Available online at www.sciencedirect.com
ScienceDirect Procedia Computer Science 70 (2015) 289 – 295
4th International Conference on Eco-friendly Computing and Communication Systems, ICECCS 2015
Virtual Instrumentation Based Fetal ECG Extraction Chinmayee G Raja*, V Sri Harshab, B Sai Gowthamic, Sunitha Rd a,b,cDept. of EIE, Amrita Vishwa Vidyapeetham, Amrita School of Engineering, Kasavanahalli,
Carmelaram, Bengaluru 560035, India dDept.
of ECE, Amrita Vishwa Vidyapeetham, Amrita School of Engineering, Kasavanahalli, Carmelaram, Bengaluru 560035, India
Abstract ECG, the Electro Cardio Gram is a very important tool for the clinician to decide the cardiac health of a patient1. Fetal ECG is similarly a parameter to decide the child’s health throughout the gestation period of the child and during labour. Fetal ECG is contaminated with noise and the mother’s ECG. The main objective of this study is to isolate and extract only the fetal ECG from the mixed signals. The present study proposes a method to effectively and efficiently isolate only the fetal ECG. The database used is available from PhysioNet which provides benchmark signals of ECGs acquired from the mother’s abdomen. The algorithm used in this study is based on Fast ICA. Graphical representation software such as NI LabVIEW is preferred over Matlab due to the preciseness, cost effectiveness and the ease at which the algorithm can be understood, constructed and implemented. The present study depends on the beats/minute, amplitude and the correlation coefficient parameters to conclude the accuracy of the algorithm. © 2015 Chinmayee G Raj. Published by Elsevier B.V. © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license Peer-review under responsibility of Organizing Committee of the International Conference on Eco-friendly Computing and (http://creativecommons.org/licenses/by-nc-nd/4.0/). Communication (ICECCS 2015). Peer-review underSystems responsibility of the Organizing Committee of ICECCS 2015 Keywords: Biomedical signal processing; Fetal Electrocardiogram; LabVIEW; Virtual Instrumentation; Principal Component Analysis; Independent Component Analysis
1. Introduction Biomedical signal processing involves the analysis of bio signals to provide vital information for clinicians to make decisions. Engineers are coming up with new ways to process these signals using a large variety of mathematical formulae and algorithms. With these assessments, we can potentially determine the state of a patient's health using non-invasive measurements2. Heart defects especially are among the most common birth defects and leading cause for birth defect-related deaths. One out of 125 babies is born with some form of congenital heart defect. It is possible that the defect is so slight that the baby appears healthy for many years, or so severe that its life is in immediate danger. Congenital heart defects originate in early stages of pregnancy starting right at the time of formation of the fetal heart and they can affect any of the parts or functions of the heart3. Extraction of fetal ECG is a special case in biomedical signal processing and conditioning. In this case, the required signal i.e., the fetal ECG is of very low *Corresponding author. Tel: +1 (408) 431-3238; E-mail address:
[email protected]
1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Organizing Committee of ICECCS 2015 doi:10.1016/j.procs.2015.10.093
290
Chinmayee G. Raj et al. / Procedia Computer Science 70 (2015) 289 – 295
amplitude and the mother’s ECG is of higher amplitude; the signal to noise ratio is relatively low thus making the process more complex3. 1.1. FECG Parameters For a fetus with a gestational age between 32 to 41 weeks, the heart rate is in the range of 80-240 BPM (beats per minute) with an expected heart rate around 140 BPM and a peak to peak amplitude ranging anywhere between 3-25μV, with expected amplitude at 20μV2. To decide the algorithm for best isolation of FECG, Second Order Blind Identification (SOBI), Joint Approximation Diagonalization (JADE) and Fast ICA were tested4. Among ICA methods, Fast ICA, π- CA, JADE and Pearson ICA were compared based on the correlation coefficient parameter5. Both prefer Fast ICA over all other methods to extract fetal ECG. Another comparison of PCA and ICA was made for the accuracy check of extracted fetal ECG6, both gave almost the same accuracy. However, ICA gave unsorted independent components and thus was necessary to check for the required signal individually7. Another method which was well scored in the PhysioNet 2013 challenge is a combination of template subtraction (TS) and ICA. This is termed the FUSE method. FUSE was defined as the combination of a subset of the valuated methods; the methods being, ICA-TS, ICA-TS-ICA, TS-ICA, ICA and TS. The best method was concluded to be FUSE-SMOOTH (FUSE with a smoothening block introduced)6. ICA is the preferred method in most of the recent methods for fetal ECG extraction with satisfying results. This work also has used ICA along with pre- processing and post-processing methods. 2. Database and methodology Two datasets were used; the PhysioNet 2013 challenge dataset (set-a) which contained four abdominal signals. The second database, also from PhysioNet, contained four abdominal signals and a direct signal acquired from the scalp electrode. LabVIEW version 2014 was used as the platform. The toolkits used were the Advanced Signal Processing Toolkit and the Biomedical Toolkit. The Biomedical toolkit was used to import the bio signals onto the platform, and to perform measurements on the signal at later stages of the algorithm.. The four abdominal signals used were ‘AB-1, AB-2, AB-3 and AB-4’.
