Estimating ECG Signal from the Hand Wrist

Current techniques for measuring Heart Bit Rate (HBR) use two far measuring points on the subject body.

Abstract
Current techniques for measuring Heart Bit Rate (HBR) use two far measuring points on the subject body. For example, a popular sport application for HBR uses two electrodes on the chest. These methods are inconvenient. Our goal is to develop technology, which can measure ECG from the hand wrist, using two close electrodes on the subject wrist. The problems with such measure is that the Signal to Noise Ratio (SNR) is very low (~ -50dB and below). In this project the subject of measuring ECG from the wrist was studied, and an algorithm based on ARX model for estimating the HBR from wrist was developed.

Introduction
For measuring the ECG signal from the wrist, two electrodes of ECG instrument (Norav Medical PCECG1200) were connected to the subject hand, around the wrist on one hand only (‘one sided’), third electrode on the other hand for reference ECG signal (‘two sided’). The first stage was to prove that there is ECG signal in the sampled signal from the wrist. So one sided signal was taken and a method of averaging was applied on it. This method takes synchronized segments of the one and two sided signals and averaging them. When averaging the two sided signal we get the two sided signal, and when averaging one sided signal, the noise is canceled and the result in signal similar to the two sided signal:

1
Figure 1 – Schematic description of the averaging process

After proving that there is a ECG signal in the wrist, an ARX model was applied for estimating the energy of the ECG signal in the one sided signal. Looking at ECG signal, one can see that the ECG signal has two main parts: one is the action of the heart (called the QRS-T complex) and one in the rest time of the heart. The ARX model looks for templates of QRS-T complexes in the one sided signal. The model can be described in the following way:

2

Where U(n) represents the observed process at time n (in our case the one sided signal) and T(n) represents a single QRS-T sequence, which derived from an ensemble average of many synchronized trials.

3
Figure 2 – Schematic description of the ARX model

The B(n) filter coefficients energy are proportional to the probability of appearance of the template T(n) in the one sided signal. By applying a moving window on the one sided signal, calculating for each window the B(n) coefficients, a graph represents the template probability in the moving window can produced.

4

Figure 3 – ARX process – The moving ARX windows calculate each window the energy of the B coefficients

Results
The next example shows a result of applying this ARX method on one sided signal. Its seen that this method generally works. But in many cases this method estimates non existing peaks, or the time deviation is too large (above 0.05 sec which is the QRS time length).

5

Conclusions

  • There is ECG signal in the one sided sampled signal. This signal suffers from very low SNR
  • The ARX method can be applicable for signal with SNR above –30dB. This SNR can be achieved by building a special apparatus for measuring one sided ECG signal. In the signal we measured an average SNR of –60dB was measured
  • Trying to filter the one sided signal, in order to get a signal which from it we can measure the HBR, can’t be done

Acknowledgment
We would like to thank our supervisor Dr. Daniel H. Lange for his guidance and Johanan Erez for his time and support.