Required in stage
First term
Second term
Both terms
Not organised
This year
Next year
Alternating years
External
Prerequisites
Language of instruction
Duration
Identical coursesSelected Topics in Biomedical Signal Processing (B-KUL-H06W1A)
Bertrand Alexander (coordinator) | Bertrand Alexander | De Lathauwer Lieven | De Vos Maarten Aims
This course consists of an in-depth study of advanced techniques for processing, analyzing and modelling of biomedical signals with applications in medical diagnosis and biomedical research. The course does not deal with details of instrumental techniques but focuses on algorithmic tools for processing biomedical signals after acquisition.
The course introduces generic signal processing and machine learning techniques/theory from the following fields:
- subspace based signal processing and data-driven multi-channel filter design
- tensor methods and blind source separation
- non-linear signal analysis and non-linear classification (incl. gaussian processes and deep learning) for time series data
and demonstrates how these are applied in the analysis of various biomedical signal modalities (incl. EEG, MEG, EMG, neural probes, ECG, NMR, polysomnography, etc.). Through computer projects, the students learn how to apply these techniques on actual biomedical signals.
At the end of the course, the student should have acquired the following skills:
- Being familiar with (and having insight in) the typical problems arising in the analysis and interpretation of biomedical signals
- A sufficient level of understanding of the techniques and theory covered in the lectures and how they are applied in biomedical contexts.
- The ability to correctly program these methods in Matlab/Python, apply them to biomedical signals, and critically evaluate their performance.
Previous knowledge
The students should have basic knowledge in statistics, linear algebra, systems theory, signal processing, stochastic processes, machine learning, and Matlab/Python programming. More specifically:
- for STATISTICS: second-order statistics (normality, variance, correlation) and distributions (Gaussian, etc.), both for univariate and multivariate distributions, principal component analysis (PCA), etc.
- for LINEAR ALGEBRA: matrix computations, vector spaces, orthogonality, eigenvalue and singular value decomposition, least squares theory, etc.
- for SYSTEMS THEORY and SIGNAL PROCESSING: Concepts such as transfer function, convolution, FIR filtering, the Discrete Fourier transform, short-time Fourier transform, adaptive filters, etc.
- for STOCHASTIC PROCESSES: Concepts such as stationarity, power spectral density (PSD), ergodicity, Wiener filtering, etc.
- for MACHINE LEARNING: basic classifiers (LDA, artificial neural networks)
- MATLAB programming: basic knowledge
- PYTHON programming: basic knowledge
(Students that do not have any experience in Python or Matlab can still follow the course, but they should take an introductory Python/Matlab tutorial at home. The preparation time to get familiar with Python/Matlab is NOT counted in the course load. If a student is not familiar with Python, it is allowed to make the Python projects in Matlab, but then no or less support will be available).
KU Leuven students who have earned credits for all of the following basic courses (or comparable courses) should have seen most of the required theoretical knowledge:
- H01A4A or I0D38B: Linear Algebra
- H01L6A and/or H05F3A: Digital Signal Processing and/or Digital Signal Processing for Communication and Information Systems, or comparable courses.
- H03I2A and/or H05I7A: Biomedical Data Processing and/or Stochastic Signal and System Analysis
- H03I2A and/or H09J2A: Biomedical Data Processing and/or Image Analysis and Understanding
Is included in these courses of study
- Master in de ingenieurswetenschappen: biomedische technologie (Leuven) 120 ects.
- Master of Bioinformatics (Leuven) (ENGINEERING) 120 ects.
- Master of Biomedical Engineering (Programme for students started before 2021-2022) (Leuven) 120 ects.
- Courses for Exchange Students Faculty of Engineering Science (Leuven)
- Master in de ingenieurswetenschappen: elektrotechniek (Leuven) (Informatiesystemen en signaalverwerking) 120 ects.
- Master of Electrical Engineering (Leuven) (Information Systems and Signal Processing) 120 ects.
- Master of Biomedical Engineering (Programme for students started in 2021-2022 or later) (Leuven) (Option: Biomedical Data Analytics) 120 ects.
Activities
4.83 ects. Selected Topics in Biomedical Signal Processing: Lectures (B-KUL-H06W1a)
Content
Information on the Lectures.
In biomedical signal processing, the aim is to extract clinically, biochemically or pharmaceutically relevant information out of (potentially low-quality) measurements in order to enable an improved medical diagnosis. Typically, the relevant information is obscured by large measurement artifacts and background noise from interfering physiological processes. Furthermore, physiological signals often exhibit a large variability (across time and across patients). Last, the recorded data can become too big to be analysed manually. Therefore, accurate and automated quantification of this information requires an ingenious combination of:
· an adequate pretreatment of the data,
· the design of an appropriate model and model validation,
· a fast and numerically robust model parameter quantification method, and
· an extensive evaluation study, using in-vivo and patient data, up to the embedding of the advanced algorithms into efficient tools to be used by clinicians.
