Pengolahan Sinyal Biomedis

(1)

Achmad Rizal

Biomedical Signal Processing & Instrumentation RG (BioSPIN RG)

School of Electrical Engineering, Telkom University

(2)

Short Bio

(3)

(4)

Outline

• Teori Sinyal Biomedis

• Teknik Dasar Pengolahan Sinyal Biomedis

• Pengolahan Sinyal Biomedis pada Domain Waktu

• Pengolahan Sinyal pada Domain Frekuensi dan Waktu Frekuensi

• Pengolahan Sinyal pada Domain Wavelet

(5)

Pendahuluan

• Sinyal Biomedis  sinyal yang dihasilkan oleh proses fisiologi dalam tubuh manusi

• Sinyal biomedis  Informasi kesehatan

• Sifat dasar  ampitudo rendah, frekuensi rendah, stokasticquasi periodic

• Pengembangan metode analisis berbasis computer  metode yang tepat

• Perlu pemahaman karaktistik sinyal dan system fisiologi terkait

(6)

Biomedical signal classification

On the basis of

– signal characteristics

technical point of view – signal source

from where and how the signal is originated and measured – biomedical application

cardiology, neurophysiology, monitoring, diagnosis,…

Classification may be helpful in the selection of processing methods...

(7)

Biomedical signal classification

(8)

Biomedical signal classification

Classification by source

• Bioelectric signals

• Biomagnetic signals

• Bioacoustic signals

• Biomechanical signals

• Biochemical signals

• Bioimpedance signals

• Biooptical signals

(9)

Biomedical Signal Examples

Measurement Range Frequency, Hz Method

Blood flow 1 to 300 mL/s 0 to 20 Electromagnetic or ultrasonic Blood pressure 0 to 400 mmHg 0 to 50 Cuff or strain gage

Cardiac output 4 to 25 L/min 0 to 20 Fick, dye dilution Electrocardiography 0.5 to 4 mV 0.05 to 150 Skin electrodes Electroencephalography 5 to 300 V 0.5 to 150 Scalp electrodes Electromyography 0.1 to 5 mV 0 to 10000 Needle electrodes

Electroretinography 0 to 900  V 0 to 50 Contact lens electrodes

pH 3 to 13 pH units 0 to 1 pH electrode

pCO₂ 40 to 100 mmHg 0 to 2 pCO₂ electrode

pO₂ 30 to 100 mmHg 0 to 2 pO₂ electrode

Pneumotachography 0 to 600 L/min 0 to 40 Pneumotachometer Respiratory rate 2 to 50

breaths/min 0.1 to 10 Impedance

(10)

Arterial Blood Pressure EEG

Biomedical Signal Examples

:

(11)

Biomedical signal processing application domains

• Information gathering

– measurement of phenomena to understand the system

• Diagnosis

– detection of malfunction, pathology, or abnormality

• Monitoring

– to obtain continuous or periodic information about the system

• Therapy and control

– modify the behaviour of the system and ensure the result

• Evaluation

– objective analysis: proof of performance, quality control, effect of treatment

(12)

Problems in biomedical signal processing

• Accessibility

– Patient safety, preference for noninvasiveness – Indirect measurements (variables of interest

are not accessible)

• Variance

– Inter-individual, intra-individual

• Inter-relationships and interactions among physiological system

– Subsystem of interest may not be isolated

• Acquisition interference

– Instrumentation and procedures modify the system or its state

(13)

Problems in biomedical signal processing

• Artefacts and interference

– Interference from other physiological systems (e.g. muscle artifacts in EEG recordings) – Low-level signals (e.g. microvolts in EEG) require very sensitive

amplifiers; they are easily sensitive to interference, too!

