L E C : 1

B i Boim
om e eddi icc aa ll
S i g nS ai gl n a l
Processing
Processing
 Course description
●
The course introduces techniques and procedures for analysis of
biomedical signals and images like:

- Cardiology signals
* electrocardiogram - ECG

- Neurophysiology signals
* electromyogram – EMG (electrohysterogram - EHG)
* (electroencephalogram – EEG; See course: Human-Computer
Interaction)
- Medical images
* computed tomography images – CT images

with the emphasis on problems of biomedical researches.

Course: 63514 Biomedical Signal and Image Processing

 Course description
●
We will recognize how we can automatically detect and classify heart beats and then, non-invasive
and punctually, within 24-hour electrocardiogram signals, detect transient ischaemic disease, which is
one of the most terrible heart diseases, and if we do not discover it punctually, it may lead to heart
infarct.

●
We will recognize how we can, using some non-linear signal processing techniques, analyze
electromyograms recorded from the abdomen of a pregnant women early during pregnancy (23 week),
estimate, or try to predict, danger of pre-term birth.

●
We will also recognize techniques of analysis of 2- and 3-dimensional tomographic images with the aim of
extraction and visualization of anatomic structures of human body organs.

Course: 63514 Biomedical Signal and Image Processing

 Course syllabus
●
Introduction to Biomedical Signal and Image
Processing
●
Data Acquisition
●
Digital Filters
●
Fourier Transform
●
The Z Transform
●
Frequency-Domain Analysis of Digital
Filters
●
Electrocardiogram (ECG) and Detecting Transient Ischaemia
●
Feature Extraction, Shape Representation and Processing of the ECG

●
Spectral Analysis
●
Electromyogram (EMG) and Predicting Pre-Term
Delivery
●
Non-Linear Signal Processing Techniques

Course: 63514 Biomedical Signal and Image Processing

 Course syllabus
●
Image Filtering and Enhancement
●
Edge Detection and Segmentation of Images
●
Computed Tomography (CT) and Visualization of Human
Organs
●
Algorithms to Detect Transient ST Segment Episodes

●
Performance Measures and Evaluation

●
(Semi-Automatic Graphic Editing Tools to Annotate Ambulatory ECG
Records)
●
(Selected topics)

Course: 63514 Biomedical Signal and Image Processing

 Topics
●
The topics cover:

- Representation of international standardized databases of signals and images

(MIT/BIH DB, LTST DB, TPEHG DB, TPEHGT DS, CTIMG DB)

- Techniques for noise extraction

- Spectral analysis, modeling

- Techniques for feature extraction from signals and images

(filtering techniques, principal components, Karhunen-Loeve transform, sample entropy, edge
detection, contour extraction)
- Analysis and visualization of diagnostic and morphology feature-vector time series, and
anatomic
structures
- Event detection, clustering, classifications

- Metrics, techniques and protocols for performance and robustness evaluation of biomedical
computer systems

Course: 63514 Biomedical Signal and Image Processing

 ●
Web classroom

●
PhysioNet site, http://www.physionet.org
(source of resources in the field of biomedical signal and image
processing)
- Databases: MIT/BIH DB, LTST DB, TPEHG DB,
TPEHGT
- Software:DS ATM tools, wfdb library, lightWAVE,
gnuplot
●
Home pages of Laboratory for Biomedical Computer Systems and Imaging
(LBCSI),
http://lbcsi.fri.uni-lj.si
- Databases: LTST DB, TPEHG DB, TPEHGT DS, CTIMG DB
●
Ubuntu 18.04 (LTS),
Linux
●
C,
(C++)
●
Matlab

