Computing in Cardiology 2011, Hangzhou, China, 18th – 21st Sept. 2011, Paper # 257 Omsai
Automatic Detection of Characteristic Points in Impedance Cardiogram S M M Naidu, P C Pandey, V K Pandey IIT Bombay, India http://www.ee.iitb.ac.in/~spilab,
Abstract: Estimation of stroke volume and several other cardiovascular indices using impedance cardiography requires error-free detection of the characteristic points in the impedance cardiogram. A technique for automatic detection of B, C, and X points, using Rpeaks in the simultaneously acquired ECG as reference, is presented. It does not require estimation of the baseline and selection of processing parameters. Use of the technique on preexercise and post-exercise recordings from healthy subjects and cardiac patients showed a very low detection error.
1. Introduction Impedance Cardiography (ICG) A non-invasive technique based on sensing the variation in the thoracic impedance Z(t) caused by variation in the blood volume in the thorax. ICG = – dZ/dt .
ICG (- d Z / dt )
( - dZ / dt )
15 % A X Tlvet Ejection time Measured |zmax|
Applications: Estimation of stroke volume (SV) & other cardiac indices - Z ( t)
ICG Characteristic Points A point: atrial contraction, before B-point, follows ECG P-wave. B point: aortic valve opening, 1st heart sound, deflection before C-point. C point: ventricular contraction, ICG peak. X point: aortic valve closure, 2nd heart sound, lowest value in ICG. O point: wide opening of the mitral valve.
PCG Time t
ICG & other related signals
2. Signal Processing Detection of B, C, and X points (the points most commonly used for calculating SV and other cardiac indices)
Beat-by-beat detection of characteristic points, without ensemble averaging. Developed after examining a large number of artifact-free and artifact-contaminated recordings. Baseline estimation & processing parameters selection not required. No restriction of record lengths. Useable in the presence of artifacts, without ensemble averaging.
Steps ICG cycle identification with reference to the automatically detected ECG R-peaks. C point: highest ICG point after the R-peak and within (R-R interval)/5. B point: first minimum preceding the C point. X point: the lowest point after the C point and within (C-C interval)/3.
3. Evaluation Material Pre- & post-exercise ICG recordings taken in supine position: 9 healthy S’s, 5 patients. Samp. rate = 500 Hz. ICG instruments: (i) developed in our lab, (ii) ‘HIC-2000’ (from Bio-impedance Tech., Chapel Hill, NC). Denoising: Respiratory artifact suppression by wavelet-based denoising (23 dB improvement in the signal-to-artifact ratio for signals highly corrupted by respiratory artifact).
Method Detection of characteristic points for unprocessed & denoised ICG. Detected points marked on the waveform. Quantitative evaluation Sensitivity = No. of correctly detected points / Total no. of points Positive predictivity = No. of correctly detected points / No. of total detected points Detection error = (No of failed detections + No. of missed detections) / Total no. of points
Detection of B, C, X points
Solid: ICG, dotted: ECG, R peak: triangle, C: inverted triangle, B point : circle, X point: diamond
(a) pre-execrcise rec. (SH9)
-0.2 0.2 (b)
(b) ‘a’ after denoising
-0.2 0.2 (c)
(c) post-exercise rec. (SH9)
-0.2 0.2 (d)
(d) ‘c’ after denoising
0 -0.2 0.1
(e) pre-exercise rec. (PT1)
0 -0.1 0.1
(f) ‘e’ after denoising
0 -0.1 0.2
(g) post-exercise rec. ( PT1),
0 -0.2 0.1
(h) ‘g’ after denoising
0 -0.1 0
Evaluation indices (%) for detection of characteristic points Unprocessed ICG
No of cardiac cycles = 545 Errors: mostly related to errors in R-peak detection
5. Conclusion Result summary: BCX detection with very low errors. Further work: Evaluation in a clinical setting for estimating SV & other indices.