To solve each of the above issues, special attention is given to the design of improved models and the development of advanced algorithms, as mentioned above, for processing multi-channel biomedical data, possibly acquired using various modalities (e.g. EEG and fMRI, EEG and sound, polysomnography).
The following advanced topics are discussed in 12 two-hour lectures, split up in 3 parts:
· Part 1: Data-driven multi-channel filter design (5x2h, incl. practical info on the course): including linear spatio-temporal MISO/SIMO/MIMO models, an introduction to linear estimation theory (ML, least squares, MMSE, BLUE, pre-whitening), data-driven spatio-temporal filter design (SNR-optimal filtering, CCA, CSP, MWF, and low-rank models thereof), beamforming, dimensionality reduction, and channel selection. Applications include (amongst others) brain-computer interfaces (auditory and motor), neural spike sorting, artifact removal in EEG, neural source localization, echography, wearable EEG, polysomnography
· Part 2: Tensor-based methods (3x2h): including tensor decompositions (CPD, MLSVD), multilinear PCA, blind source separation and ICA. Applications include (amongst others) harmonic retrieval (NMR), excitation-emission spectroscopy (amino acids), detection of epileptic seizure in EEG, fetal ECG extraction.
· Part 3: Non-linear signal analysis and deep learning for time series (4x2h): including an introduction to fractal and chaos theory. Self-similarity, multifractal analysis, power-law type behavior, detrended fluctuation analysis are discussed. Also advanced machine learning techniques including common deep learning and causal ML techniques for medical data and time series will be discussed. Applications include (amongst others) non-linear heart rate variability analysis, speech processing and automated sleep staging.
Course material
slides + online material on Toledo
Format: more information
12 lectures of 2 hours
1.17 ects. Selected Topics in Biomedical Signal Processing: Exercises (B-KUL-H06W2a)
Content
Computer sessions:
This study activity consists of 3 computer sessions of 2,5 hours (in Matlab and Python), during which a practical project is introduced. These projects offer practical experience with life-like signals and are essential to understand and appreciate the theory. The projects are to be solved either individually or in teams of 2 students during the introduction session and continued at home. The students are asked either to apply the offered Matlab/Python programmes on biomedical signals and to analyze the results, or to further expand on or develop own computer programmes. A paper + electronic version of the reports of each computer session has to be handed in before the start of the Christmas vacation (1 report per team, deadline will be communicated through Toledo). In addition, the corresponding computer code has to be uploaded on Toledo (same deadline).
Course material
Slides, Matlab/Python software, datasets, Tensorlab (www.tensorlab.net).
Format: more information
3 exercise sessions in Matlab/Python on computer, which introduce 3 project tasks to be finished at home
Evaluation
Evaluation: Selected Topics in Biomedical Signal Processing (B-KUL-H26W1a)
Explanation
COMPUTER SESSIONS: Before attending each of the three computer sessions, the student studies the required material and fills in a short online quiz at home to evaluate whether the required prior knowledge has been acquired to be able to solve and implement the computer problems during the session. The student’s score on the homework quizzes will be monitored by the didactical team but will not be taken into account for the final grade. However, submitting the answers to the quiz before the deadline is obligatory (failure to do so might be taken into account in the final grade).
REPORTS: Before the start of the Christmas holidays the student hands in a report of each computer session (3 in total) and uploads the corresponding computer code through Toledo. The deadline is communicated through Toledo.
EXAM: During the final examination, the student will be questioned on each of the 3 course parts: (1) Data-driven multi-channel filter design, (2) Tensor methods, (3) Nonlinear signal analysis and deep learning. Example questions will be provided on Toledo for each part. The emphasis of the exam is on understanding and insight; perfectly reproducing the relevant course notes is not a sufficient condition to pass. The exam is ‘closed book’, yet a formularium is available (no other material is allowed).
The weights for each part in the total score are the following:
Part 1 (Data-driven multi-channel filter design): 35%
Part 2 (Tensor methods): 30%
Part 3 (Nonlinear signal analysis and deep learning): 35%
In all parts, the grades for the theory and the project report are weighted as 60% and 40% respectively.
Note on plagiarism: all reports and computer code will be scanned for plagiarism. Software plagiarism also includes copying/manipulating pieces of code from other students (or other sources) without mentioning the source. Hiding software plagiarism by manipulating/reworking existing code is treated as fraud. All such plagiarism cases will be forwarded to the faculty exam committee.
Information about retaking exams
The student is asked to remake the same 3 exercise sessions and update the reports. Feedback on the reports is given upon request, yet only within the foreseen time span as mentioned in the examination rules of the university.
The reports need to be submitted 10 days before the examination date at the latest. The theory and the written reports are re-evaluated similarly as for the initial exam.