My diploma thesis was made at Innomed Medical Zrt. First of all, I did literature research in filtering out the artefact what comes from chest movements. During resuscitation the ECG becomes extremely noisy because of the chest compressions. Currently the heart rate analysis can be done only when chest compressions are stopped, otherwise the result could be invalid. That’s why we need a filtering algorithm, which helps to restore the original (underlying) rhythm and then the shockable rhythm can be detected during resuscitation.
I’ve found a lot of articles in this theme and I chose two filters: LMS (Least Mean Squares) and NLMS (Normalized Least Mean Squares), and I implemented these filters in Matlab. The CPR (cardiopulmonary resuscitation), which occurs during resuscitation, is modelled by a sinus wave. To test the filters I used real ECG signals. The sum of ECG and CPR signals is the noisy ECG signal.
The LMS and NLMS are adaptive filters, which need the noisy ECG signal and the reference CPR signal to restore the original ECG. While adjusting the parameters and analysing the records I came to the conclusion that the NLMS filter fits the task better, because it has got faster convergence and it follows better the signal to be modelled.
In the next step, I discovered the VF detector used in the devices of Innomed Medical Zrt. Then, I created a software in Qt, that has a good user interface and replaces the previous Matlab program. The VF detector was integrated into this software. After that, I tested the filtering algorithm by comparing the result of the VF analysis of the filtered and original ECG. I found that in most cases, the filtering was good, the shock or no shock advice was the same for the original and the filtered signal.
Overall, I created an algorithm that uses a reference signal (CPR) to filter out the noisy ECG signal, so it is possible to perform the analysis during chest compressions and to indicate the user whether the rhythm is to be shocked or not.