– Limited possibilities for shielding or other protection

• Nonlinearity and obscurity of the system under study

– basically all biological systems exhibit nonlinearities while most of the methods are based on the assumption of linearity → approximation

– exact structures and true function of many physiological systems are often not known

(14)

Pokok Bahasan 2

Pengolahan Sinyal Biomedis pada Domain Waktu

(15)

Sinyal Biomedis pada Domain Waktu

• Sifat sinyal secara alami  f(t)

• Beberapa proses yang dilakukan:

• Proses Normalisasi: DC removal, normlaisasi amplitudo

Normalisasi Filtering Ekstraksi

ciri Klasifikasi

(16)

Sinyal Biomedis pada Domain Waktu

Filtering = removing unwanted frequency component (noise) from the signal

• Linear Filter : LPF, HPF, BPF, BSF, notch filter

• FFT Filter : use FFT to remove signal  IFFT

• Non Linear Filter  phase-locked loops, detectors

• Mixers, median filters, ranklets

• Adaptive filter  RLS, LMS, etc

(17)

Perancangan filter

• Parameter perancangan filter

• Frekuensi sampling

• jenis filter (LPF, BPF, dsb), (Butterworth, Chebychef atau Elliptic)

• Frekuensi yang akan dihilangkan (frekuensi stop band, frekuensi passband)

• Ripple pada pass-band dan ripple pada stopband

• Keluaran perancangan filter

• orde filter

• koefisien filter

• Respon frekuensi

(18)

Perancangan filter

Example 1:

% For data sampled at 1000 Hz, design a lowpass filter with less than

% 3 dB of ripple in the passband defined from 0 to 40 Hz and at least

% 60 dB of ripple in the stopband defined from 150 Hz to the Nyquist

% frequency (500 Hz):

Wp = 40/500; Ws = 70/500;

Rp = 3; Rs = 60;

[n,Wp] = cheb1ord(Wp,Ws,Rp,Rs) % Gives minimum order of filter [b,a] = cheby1(n,Rp,Wp); % Chebyshev Type I filter

freqz(b,a,512,1000); % Plots the frequency response

(19)

Perancangan filter

Example 2:

% Design a 6th-order Elliptic band-pass filter which passes

% frequencies between 0.2 and 0.5, and with 5 dB of ripple in the

% passband, and 80 dB of attenuation in the stopband

[b,a]=ellip(6,5,80,[.2,.5]); % Bandpass digital filter design h = fvtool(b,a); % Visualize filter

(20)

Ekstraksi ciri

• Ciri statistic : max, min, mean, variance, entropy, skewness, kurtosis

(21)

HJORTH DESCRIPTOR

• Hjorth Descriptor (Hjorth 1970) Variasi sinyal orde 1

Variasi sinyal Orde 2

(22)

Pengukuran Entropy

• Shannon Entropy

• Spectral Entropy

• Renyi Entropy

• Tsallis Entropy

(23)

Pengukuran Entropy

• Wavelet Entropy

• Approximate entropy

• Sample entropy

(24)

Ekstraksi ciri

Time domain feature

• Zero-crossing

• Root mean square

• Log detector

• Mean Absolute value (MAV)

𝑍𝐶 =

𝑛=1 𝑁

𝑠𝑖𝑔𝑛 𝑥 × 𝑥_𝑛+1 ∩ 𝑥 − 𝑥_𝑛+1 ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑

𝑠𝑔𝑛 = 1, 𝑖𝑓 𝑥 ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 0, 𝑜𝑡ℎ𝑒𝑟 𝑤𝑖𝑠𝑒

𝑅𝑀𝑆 = 1 𝑁𝑛=1

𝑁

𝑥_𝑛²

𝐿𝑂𝐺 = 𝒆^𝑵¹ ^𝑛=1^𝑁 ^{log 𝑥}^𝑛

1 ^𝑁

(25)

Ekstraksi ciri

Time domain feature

• Waveform length (Wave)

• Standard deviation (STD)

• Slope-sign change (SSC)

𝑉𝐴𝑅 = 1 𝑁 − 1

𝑛=1 𝑁

𝑥_𝑛² 𝑊𝐿 =

𝑛=1 𝑁

𝑥_𝑛+1 − 𝑥_𝑛

𝑆𝑇𝐷 = 𝑉𝐴𝑅

𝑁

1, 𝑖𝑓 𝑥 ≥ 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑

(26)

Contoh Aplikasi

• ECG signal classification using Hjorth Descriptor

A. Rizal and S. Hadiyoso, “ECG signal classification using Hjorth Descriptor,” in 2015 International Conference on Automation, Cognitive Science, Optics, Micro Electro-Mechanical System, and Information Technology

(27)

Contoh Aplikasi

• T. Yingthawornsuk, “Classification of ECG Signals using Modified Hjorth Descriptors,” in 2018 14th International Conference on

Signal-Image Technology & Internet-Based Systems (SITIS), 2018, pp. 345–350.

