I'm not sure if a consistent signal will exist between people. A lot of variables can possibly interfere with your ability to establish a canonical waveform like this. Device placement, proportional to physiology is on. Likewise various cardiac parameters (ejection fraction, for instance), individual peculiarities in the vasculature, body mass, hemodynamic delay, and so on. I would start on easier questions first: a) can you achieve a consistent signal *within* participants, and get a BCG-derived interval which is a reasonable description of the R-R interval? b) can you establish any covariate which determines the between-patient changes in the BCG?

For a question like this, it's good to have panels of the signals you're describing as well. Trying to outline a signal in the abstract is hard.

I'm not sure if a consistent signal will exist between people. A lot of variables can possibly interfere with your ability to establish a canonical waveform like this. Device placement, proportional to physiology is on. Likewise various cardiac parameters (ejection fraction, for instance), individual peculiarities in the vasculature, body mass, hemodynamic delay, and so on. I would start on easier questions first: a) can you achieve a consistent signal *within* participants, and get a BCG-derived interval which is a reasonable description of the R-R interval? b) can you establish any covariate which determines the between-patient changes in the BCG?

For a question like this, it's good to have panels of the signals you're describing as well. Trying to outline a signal in the abstract is hard.

Hi Mark,

I agree with James that it might be difficult to obtain a BCG between subjects and at first trying to establish signals within participant makes sense. Generally, just averaging beats (even if the amplitude is normalized) might distort the resulting waveform. A potential solution, either within participants or over participants, might be to use warping (see "Evaluation of respiratory- and postural-induced changes on the ballistocardiogram signal by time warping averaging" or the mathod we proposed to warp signals under some constraints described in "Two-Dimensional Warping for One-Dimensional Signals—Conceptual Framework and Application to ECG Processing").

Greetings, Sebastian

Hi. Thanks both for your responses. I have been able to generate an average BCG signal for each patient, so there may be some interesting information within those signals. I will try the methods that sebastian recommended and see how things go.

Hi Mark. James and Sebastien had great input and I would agree with them. Methodologies for acquiring BCG vary greatly between studies, and comparing absolute values might not be feasible. I might have two other ideas that could be of interest to you.

1. Once you know which variables you will be comparing between subjects, why don't you compare the means these values directly, instead of going through some mean signal? In line with what has already been said, I would tend to choose relative values over absolute ones (I/J, for example, instead of I and J).

2. You could try to create some black-box 'model' of the signal, just some function of time that represents the part of the signal that you are interested in. Something like is suggested in the paper below maybe? If you can do that, then you'll have a good handle on individuals' whole signal and might be able to compare them more easily between subjects.

http://ieeexplore.ieee.org/abstract/document/7408577/

Dear Mark,

As Alex, James, and Sebastien have already commented on this it might be difficult to find a fully similar pattern/shape/waveform between different persons. Inter- and Intra subject variability should be always considered when dealing with BCG/SCG/GCG signals. As James truly commented many factors influence on the outcome signal. Sensor quality/resulotion, filtering, sampling rate etc. all also affect on the shape of GCG signal or waveform. Moreover, the way you attach the sensor on the body is also important to get a high quality signal, double sided tape seems to be a good solution for example.

Regarding ensemble averaging, there are many available approaches, I usualy use a simple robust peak detection over GCG/SCG signal and make a kind of deterministic approach by segmenting signal from each heartpulse untill the begining of the following pulse so that all cardiac cycles are merged in a matrix with certain size. But of course this may not be a applicable in all cases because of HRV, false/missed peaks etc. Same might apply even for ECG, some people may have extra beats, strong respiration induction over ECG signal, high HRV etc. In order to avoide wrong segmentation, as you also considered, I benefit from a back up signal, either SCG or ECG if available, to double check if each pulse is correctly detected and so on.

Finally, after processing/analyzing handreds of cases, I strongly believe that GCG possess less waveform variations as compared to SCG/BCG, meaning that gyro gives you kind of semi-stationary waveforms than other mechanical sigals. Note that there is no stationary cardiac signal, as physiology says.

I guess my comment is a bit late and you have probably came up with an applicable solution but feel free to ask any question regarding GCG/SCG.

Hi Mark,

I also agree that there would be a big variation between the "Average Waveforms" of different subjects. As an example I've tested the variation in BCG waveform/template due to the age (cardiac health) in this link: Conference Paper Investigating the Interaction between Ballistocardiogram and...

Going back to the old papers from Isaac Staar, they always grouped subjects by their ages and report results for different age groups, separately. So maybe you also want to group your subjects based on Age and Gender.

Best.