Hello Everyone, I need some information about how to train the cycleGAN model using MRI Images and a Continuous Wavelet Transformer (CWT).
Dataset: OpenNeuro
Basically, In this dataset, we have 25 different subjects (participants) data of both MRI Images and EEG signals consisting of 120 Channels. Now my task is to give this dataset into cycleGAN, In cycleGAN we have two domains (Domain-A, I assign to MRI Images), and (Domain-B, I assign to EEG signals consisting of 120 channels). Now the problem is that cycleGAN typically takes image data not signal data. So I convert EEG signals into Continuous wavelet transform (CWT). But here I face another problem, Normally in CycleGAN we give one image to domain-A and expect to predict the one image to the domain-B or vice-versa. But here because we have 120 different channels of EEG signals, I fuse this (I take the mean of the different channels 120 channels and produce only a single channel signal) Then I give this single channel into Continuous wavelet transform (CWT), and it produces a single image.
Now this task is image-to-image translation, I tried to train this into typical CycleGAN but it is not working because I think in cycleGAN we have to give related data to both domain-A and domain-B but here both are different. Please guide me about that. How can I train this? Thanks in advance.
Note: I found one Research Paper about this topic but it is not based on cycleGAN it is based on CNN. In this research paper written that, If we use CNN + LSTM layer it gives best result specially for CWT data.