fernandoandreotti/fecgsyn

FECGSYN toolbox for ECG and fetal ECG simulation

• 👋 Hi, I’m @fernandoandreotti
• 👀 I’m interested in machine learning, data engineering, electronic health records, smart contracts, amongst others
• 💞️ I’m looking to collaborate on intersting projects
• 📫 How to reach me [email protected]

fecgsyn's Issues

Ground electrode

I commented out the calculus using the ground electrode:

ground = repmat(mean(mixture),NB_ELEC,1);
mixture = mixture - ground;
mecg = mecg - ground;
fecg = cellfun(@(x) x - ground,fecg,'UniformOutput',0);
noise = cellfun(@(x) x - ground,noise,'UniformOutput',0);

A better solution then averaging the channels should be found. e.g. defining a grond electrode somewhere on the models back.

Improve input arguments FECGSYN_main_extract()

Enable the selection of a subset of methods

How to generate signals from different people?

Hi, I wonder how to generate signals from different people?
There are 10 pregnant women mentioned in the introduction of fecgdb data set. How are they generated? What's the difference?

out.mixture in run_ecg_generator

This is more question (or a problem with the documentation) than an issue but I wasn't sure how else to raise my question.
In the function run_ecg_generator several outputs are generated. Among them out.mixture defined as "generated ecg mixture [NB_EL x param.n matrix]".

• I am not sure what NB_EL represents, is it the number of electrodes capturing signals?
• Is it the same for the mother ecg and foetus ecg?
• Is there is a special meaning for the first one (it looks a bit different to the others)?

Publish on Physionet

v1.0 release

Generate short segments

Test if 1-5 second segments' generation works.

pcorr2 is missing

Originally used this code for generating faster correlations. Have to check permissions to include it here.

where's the data for this article?

could you tell me where the data for this article:

<Non-invasive Fetal ECG Signal Quality
Assessment for Multichannel
Heart Rate Estimation>

A total of n = 259 mul- tichannel recordings from 107 singleton pregnancies (diverse pathophysiological states and gestational weeks ranging from 17 to 39 weeks) were collected.

Thanks for your help!

Calibration depends on heart rate

Calibration of cardiac signals depends on the heart rate of the signals, it should be normalized using the number of peaks

@GUI: Stays busy after crashing

After an error on the run the GUI stays "Busy ..."

@GUI: Number of electrodes

It should be a field which assumes any number. If necessary, we can randomize the initial value, or put them as a sequential array around the belly.

Additionally, we should make another button and some presets, e.g.:
Foetal 32 + 2 (new scheme) - should be the default
Foetal 5 (old scheme original)
Adult 3 (Eindhoven)
Adult 12 (12-Lead)

GUI compatibility with fecgsyn v.1.1

The toolbox was largely updated and the GUI may not work properly. First steps are:

• update GUI for data generation (i.e. simulating and outputing signals)
• update GUI for FECG extraction (using any of the 8 available methods, see documentation).

GUI plots xlim

Add ability to scroll or modify plots xlim

Ectopic VCG Models

We only have one activated, since the other ones were problematic. Should we add some more?

Documentation is unavailable

Documentation available at: www.fecgsyn.com
I can't access the website, please check out.

Nonlinear phase between beats

In order to make the model more realistic and make things more interesting for extraction methods, a nonlinear phase should be designed. In order to do so, some nonlinear phase - RR relationship is required.

Realistic EHG

Uterine actiivity could be enhanced by using a real EHG dataset such as the newly released The Term-Preterm EHG Database (TPEHG DB): https://www.physionet.org/pn6/tpehgdb/

Update Naive Bayes to use fitcnb from Matlab

Warning: NaiveBayes will be removed in a future release. Use the predict method of an object returned by fitcnb instead.

In NaiveBayes/posterior (line 46)
In classreg.learning.internal.DisallowVectorOps/subsref (line 21)
Warning: NaiveBayes will be removed in a future release. Use the predict method of an object returned by fitcnb instead.
In NaiveBayes/posterior (line 46)
In classreg.learning.internal.DisallowVectorOps/subsref (line 21)
Warning: NaiveBayes will be removed in a future release. Use the predict method of an object returned by fitcnb instead.
In NaiveBayes/posterior (line 46)
In classreg.learning.internal.DisallowVectorOps/subsref (line 21)

Can the emulator customize SNRfm and SNRfn?

fetal/maternal (SNRfm), the fetallGaussian noise (SNRfn)

Problems generating data using sampling frequencies below 250 Hz.

I believe there is something on the noise generation

Include segmentation algorithm other than ECGPUWAVE

ECGPUWAVE somethings encounters problems while generating templates, an alternative solution is very welcome.

GUI Initial screen with easy access to FECGSYN functionalities

This is up for discussion, we have four main functionalities to which we should facilitate the access. This could be done by creating an initial screen with 4 large buttons:

• Data Generation (call run_ecg_generator() with specified param - redirects to current version of GUI)
• FECG Extraction (call specific method, or FECGSYN_main_extract() to extract with multiple methods)
• FQRS Detection (call qrs_detect() with desired thresholds and refractory period)
• Morphological Analysis (call FECGSYN_benchMorph() to selected path and channels - assuming out struct as input)

Multiple file conversion using fecgsyn2wfdb

The function could enable multiple file conversion.

PQRST Segmentation

The model could allow for segmentation algorithms a good training platform. In order to do so, we should spit out the PQRST waves on-set/peak/offsets. Maybe use ecgpuwave on the raw VCG? Points can be simply expanded using phase information.

@GUI: Allow noise modulating functions

The fecgsyn allows any funtion to modulate its noise signals. The GUI should be able to do the same.

Rethink how to maintain header

Need automated form of maintaining header in files. This pertains authorship and license information.

Some of the things that are currently inconsistent:

• VERSION_NUMBER
• VERSION_DATE
• COPYRIGHT_YEAR
• "Last update" useless since we use git
• Website should move all to www.fecgsyn.com

HRV Generating Unexpected Results

NSR does not seem physiological.

GUI Export code for data generation

Allow user to export the code for the designed experiment for semi-automated data generation.

add_noisedipole with octave

While using octave, and following the section "Initial NI-FECG simulation example" from the getting started webpage I encountered an error stating that the pca function is missing in the statistics package.

I noted that there is a "princomp" function in the package, and changed the line 166 from
[~,pc] = pca(noise_ar);
to
[~,pc] = princomp(noise_ar);

It resolved the issue, however I am new to octave/matlab, so I am not 100% sure that the function are interchangeable. Could someone kindly indicate that this change is making sense?

Improve jqrs thresholding by running it on 1-min segments

Fix this on QRSmincompare.m, which at the time processing the whole recordings (for FECGSYNDB = 5min)

@GUI: Posdev not defined

Posdev allows (or not) that the position of fheart and mheart vary within a small sphere. It should be included in the GUI.

add_noisedipole

Has a problem with sampling frequency under 360 Hz

@GUI: Warning by refreshing electrodes preview

Here is the error message:
Warning: linkaxes requires 2-D axes as input. Use linkprop for generic
property linking.

In linkaxes at 72
In FECGSYN_UI>update_axes at 1247
In FECGSYN_UI>update_geometry_preview at 1299
In FECGSYN_UI>cb_bt_geo_preview at 2166

Generating new Dataset- problem with saving ouputs

Hi,
I've run the following code in order to generate new data ( as shown in your examples):

param.fs = 1000; % data generation at 1kHz
param.n = 15*param.fs;
param.SNRfm = -3; % fetal maternal SNR
param.SNRmn = 8; % maternal to noise SNR
param.ntype = {'MA'}; % adding muscular artifact type noise
param.noise_fct = {1}; % noise at constant SNR level
debug = 5; % debug level (5 = max number of plots generated)
out = run_ecg_generator(param,debug);
FECGSYN_plotmix(out) % plots random channels of one of the created datasets

In the examples it's written that "The code will produce simulation files". I specified the desires path on the Matlab root with cd savedir but i don't get any simulation files.
Could anyone help me out?
Thank you in advance,

Rebecca

Update RLS to Matlab's newer versions

I was using a Matlab 2016 and this warning popped up:

Warning: adaptfilt.rls will be removed in a future release. Use dsp.RLSFilter instead.
> In adaptfilt.baseclass/deprecationMessage (line 12)
  In adaptfilt.rls (line 49)
  In FECGESN_lmsrls_canceller (line 96)
  In FECGSYN_adaptfilt_extraction (line 114)
  In parallel_function>make_general_channel/channel_general (line 914)
  In remoteParallelFunction (line 38)

On future releases should be fixed..

Calibration depends on electrodes position

Since both fetal and noise calibration depends on average maternal, foetal or noise power throughout the propagated channels, the final values (e.g. SNRfm = -6 dB) depend on the electrode configuration. Maybe considering the total power on the reference electrode may be a good alternative?

SNRmn is acting wierd

After changing electrode positions, calibration does not seem to be working properly.

Update wfdb2fecgsyn new annotation format

Update to match released FECGSYNDB