jrevels
commented
2 years ago Hi! Thanks for using the package.
It'd be good to tag in a more current maintainer to check my guesses, but if I had to bet, it seems like the main problem is that OndaEDF doesn't seem to support BDF; it has a hardcoded assumption that signal element types will be Int16 (EDF's only signal type) whereas BDF support 24-bit.
It seems like it would be pretty easy to update OndaEDF to support this, though. I opened a draft PR to do this here: https://github.com/beacon-biosignals/OndaEDF.jl/pull/35
With this PR (using the data you provided):
julia> using EDF, OndaEDF, Onda
julia> bdf = EDF.read("data.bdf");
julia> bdf_samples = reduce(vcat, (transpose(EDF.decode(signal)) for signal in bdf.signals));
julia> onda, _ = OndaEDF.edf_to_onda_samples(bdf);
julia> eeg = only(s for s in onda if s.info.kind == "eeg");
julia> onda_fp1 = decode(eeg["fp1", :]);
julia> bdf.signals[2].header.label # just to prove that the second BDF signal is Fp1
"Fp1"
julia> bdf_fp1 = transpose(bdf_samples[2, :]);
# there may be slight inconsequential differences due to redithering that occurs during
# conversion if the conversion necessitates re-encoding the signals; in that case the two
# vectors won't be `==` but should still be `isapprox`
julia> isapprox(onda_fp1.data, bdf_fp1)
trueThis resolves the numerical difference in the converted data. Note that the data layout is different on purpose, though, as a result of translating to Onda. A few things to call out just in case it wasn't apparent:
-
OndaEDF only extracts EDF signals that it is "told" about, and will always try to sort extracted signals into the groups/order that it was "told" to. This is controlled by the
custom_extractorsargument, which is populated by default with the well-known signal labels defined by the EDF+ specification -
Since Onda has a notion of multichannel signals, OndaEDF splits
eegandecginto two separate multichannel signals as a result of the conversion (you can see I specifically grabbed theeegsignal above viaonly(s for s in onda if s.info.kind == "eeg")).
Finally, note that you can also just extract things yourself to Onda, and skip a lot of the stuff that OndaEDF tries to do for you. Here's how to do that (note, this should work on the current OndaEDF.jl release without my PR):
julia> using Tables
# grab the Onda.SamplesInfo and the relevant EDF.Signals for eeg
julia> eeg_info, eeg_channels = OndaEDF.extract_channels_by_label(bdf, ["eeg"], OndaEDF.STANDARD_LABELS[["eeg"]])[1][1];
# replace the computed encoding info with an unscaled floating point encoding
julia> eeg_info = Onda.SamplesInfo(Tables.rowmerge(eeg_info; sample_resolution_in_unit=1.0, sample_offset_in_unit=0.0, sample_type=Float32));
# grab our decoded data matrix
julia> eeg_data = reduce(vcat, (transpose(EDF.decode(c)) for c in eeg_channels));
# construct Onda.Samples object with our manually decoded data matrix and metadata
julia> eeg = Onda.Samples(eeg_data, eeg_info, false)
likanzhan
In the Dropbox Folder, I uploaded a sample data recorded in our study. We have a data file
data.bdfand a trigger data fileevt.bdf. Normally the data is processed with Matlab,readbdfdata.m.For now, we can use the
EDF.read()to read the data, and theEDF.decoded data are similar to the results obtained with MatLab (appendix 1)But the output of the
edf_to_onda_samples()function is totally different from that of the supposed results (appendix 2):What is the problem here?
Appendixes:
data_edf:data_onda: