Sie sind auf Seite 1von 3

Rem moving Basel Wander in EC Wa g line W r CG aveform ms

S. Akhter S Virginia Comm V monwealth U University 401 We Main Str est reet Richmo ond, VA 232 284 11/14/11 1
Abstract The Electroca ardiogram(ECG is a standard G) d physiologic biomedical signal used to diagnose the cal o e abnormalit ties of heart beat rhythms. This signal is b s distorted by a number of artifacts; the artifact tackled f a d in this pa aper is the ba aseline wander artifact. The r e method un ndertaken is a baseline wan nder estimation n using samp rate alteratio and a low-pass filter which ple on h is then rem moved from the original signal. The root-mean n squared(RM error and the ST distortion is compared MS) d to that of th Kalman filte used by MIThe er -BIH to remove e the baselin wander. Th method sho ne he ows promising g results for a low-end emb bedded filter application that a t ily w t can be easi replicated with electronics, yet falls short of the MIT-BIH standard. . I. INTRO ODUCTION iography (ECG) is a non-invasiv measurement ) ve t Electrocardi of the elect trical activity ar round the heart. Specifically, it . t measures th depolarizatio of the heart muscle through he on h small chang in the skin po ges otential. The cha aracteristic wave e of the Electr rocardiogram is shown below: Hype ercalcaemia or Hypocalcaemi for examp ia ple, or flatten ned or inverted T waves cou indicate Co d uld oronary Ischem mia, or Ventri icular Hypertro ophy. ECG tur rns the monit toring for such diseases into an automated t h task of determ mining if a b bio signal is behaving expe ectedly. Howe ever usual ECG signals are sub G bjected to artifac due cts to res spiration, power r-line interferenc muscle activ and ce, vity baseli drift. These are represented in Figure 2 and can be ine order of magnitudes larger than the r rs relevant signal.

Figur 2: Baseline Drift Artifact Due to Movement in ECG re e n These artifacts usua e ally have a m much lower frequency comp pared to the orig ginal signal and can thus be re d emoved using linear filters. There are three filter structur that g e res work well in removing the baseline a e artifact: forwa ard/backward IIR filter, a comb FIR filter or a sample R rate a alteration. The standard used by the Massac chusetts Institu of Technolo ute ogy/Beth Israel Hospital is a f forward /back kward Kalman f filter [1]. The m model presented in this paper is an estimatio of the baseline by decimati the r on ing wave form and low-p passing to smoot any remnants of the th s QRS complex. The b baseline estimate will be remove from e ed the s signal to get th periodic ECG signal and w be he G will comp pared to standard d. II. METHODS S A. Co oncept The s signal is parallel processed thro ough a delay filter and conse ecutively decima ated, low-passed and interpolate The d ed. result of the second b t branch is remove from that of t first ed the as sho own in Figure 3. .

C W Figure 1: ECG Characteristic Wave h The wave has peaks which are denoted by the letters P, Q, S, R and T which refer to ke electrical eve w ey ents. The P-wave e occurs due to the depolariza t ation of the atria muscles; atrial al l repolarizatio is obscured due to ventricula depolarization on d ar n a few mill liseconds later. This delay happens because e conduction in the atrial muscles are prevented from p m e e s reaching the Ventricles by non-conductive tissue and thus have to trav through the Atrio-Ventricular (AV) node and vel A r d Purkinje fibers. The QRS complex represents the f Q e ventricular depolarization and the T wave represents w s r ventricular repolarization. ws iguities in heart t ECG allow clinicians to diagnose ambi behavior. Unusual QT intervals could indicate T c e