Course: 63514 Biomedical Signal and Image Processing

 Literature
●
Sornmo Leif, Laguna Pablo, Bioelectrical Signal Processing in Cardic
and
Neurological Applications, 2005, Elsevier Inc.
●
Clifford Gari D, Azuaje F, McSharry Patrick E (Editors), Advanced
Methods
and Tools for ECG Data Analysis, 2006, Artech House Inc.
●
Proakis J G, Manolakis D G, Digital Signal Processing, 2014, Prentice Hall (in our library)
Inc.
●
Gonzales Rafael C, Woods Richard E, Digital Image Processing, 3rd Edition,
2008,
Pearson Prentice Hall.
Gonzales Rafael C, Woods Richard E, Digital Image Processing, 4th Edition,
2018,
Pearson Prentice Hall.
●
Oppenheim Alan V, Schafer Ronald W, Discrete-Time Signal Processing, Third Edition,
2014,
Pearson Education Limited. (in our library)
●
Lyons Richard G, Understanding Digital Signal Processing, Third Edition, 2011, Pearson
Education,
Inc. (in our library)
Course: 63514 Biomedical Signal and Image Processing
 Laboratory sessions, quiz, exam, grading
●
Laboratory assignments (each assignment has to be submitted and explained ongoing):
1. Analysis of electrocardiogram (ECG) signals
and analysis of electromyogram (EMG) signals (Max: 45 - 55 points)
2. Analysis of computed tomography (CT) images (Max: 55 - 65 points)
→ Obligatory: 50 points; Max: 120 points

●
Quiz during the semester (by your choice) → Max: 20 points
●
Exam at the end of semester: → Obligatory: 50 points; Max: 100 points

●
How the preliminary grade (5 – 10) will be composed?

Total score = round ( ( Laboratory points + Quiz + Exam ) /2

)
Total score ≤ 49 → 5
50 ≤ Total score ≤ 59 → 6
60 ≤ Total score ≤ 69 → 7
70 ≤ Total score ≤ 79 → 8
80 ≤ Total score ≤ 89 → 9
90 ≤ Total score → 10

●
Oral exam is obligatory.
Course: 63514 Biomedical Signal and Image Processing
 Physionet / Computing in Cardiology Challenges
●
Physionet / 2014 Computing in Cardiology
Challenge
- Robust Detection of Heart Beats in Multimodal Data

- Students at this course participated at the prototype challenge

- The actual challenge started on January 2014 and ended on January 2015

- Urška Pangerc (student at this course) obtained the 2nd, 3rd, 6th, and the final 1st place during
the phases I, II, III, and the follow-up phase (end of February 2015) of the competition of
the actual challenge (which was not student competition) among 47 international teams
( http://physionet.org/challenge/2014 )

Course: 63514 Biomedical Signal and Image Processing

 PhysioNet / 2014 Computing in Cardiology
Challenge
Focus issue entry Score(%)

Pangerc and Jager (2015) 93.64

Johnson et al (2015) 91.50
Antink et al (2015) 90.70
DeCooman et al (2015) 90.02
Galeotti et al (2015) 89.73
*VollmerM 89.55
Pimentel et al (2015) 89.13
Mollakazemi et al (2015) 88.85
*Krug J 88.34
Gieraltowski et al (2015) 88.07
C-code sample entry 87.38
M-code sample en try 85.04

Source: Silva I, Moody B, Behar J, JohnsonA, Oster J, Clifford GD, Moody Editorial: Robust Detection
GB, of Heart Beats in Multimodal Data, Physiological Measurement, 36:1629–1644,
2015.
( http://iopscience.iop.org/0967-3334/36/8/1629 )
Course: 63514 Biomedical Signal and Image Processing
 Physionet / Computing in Cardiology Challenges
●
Physionet / 2020 Computing in Cardiology
Challenge
- The topic has not been decided yet

●
Participating at the topics from the past challenges is possible
(see: http://physionet.org/challenge/ )

Course: 63514 Biomedical Signal and Image Processing

 DATA ACQUISITION

●
Data acquisition
●
Continuous-time sinusoidal signals
●
Sampling a continuous-time signal
●
Quantization
●
Analog to Digital (A/D) and Digital to Analog (D/A)
conversion
●
Sampling a sinusoid – aliasing
●
The Nyquist sampling theorem
●
Relations among frequency variables
●
(Reconstructing continuous-time signals)

Course: 63514 Biomedical Signal and Image Processing

 Data acquisition
●
Data acquisition typically consists of three stages:
* Transduction (in general conversion of one form of energy to electrical energy which is suitable for
encoding into a computer)
* Analog signal conditioning (amplifying and filtering the analog signal measured with a transducer to
provide a goodmatch between the typically low-amplitude, wide- bandwidth transducer signals and the
analog-to-digital converter)
* Analog-to-digital converter (transforms a continuous-time signal into a digital signal: sampling – taking
amplitudes of continuous-time signal at the discrete times, quantization – sample amplitudes can only
take a finite set of values)

