6.4 Overall results and conclusions

Using a function to repeatedly analyze ECGs with our algorithm, we obtained rates of correctly identifying theright type of rhythm and heart rate for specified noise levels. We then plotted these rates versus the corresponding SNR value foreach noise level to see how much noise we could tolerate before we could no longer accurately detect the signal’s information. Beloware these particular graphs for an 80 BPM Sinus wave, a 100 BPM Inverted T wave, and a 108 BPM Ventricular Tachycardia wave.

As shown in the graphs, all signals can be correctly identified with a SNR as low as 4, at which point theSinus wave starts to look like a Ventricular Tachycardia wave due to noise. The other two signals we tested correctly identify therhythm all the way down to a SNR of about 1.5. As far as heart rate goes, the Sinus and Inverted T wave always calculate the correcthear rate within two beats per minute for all signals with a SNR as low as 3, while a Ventricular Tachycardia signal can only do so.For practical purposes the one to two beat deviation in beats per minute may be considered negligible as all patients exhibit slightvariances in heartrate from one moment to the next and such a small variance is insignificant for medical purposes anyway.

The fact that the filter and the deciphering algorithm can tolerate such low signal to noise ratios shows greatpromise. Though this success rate may be lowered slightly as we add more signal types to be identified, it seems fairly certain thateven for a relatively large number of signal types we would still be able to identify the correct rhythm for any reasonable SNRvalue.