• S.-H. Oh, Y.-R. Lee, and H.-N. Kim, “A Novel EEG Feature Extraction Method Using Hjorth Parameter,” Int. J. Electron. Electr. Eng., pp.

106–110, 2014.

• A. Rizal, R. Hidayat, and H. A. Nugroho, “Entropy Measurement as Features Extraction in Automatic Lung Sound Classification,” in the 3rd International Conference on Control, Electronics, Renewable Energy, and Communications 2017 (ICCEREC 2017), 2017.

(28)

Pokok Bahasan 3

Pengolahan Sinyal Biomedis pada Domain

Frekuensi dan Time Frekuensi Domain

(29)

Mathematical Transformation

• Why

• To obtain a further information from the signal that is not readily available in the raw signal.

• Raw Signal

• Normally the time-domain signal

• Processed Signal

• A signal that has been "transformed" by any of the available mathematical transformations

• Fourier Transformation

• The most popular transformation

(30)

Transformasi Fourier

• One way to find the frequency content

• Tells how much of each frequency exists in a signal

  ^N   _N^kn

n

W n

x k

X   ^  

 1

0

1 1

  ^N   _N^kn

k

W k

N X n

x ^



 ^ ^



 ¹

0

1 1 1



 



 

 ^j ^N

N e

w

 2

 f x t e dt

X ^²^j^ft











 t X f e df x  ^  ²^j^^ft





(31)

Sinyal EKG

(32)

Sinyal Suara Paru

(33)

Sinyal suara paru

• Stridor & specturm

(34)

Fitur pada domain frekuensi

• Peak frequency

• Quantile frequency

• Mean frequency

• Median frequency

• Maximum to Minimum Drop in Power Density Ratio

• Signal to noise ratio

• Power Spectrum deformation

• Signal to Motion artifact ratio etc

(35)

Praktek Matlab

t = 0:.001:.25;

x = sin(2*pi*50*t) + sin(2*pi*120*t); plot (t,x);

%tambah random sinyal dengan standard deviasi 2 y = x + 2*randn(size(t));

plot(y(1:50))

title('Noisy time domain signal')

%analisis komponen frekuensi Y = fft(y,256);

Pyy = Y.*conj(Y)/256;

f = 1000/256*(0:127);

figure;

plot(f,Pyy(1:128))

• %zoom hanya sampai 200Hz plot(f(1:50),Pyy(1:50))

title('Power spectral density') xlabel('Frequency (Hz)')

(36)

SFORT TIME FOURIER TRANSFORM (STFT)

• Dennis Gabor (1946) Used STFT

• To analyze only a small section of the signal at a time -- a technique called Windowing the Signal.

• The Segment of Signal is Assumed Stationary

• A 3D transform

 ^t ^f  ^x ^t ^t ^t  ^e ^j ^ft^dt

t





    ^*    ²

X ,

STFT

 t : the window function



A function of time and frequency

(37)

DRAWBACKS OF STFT

• Unchanged Window

• Dilemma of Resolution

• Narrow window -> poor frequency resolution

• Wide window -> poor time resolution

• Heisenberg Uncertainty Principle

• Cannot know what frequency exists at what time intervals

Via Narrow Window Via Wide Window

(38)

Stockwell Transform

• S-Transform : special case of STFT

• General equation :

• Window size depend on frequency

(39)

Cohen’s Class Distribution

• General equation:

• g(v,f) = 2D filter to reduce cross-product

• Wigner-Ville distribution (WVD), g(v,f) = 1

• Drawback of WVD  cross-product

• Other Cohen’s Class Dist : ZAM, Choi-William, Rihazcek, etc

(40)

Contoh aplikasi

• Classification of Pulmonary Crackle and Normal Lung Sound using

Spectrogram and Support Vector Machine (ICEBEHI 2021, accepted)