.A/D CONVERTER
.x(I') I .x[.1r:1 I I ,- '
I I
]1 G::;i(n
'TRANSOUC:
I. CONDITIONE Sli.MPL.ER - QUANTIZER
Efl
R I
I

Course: 63514 Biomedical Signal and Image Processing

 Continuous-time sinusoidal signals
●
Cosine signal
xa ( t ) = A cos (Ω t + θ ) , −∞ < t < ∞
A is the amplitude
Ω is the frequency in radians per second [rad/s], Ω = 2π F
Θ is the phase in radians [rad]
Tp is the duration of one cycle in seconds [s]
F = 1 / Tp is the frequency in cycles per second or Hertz [Hz], Hz = 1/s

Course: 63514 Biomedical Signal and Image Processing

 Sampling a continuous-time signal
● Discrete-time signals are obtained by sampling a continuous-time signal x(t)at
regular intervals

●
Ts is the sampling interval or sampling period in seconds [sec], [s]
● Fs is the sampling frequency or sampling rate in samples per second [smp/s]
or in [Hz], [1/s]
x[0]

x[-1] x[1]

x[2]

Course: 63514 Biomed cai Sl gna

i andl mageI P ocess rng i n
 Quantization

●
A quantizer takes x[n] and produces a signal xq[n] that can only take a finite
number of values
● The quantizer output xq[n] is usually equal to the closest integer
● The number of quantization steps is a power of two
● The quantizer encodes signals whose values lie in the range:

−V max ≤ x [n] < V max

●
Where Vmax is related to the number of quantization steps by:
B−1
●
and B is the number of bits of the quantizer V max = 2

Course: 63514 Biomedical Signal and Image Processing

 Quantization
●
xq[n] as a function of x[n] for B = 4 corresponding to Vmax = 8

9
8
7
6
5
-c
J 4
z 3
(!)
V)
2
1
~ 0
N-1
~-2
~-3
0-4
-5
-6
-7
-8
-9
-12-\,- .!e ..!7 .!e ..!a..:...:., ..!2 ..!, b i II I
a e
I I I
10 11
I

INPUT SIGNALI .J 4 12
\0.!e >( [n "]

Course: 63514 Biomedical Signal and Image Processing

 .
.

Analog to Digital (A/D) and Digital to Analog (D/A)

conversion
●
Continuous-time signal, xa(t)
● Discrete-time unquantized samples, xa(nt)
Amplitude
● A/D → Discrete-time quantized samples, xq(nt) Time Discretization
Discretization
● Output of D/A converter, xq(t)

I I
-, I
/ Quantization
Level

.-1-0riginal -, Analog
/ Signal/
xa(nT) → x[n] 4.1
/
-+---+--'---t---;;r-i

Discrete signal 16 .:
1---+- _ ,,,.. .....~
~,
. xa(t)

)
/
26 1-- ) ~'~ '~- - \
I
Unquantized Samples / , ·
xa(nt ) J...--L.---if---/ ,,t-t·-- Quantizatio. n

xq(nT) → x[n] I --+----t- \/

_,... v
--->e;~-:;-;r/1 I i 6 Step
.1
t
0

·E- I
"O
.:.: Quantized~ "\ Output of Zero-Order /
Digital signal .. 0
:: Samples --1--\--t--::7"",-~HoldDIA Converter •,,---1
xq(nt) \ »> xqCt) ) 1
<,
-
\
-tl Range of the
0. Quantizer
L-i-
L-i---+-+--+-Pri"'r--t--~J(-r-11
I
-2.1

- ~
[Proakis, Manolakis] ~D,.L---l~--i-~-+~-+~-+-~-t-
36

~'t'"--t---"1'---r-
0 I 21 4I 57 6 7 87 9T
t
Course: 63514 Biomedical Signal and ImagTeimPreocessing 7 7
 Sampling a sinusoid - aliasing
●
Sampling a continuous-time sinusoid:
F
FS = f
x[n] a oos(27rnF / F; + ¢)
●
x[n] hides a difficulty arising from the ambiguity of frequency for discrete-time
sinusoids:
* It is not possible to know if the frequency of the original continuous-time signal x(t) was
F, or F + Fs, or F + 2Fs, etc; or, Fs – F or 2Fs – F, etc
●
This phenomenon is known as aliasing because frequencies may not be what they appear to be
once a continuous-time signal x(t) is sampled
●
Aliasing – the error in a signal arising from limitations in the system that generates or
processes the signal (Collins English Dictionary)