Figure 1
Chinmayee G. Raj et al. / Procedia Computer Science 70 (2015) 289 – 295
a. Pre-Processing Pre-processing was performed to make the data independent and thus ready for ICA,. PCA was used5. PCA block used was available with the Advanced Signal Processing Toolkit. However, it was found that before the usage of the PCA block, the noise due to the baseline wander needed to be removed. For this a HPF block with an upper cut-off frequency of 3Hz and a lower cut-off frequency of 0.5Hz was used with the Kaiser windowing function as an input to the filter block.
Figure 2 50 Hz power line interference was expected as almost all ECG signals have it9. To remove it, when Notch filter of 50 Hz was used, the waveforms however remained the same before and after the filter. To further reduce noises, the Wavelet de-noise block was used. Post de-noise, the signals were considerably noise free and smooth. After PCA, to check if the PCA made the data independent, the Whiteness Test block was used. After PCA, the confidence level of data whitening was found to be 99% indicating that the autocorrelation plot is almost entirely within the confidence region, unlike the correlogram plot before PCA.
Figure 3
291
292
Chinmayee G. Raj et al. / Procedia Computer Science 70 (2015) 289 – 295
Also, another test –Randomness test was performed to check for non-gaussianity. The Boolean indicator glowed; indicating that the data is thus randomized.
Figure 4 b. Processing The TSA ICA block available within the Advanced Signal Processing Toolkit in LabVIEW was used. The block has an input which allows the user to switch between two methods - Fast ICA and MatrixPencil ICA. As explained above, we chose the Fast ICA method as our main algorithm to extract fetal ECG. At the Index (number of ICs to search) input to the ICA block, our input was ‘-1’so that all the possible independent components are obtained separately at the output.
Figure 5 Four components were obtained at the output. Two components are periodic and similar to ECG. The other two components are noises as they are non-periodic and have irregular amplitudes.
Figure 6
Chinmayee G. Raj et al. / Procedia Computer Science 70 (2015) 289 – 295
Figure 7 c. Post-Processing & Measurements In spite of the initial pre-processing of signals, the noises still existed. Thus, the Wavelet de-noise block was used to further smoothen the signal. Post-processing was performed on the two components achieved after ICA.