• Segmentasi : 25-20/250-200/500-475

• FFT : 256, 512

(41)

(42)

Analisis Wavelet

(43)

PRINCIPLES OF WAELET TRANSFORM

• Split Up the Signal into a Bunch of Signals

• Representing the Same Signal, but all Corresponding to Different Frequency Bands

• Only Providing What Frequency Bands Exists at What Time Intervals

(44)

DEFINITION OF CONTINUOUS WAVELET TRANSFORM

• Wavelet

• Small wave

• Means the window function is of finite length

• Mother Wavelet

• A prototype for generating the other window functions

• All the used windows are its dilated or compressed and shifted versions

      ^dt

s t t

s x s

s _x

x 



 



  













 ^ 

 1 *

, ,

CWT

Translation

(The location of the window)

Scale

Mother Wavelet

(45)

RESOLUTION OF TIME & FREQUENCY

Frequenc y

Better time resolution;

Poor

frequency resolution

Better

frequency resolution;

(46)

COMPARSION OF TRANSFORMATIONS

(47)

SUBBABD CODING ALGORITHM

• Halves the Time Resolution

• Only half number of samples resulted

• Doubles the Frequency Resolution

• The spanned frequency band halved

0-1000 Hz

D₂: 250-500 Hz

D₃: 125-250 Hz Filter 1

Filter 2

Filter 3

D₁: 500-1000 Hz

A₃: 0-125 Hz A₁

A₂

X[n]

512

256

128

64 128

256

S S

A₁

A₂ D₂

A D

D₁

(48)

Multiskala

(49)

Proses Multiskala

• Proses mendekomposisi menjadi beberapa sinyal pada skala/level/subband yang berebda

• Coarse grained procedure (Costa et.al 2002)

(50)

Hasil coarse-grained procedure

(51)

Proses Multiskala

Proses multiskala yang lain

• Dekomposisi wavelet  DWT, WPD  subband

• Empirical mode decomposition  IMFs (Huang et al., 1998)

• Variasi lain dari EMD  EEMD, VMD etc

• Multidistance signal level difference (MSLD) dan variasi nya (Rizal et al, 2017, Rizal et al, 2019 )

(52)

EMD

• Empirical Mode Decomposition (EMD) (Huang, 1998)

membentuk deretan intrinsic mode function (IMF) dari sinyal Langkah-langkah

1. Hitung upper dan lower envelope sinyal, hitung rata-ratanya, 2. Jika h1(t) bukan IMF, ulangi langkah 1 dan hitung

(53)

EMD

3. Jika terjadi IMF, m1k  0 maka h1k(t) = c1(t) atau IMF pertama 4. Akan didapat res1(t)= x(t)-c1(t). Selanjutnya res1(t) akan menjadi

masukan untuk mencari IMF selanjutnya, sampai envelope monotonik

5. Sinyal x(t) dapat dinyatakan sebagai:

Dengan c1(t)+…+ck(t) adalah IMF

(54)

Hasil Empirical Mode Decomposition

Suara Bronchial  5 IMF

(55)

Strategi pengukuran kompleksitas sinyal multiskala

Sinyal didekomposisi Sinyal tetap

(56)

Contoh Penerapan Strategi 1

Rizal, Hadiyoso, Wijayanto, 2020

Rizal, A., Hidayat, R., & Nugroho, H. A. (2019b). Lung Sound Classification Using Hjorth Descriptor Measurement

(57)

Contoh Penerapan Strategi 2

Rizal, A., Estananto, E., & Atmaja, R. D. (2019). Epileptic EEG Signal Classification Using Multiresolution Higuchi Fractal Dimension. International Journal of Engineering Research and Technology, 12(4), 508–511.