●
F – continuous-timefrequency in cycles per second [cyc/s], [Hz]
●
f - discrete-timefrequency in cycles per sample [cycles/sample], [cyc/smp]

Course: 63514 Biomedical Signal and Image Processing

 Sampling a sinusoid - aliasing

Course: 63514 Biomedical Signal and Image Processing

 The Nyquist sampling theorem

●
How to avoid aliasing?

●
Regarding the previous example, what is the minimum number of samples
per sinusoid, N, that would still approximate a sinusoid? N = ?

●
Regarding the previous example, what is the highest frequency F (expressed
with Fs) of a sinusoid that would still be approximated, if using sampling
frequency Fs?

Since: Fs / F = N and N ≥ 2, follows: F = Fs / 2

●
If Fs = 2.F, the Fs is said to be Nyquist frequency

Course: 63514 Biomedical Signal and Image Processing

 The Nyquist sampling theorem

●
Sampled analog signal x(t) should not contain frequencies higher than
Fs / 2;

●
Sampling frequency Fs should be higher than twice the highest frequency
F present in the analog signal, Fs ≥ 2. F

1
=> Principal value of discrete-time frequency 0 ≤f ≤
2

Course: 63514 Biomedical Signal and Image Processing

 The Nyquist sampling theorem
●
In practice always avoid aliasing by low-pass filtering the continuous-time
signal x(t) before sampling
● In practice sample signals at about Fs = (3-4).F

NYQUIST SAMPLING WITH ANALOG

LOWPASS FILTER
DIGITAL
ANALOG SIGNAL
LOWPASS ADC
FILTER
•
DATA
DATA RATE z f
\,
'•top='•
... ,,;,
/2

,.

Course: 63514 Biomedical Signal and Image Processing

 Relations among frequency variables
Lowpass filtering and sampling
xa(t) = A sin(2πF t + θ) xa[nTs] = A sin(2πF/Fs n + θ) = x[n]
= A sin (Ωt + = A sin (ωn + θ) = x[n]
θ) Ω= ω= 2πF/Fs → ω= 2πf
2πF
I Spectral band
I< of interest
(Fs = 6 kHz) F = f
Ω, - ∞ < Ω < ∞, Ω = 2 π F, Spectral band
the frequency in radians per sec [rad/s] of interest FS

F, - ∞ < F < ∞,
-3 - 0 1 3 4 6 7 Fre
the frequency in cycles per sec or Hertz [Hz] 2 q
(kHz)
f , - 1/2 ≤ f ≤ 1/2,
the frequency in cycles per sample [cyc/smp] −F S / 2 0 F S/ FS F
ω, - π ≤ ω ≤ π, ω = 2 π f, −1/ 2 0 2 1 f

the frequency in radians per sample [rad/smp] −π 0 1/2 2π ω

π
Course: 63514 Biomedical Signal and Image Processing
 (Reconstructing continuous-time signals)
●
If a continuous-time signal x(t) contains no frequency components higher than F,
it can be exactly reconstructed from samples taken at a frequency Fs > 2F
●
The Nyquist theorem gives an explicit interpolation formula for reconstructing x(t)
from the discrete-time signal x[n]:
00

x(t) L ,~r[n] ¢,(t -

nT~)
n=-oo
with a basic function Φ(t)t): sin(1rFst
¢,(t )
)
1rF,.
,t
●
Time-dependent weights Φ(t)t - nTs) are obtained by delaying the basic function Φ(t)t)

Course: 63514 Biomedical Signal and Image Processing

 (Reconstructing continuous-time signals)

●
The basic function Φ(t)t):

●
This function verifies the property:

1 ifn=O
¢,(nT.~) - { O otherwise
1

●
This property implies that x(t) = x[n] for t = nTs
● The signal is said to be sampled at Nyquist frequency if Fs = 2F

Course: 63514 Biomedical Signal and Image Processing