Figure 3: Bloc Diagram Mod ck del The integer decimation fa r actor, D, is ch hosen based the e bandwidth of the original signal. The signa must be band o s al d limited to | /D to pre | event aliasing as any frequency s y above that is to be eliminate The cutoff frequency for the i ed. fr e Low-pass filter portion is D times the pre fi evious cutoff as s given by th equation he . The in nterpolation is a linear interp polation where zeros are padd between the ded e samples. Th delay is design proportional to the length of he ned l f the Low-pass filter and com mpensates for th delays for the he e other comp ponents in the second branch. The power-line s e interference is measured using an FFT and removed with a e simple FIR notch filter. Th RMS error an ST deviation he nd n ompared to tha using the Kalman filter to at K o will be co determine how well the met h thod fared. B. Implemen ntation The datasets used were pro ovided by the bi iomedical signal l processing graduate class at Virginia Commonwealth s h University containing an id ECG wavef c deal form, one where e an artificial baseline was added and an actual measured l a d dataset with baseline wande The heart rat was relatively h er. te y constant thu a time-varyin filter was not necessary. The us ng t e FFT of the signal was computed for bot the ideal and e th d artificial EC datasets as sh CG hown in Figure 4. 4 Fig gure 6: Magnitud Response of F Low-Pass F de FIR Filter The o output of this filt was then imp ter plemented on an actual ECG signal with base eline drift and its results were s pared to the Kalm filtered sign used by MIT/ man nal /BIH. comp III. RESULTS S g mparing to the signal Using the artificial dataset and com befor the artificial baseline and af re fter it was add and ded ults Figure 8. The b baseline filter gives the resu shown in F mate ove ure estim used to remo the artifact is shown in Figu 7. F Figure 5: Tempo and Spectral Representation of oral l D Decimated Sign nal A low w-pass filter wi the magnitu response sho ith ude own in Figur 6, was then ap re pplied to the sig gnal and it was l linearly interp polated back to t original 600 the 0Hz and remove from ed the si ignal with an arti ificial baseline.

Figure 4: The Spectral Com T mponents of a 600Hz Sampled ECG (bl lue) and Artificial Baseline Artif (yellow) fact The frequen of the baseli as shown is less than 0.5Hz ncy ine z thus a larg decimation factor of 150 can be used. ge Decimating the signal by this factor yie elds the results s shown in Figure 5.

igure 7: Baseline Estimate (yello and ECG (bl e ow) lue) Fi

IV DISCUSSIO V. ON As se een from Figur 10, the Kalm obviously does a re man better job of estimati the baseline yet consists of much r ing e, f greate filter logic. T sample rate alteration estimate can er The be m made with simp D-A and A ple A-D converters and a relativ vely low order F filter. The inability of the system FIR in to tra the changes i the MIT/BIH dataset may be due to ack H e ampling frequen as the curren system was de ncy nt esigned the sa for 6 600Hz sample r rate or that the heart rate had been e d altere and thus a time-variant low-pass filte was ed er neces ssary. Futur work would b to implemen a time-varyin filter re be nt ng that i proportional to the heart r is rate or to modi the ify param meters to be ad daptable to diff fferent sampling rates. g Appli ications of such a filter mechan nism could be in a lowlevel monitoring se etting where a computer capa able of rating higher-ord Kalman filter are not presen der rs nt. gener V CONCLUSIO V. ON teration is a si imple estimate of the The sample rate alt baseli that works p ine pretty well comp pared to its comp petition yet is not fully adap s ptive to handle all different fo orms of ECG signals and cond ditions. V REFERENC VI. CES
[1] M Kaur, B. S M. Singh, and See ema, Comparisons of Differ rent Approaches for Removal of Baseline Wande from er nternational Con nference on Em merging ECG signals 2st In ds y. a, eburary Trend in Technology Mumbai. India pp. 30-36, Fe 2011. . se-Line [2] T.S Schilder and J.A. Alste van, Removal of Bas S. Wand der and Power-Line Interference from ECG by an Efficie enct FIR Filter w a Reduced Number of Taps IEEE with s Trans sactions on Biom medical Enginee ering, 32(12) pp. 10521060, April 1985. ,

Figure 8: Comparison Between Unalter ECG and 8 B red Filtered Artificial ECG A The RMS error was 54.3 which is 6.19% relative to the e % e amplitude of the signal an the S-T deviation was only o nd y 0.23%, thus the filtering see adequate. Th actual dataset s ems he t with baselin drift also perf ne formed very well under the filter r conditions. The result is sho in Figure 9. own

gure 9: Effect of Filter on Actual Signal f l Fig Finally, the baseline estima using this mo ate odel is compared d to that generated by the Kalman filter co used in the K ode e MIT/BIH fo a different EC dataset shown in Figure 10. or CG n
B. ECG Baseline W Wander [3] B Mozaffary an d M. Tinati, E nation using W Wavelet Packets World Acade s emy of Elimin Scien nce, Engineering and Technolo g ogy, pp. 14-16, March 2005. .

Figure 10: Comparison btw Kalman and Rate Alteration w. R

Das könnte Ihnen auch gefallen