Figure 8 The amplitude of the two independent components plays a vital role in deciding which is the fetal ECG. To be more certain, after post-processing, a Peak-Detector block was introduced to count the number of peaks in the signal. The signal which gave the number of peaks in the range of 80-240 was chosen as the required fetal ECG. Number of peaks is the beats per minute parameter. Also the amplitude of the fetal ECG did lie in the expected range of around 20μV. The values of heart rates (HR) obtained are tabulated. Since the direct ECG obtained from the scalp electrodes is available from the PhysioBank, the correlation coefficient of the Directsignal and the extracted fetal ECG have been calculated. These values are tabulated as well. 3. Results and Discussions This work comes under the very few fetal ECG extraction algorithms developed using LabVIEW platform, almost all of the FECG extraction algorithms are constructed using Matlab. LabVIEW being a Graphical User Interface (GUI) software, it is comparatively easy to construct and understand the algorithm, easy to trouble shoot and modify the constructed algorithm as desired. Also with the use of LabVIEW, integrating the required hardware is simple and
293
294
Chinmayee G. Raj et al. / Procedia Computer Science 70 (2015) 289 – 295
straight forward. LabVIEW acts as the professional development environment of virtual monitoring and control equipment, has good man-machine interface, and easy to setup system and reconstruct system and make custom function10. The BPM, amplitude and the correlation coefficient parameter have been used in this work to determine the extracted fetal ECG signal. The values of the same along with the dataset used are tabulated below. Table 1 PhysioBank Dataset Database
BPM found
Correlation coefficient(CC)
r01
114
0.7019
r08
104
0.9109
r10
165
0.6186
Average CC
0.7483
Table 2 PhysioNet 2013 Challenge Dataset (set-a) Database
BPM found
a01
174
a02
102
a03
204
a04
120
a05
138
a07
138
a08
120
Average of BPM found in both tables
137.9
Since the PhysioNet 2013 Challenge Dataset doesn’t have the direct scalp electrodes, it was not possible to calculate the correlation coefficient. Only the BPM calculation was done. The average value of all the BPM values was found and it lies close to the expected value of 140. 4. Conclusions The accuracy of the method used in this paper is dependent on the bio signal database. None of the techniques guarantee the same accuracy for all acquired ECG abdominal signals. The algorithm in this work will have to be slightly altered in two scenarios. If the noises have increased due to faulty ECG acquisition techniques, the pre-processing cut off frequencies will have to be altered accordingly. And, the decision as to which independent components contain the fetal ECG and the mother’s ECG is a visual task. To be more certain, both are given to the Peak detector
Chinmayee G. Raj et al. / Procedia Computer Science 70 (2015) 289 – 295
block. The values at #found terminal help in deciding this affirmatively. In most of the cases, the second independent component was the fetal ECG. In some cases, the third independent component was the fetal ECG. Thus, a technique which works equally well irrespective of the dataset and irrespective of the independent component obtained is yet to be developed.
References 1.
Biomedical Signal Processing [Internet]. [cited 2015 May 10]. Available from: http://www.embs.org/aboutbiomedical-engineering/our-areas-of-research/biomedical-signal- processing/
2.
Rooijakkers MJ, Song S, Rabotti C, Oei SG, Bergmans JWM, Cantatore E, et al. Influence of electrode placement on signal quality for ambulatory pregnancy monitoring.Comput Math Methods Med. 2014;2014.
3.
Sameni. A Review of Fetal ECG Signal Processing Issues and Promising Directions.The Open Pacing, Electrophysiology & Therapy Journal. 2010.
4.
Varanini M, Tartarisco G, Billeci L, Macerata A, Pioggia G, Balocchi R. A multi-step approach for noninvasive fetal ECG analysis. Computing in Cardiology Conference (CinC), 2013. 2013. p. 281–4.
5.
Martín-Clemente R, Camargo-Olivares JL, Hornillo-Mellado S, Elena M, Román I. Fast technique for noninvasivefetal ECG extraction. IEEE Trans Biomed Eng. 2011;58:227–30.
6.
Llamedo M, Martín-yebra A, Laguna P, Martínez JP. Noninvasive Fetal ECG Estimation Based on Linear Transformations. 2013;285–8. Available from: http://www.cinc.org/archives/2013/pdf/0285.pdf
7.
Petrolis R, Krisciukaitis A. Multi Stage Principal Component Analysis Based Method for Detection of Fetal Heart Beats in Abdominal ECGs. 2013;301–4.
8.
Behar J, Oster J, Clifford GD. Non-invasive FECG extraction from a set of abdominal sensors. 2013;297– 300.
9.
Pankaj K.Gakare, V. D. Jaiswal AKS. Baseline wander and power line interference removal in ECG signal. Academia [Internet]. 2010; Available from: Baseline_Wander_and_Power_Line_Interference_Removal_in_ECG_Signal
10.
Zhang LZL, Jiang XJX. Acquisition and analysis system of the ECG signal based on LabVIEW.2009 9th IntConf Electron Meas Instruments. 2009;
295
