithms (matrix pencil, ESPRIT), subspace tracking, principal and minor component analysis for signal-noise separation are described, as well as total least squares techniques for measurement error modelling.
- Blind source separation (3x2h): including principles of ICA, higher order statistics, multilinear algebra and multiway analysis. Methods based on PCA, ICA, tensor decompositions (PARAFAC, Tucker-3 model) are discussed, as well as methods based on non-Gaussianity or spectral colour.
- Nonlinear signal analysis (3x2h): including principles of fractal and chaos theory. Basics of fractal geometry, self-similarity, multifractal analysis, power-law type behavior, chaotic time series analysis (time delay embedding, Lyapunov exponent) and multiscale analysis are discussed.
- Source localization and estimation (3x2h): including the forward and inverse problems and methods for solving these using various source configurations.
- Multimodal signal processing (1x2h): including an introduction to techniques for fusing heterogeneous sources of biomedical data such as EEG and fMRI, EMG ECG, and respiration, polysomnography, etc.
Description of learning activities
14 lectures of 2 hours
slides, selection of articles from literature, chapters from the book: J. Gao, Y. Cao, W.-w. Tung, J. Hu, ``Multiscale analysis of complex time series: Integration of chaos and random fractal theory, and beyond'', J. Wiley & Sons, Inc., New York, 2007.
This study activity consists of 4 exercise sessions of 2,5 hours. These sessions offer practical experience with life-like signals and are essential to understand and
appreciate the theory. These sessions consist of solving problems on the computer: either individually or in groups of 2 students. The student is asked either to apply the offered Matlab programmes on biomedical signals and to analyze the results, or to further expand or develop own Matlab programmes. A report of each computer exercise has to be handed in: 10 days before the examination at the latest.
Description of learning activities
4 exercise sessions in Matlab on computer.
slides, selected articles, Matlab software, datasets
The student hands in a report of each exercise session (4 in total), 10 days before the examination at the latest. The evaluation is based on an oral defense of these reports during the examination session.