Rizal, A, Hidayat, R., & Nugroho, H. A. (2018). Multilevel wavelet packet entropy: A new strategy for lung sound feature extraction based on wavelet entropy. In Proceeding of 2017 International

(58)

Beberapa Metode Multiscale dan

Pengukuran Kompleksitas Sinyal

(59)

MSLD dan multiskala lainnya

GLDM

Sinyal 2D

H(g|θ) = |x(i±d,j) – x(i,j±e)| H(g|0) = |x(i) – x(i+D)|

Modified GLDM

Sinyal 1D, θ = 0

Multidistance Signal

level difference (MSLD-A)

𝑦^𝐷 𝑖 = 𝑥 𝑖 − 𝑥(𝑖 + 𝐷 ,) 𝐷 = 1, 2, … , 𝐾

Sejumlah K sinyal baru hasil nilai

absolut selisih sampel sinyal pada jarak D

MSLD-B

𝑦^𝐷 𝑖 = 𝑥 𝑖 − 𝑥(𝑖 + 𝐷 ,) 𝐷 = 1, 2, … , 𝐾

Sejumlah K sinyal baru hasil selisih sampel sinyal pada jarak D

MStepLD

𝑦^𝐷 𝑖 = |𝑦^𝐷−1 𝑖 − 𝑦^𝐷−1 𝑖 + 𝐷 | 𝑦¹ 𝑖 = 𝑥 𝑖 − 𝑥(𝑖 + 1)

𝐷 = 1, 2, … , 𝐾

MSDownLD

𝑦^𝐷 𝑖 = | 𝑥 𝑖 − 𝑥 𝑖 + 𝐷 | ↓ 𝐷, 𝐷 = 1, 2, … , 𝐾

Proses downsampling untuk meniru

(60)

Variansi MSLD dan Variansinya

(61)

Multilevel signal complexity

• E₁ = energy of D₁

• E₂ = energy of D₂

• E₃ = energy of D₃

• E₄ = energy of A₃

• E_tot = ∑E_i i = 1,…,4

Level 1

Level 2

Level 3 A3

Original Signal

D1

A2 D2

D3 A1

Wavelet Entropy (Rosso et al., 2001): contoh perhitungan WE level 3

𝑀𝑊𝐸

_𝑁

= 𝑊𝐸

₁

, 𝑊𝐸

₂

, … , 𝑊𝐸

_𝑁

Multilevel wavelet entropy :

(62)

Multilevel signal complexity

𝑀𝑊𝑃𝐸_𝑁 = 𝑊𝑃𝐸₁, 𝑊𝑃𝐸₂, … , 𝑊𝑃𝐸_𝑁

Wavelet entropy  DWT

Wavelet packet entropy WPD

Multilevel packet wavelet entropy :

Rizal, Hidayat, Nugroho, 2017

(63)

Hasil MWE dan MWPE untuk Suara Paru

Multilevel wavelet entropy dan multilevel wavelet packet entropy

(64)

Aplikasi untuk sinyal lainnya

• PVC pada sinyal ECG  MWE

A. Rizal, R. Riandini, and T. Tresnawati, “Premature Ventricular Contraction Classification based on ECG Signal using Multilevel Wavelet entropy,” in The 2018 International Conference on Enhanced Computer Research, Engineering, and Advanced Multimedia, 2018, pp. 1–5.

(65)

Aplikasi untuk sinyal lainnya

Epileptic EEG  EMD+Entropy

(66)

Kemungkinan Pengembangan

• Penentuan skala dan subband wavelet menggunakan metode yang lebih baik

• Reduksi ciri dilakukan menggunakan FSS

• Kolaborasi ML yang lebih maju untuk klasifikasi

• Modifikasi pengolahan citra untuk pengolahan sinyal biologi

• Pemanfaatan untuk kasus yang lain (bearing fault, machine diagnostic system, speech)

(67)

Penutup

• Pengolahan sinyal digital memerlukan pemahaman karakteristik dari sinyal

• Signal complexity salah satu cara memahami sifat sinyal biomedis

• Sifat multiskala dari sinyal biomedis menarik untuk di eksploitasi lebih lanjut

• Gabungan pengolahan sinyal dan AI akan powerful dalam memecahkan masalah klasifikasi sinyal biomedis

•

(68)

(69)

WPD orde N

• T = wpdec(x,3,'db2')

(70)

Wavelet packet decomposition

(71)

Sinyal Suara Paru

(72)

Sinyal suara paru

• Stridor & specturm