In order to validate a Doppler radar used for Heartbeat Rate and Heart Rate Variability extraction, an ECG (electrocardiograph) is often used as reference. Which is the accuracy of an ECG? Can be the Doppler radar more accurate than the ECG? How the accuracy of an ECG could be determined?
If the doppler waveform indicates the blood flow and target is HR, then I would say the accuracy might be the rms error of two HR trace. one from ECG R peak detection, following HR estimation, and another one from peak detection on doppler waveform which should indicates the moment of systolic flow, namely the rhythm of heart pumping.
Thank you for your answer. We used the proposed method in our papers. In fact, I need to know how the accuracy of a real ECG was (or can be) measured. I would like to know the value of the accuracy of a real ECG. Was this parameter already determined? If yes, how ? The main question is: "Can be the Doppler radar more accurate than the ECG?"
I guess the question would be : what is the ground truth, if 'accuracy of real ECG' is mentioned, or what's the problem you are trying to solve. If my understanding was right, it was comparing alternative HRV estimation method by doppler with HRV estimation by 'real ECG', in that case, I would assume the signals from ecg leads, are ground truth.
Thank you for the sent paper. I don't know this paper, I must read it. However, my questions are the same: "Can be the Doppler radar more accurate than the ECG?" and "Which is the accuracy of a real ECG? Was this parameter already determined? If yes, how ?"
Hey Ghoerghe,
If your question is 'Can be the doppler radar more accurate than ECG', your potential assumption is you have an alternative ground truth of HR. Otherwise, you can't claim which one is better, and the only question one can answer is 'how likely dopper radar can get compatible measurement with real ECG on HR estimation'.
and as for the accuracy, I would say it has different meaning, for the classification problem, the accuracy is the percentage of the sum of diagonal counts over the whole counts in confusion matrix (table). In the regression problem, the accuracy could be defined as the root mean square (rms) error or likely. anyway, depend on the problem itself.
in our scenario, I would assume the rms on two HR traces, one reported by ECG R peak detection and another one from the systolic pulse.
If you are looking at another problem, like predict malfunction of valve, on ECG or on echocardiogram, you may facing a classification problem, and accuracy in this case may defined on the confusion matrix.
In my study, the ECG signal is used as reference, so it is assumed to be without error. Is this true? I don't know! I assume the ECG signal is "accurate". However, I can create a set-up in order to measure the accuracy of the Doppler radar. For example, it is possible to generate a periodic movement of a matallic surface with a given (and known) frequency and to measure this frequency with the Doppler radar. If this measure is accurate (let say 2%) but the rms error is 10% for HR measurements using ECG and the Doppler radar, which may be the conclusion concerning the accuracy of my reference?
Thank you for your answers. It is always useful to know another point of view. Are you working on this topic?
ECG records electrical activity of the heart beat by beat. Even when no pulse is detected (pulseless electrical activity) or if there is some interference due to a metal or excesive movement (in this case it will display a distortioned waveform or artifact). For that reason, ECG can be used as the gold standard for HRV and heartbeat rate evaluation
The question of accuracy of an ECG for HRV detection is, by itself already a treacherous one. Most of the time the underlying aim is not to look at the (ir-)regularity of the beating heart, but to find out how the autonomic system (ANS) is influencing the origin of the heartbeat, i.e. the firing of the sinus node in the right atrium. That signal is, sort of, picked up by looking at the initiation of the P-wave of the ECG, a difficult to exactly pinpoint moment in time, in view of its slow beginning and low amplitude.
Instead, people are looking at the R-wave of the ECG, neglecting the influence of the ANS on the conduction time of the electrical activation from atrium to the ventricle. The R-wave is better detectable; however, it shows variations in shape, due to the movement of the heart in the thorax with respect to the measuring electrodes, mostly due to the respiratory movements. And, when looking at the QRS-complex, one may still wonder which point of the 80-120 ms should be the fiducial point: the peak of the R-wave? That will change by the movements of the heart just mentioned. However, not very much; I have no studies at hand, but I would guess that 10-20 ms uncertainty would be maximal, depending on electrode positions.
Now, about the uncertainty in detection of a peripheral pulse wave by whichever method, this will be at least of the same order of magnitude as the uncertainty in the QRS-detection. However, its initiation is dependent on factors as the filling of the heart and the pressure in the aorta that is met by the contracting ventricle before it can open the aortic valve. In short, the uncertainty of the pre-ejection period.
All in all, I would go for the ECG, if available, if I want to have the best detection of HRV. If not, inaccuracies accumulate.