3
$\begingroup$

I'm currently a sophomore in college and pretty new to the field of research. I'm currently working on an existing algorithm for MRI de-noising and the results are nothing that great. I can't see how I can improve the results.

I've tried a few different tweaks and played with the code and parameters for sometime now. I also saw some papers on using neural networks for image de-noising as well.

I'm thinking of trying the same but I don't know if it makes sense to use neural networks on an already existing algorithm. Won't the neural network overpower my algorithm and all the results be based on it only? So if I use a pre-existing MatLab filter and use neural networks in it, will it differ much if I first use my algorithm and then use neural networks in it?

I have little experience with machine learning and hope that I made sense in my question. Any kind of help will be appreciated. Thanks.

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ Be advised that MRI is a quite special field of imaging and you can experience some unique difficulties, such as: - images in MRI can be complex - images are usually combined from several receive coils with spatially varying sensitivity - MRI DICOM images can be the result of several combination algorithms that already take into account some noise statistics - MRI images can be (truely) 3D (ie not simply stacking 2D slices) - noise statistics do depend heavily on the combination of the several receive coils $\endgroup$
    – M529
    Jul 7, 2019 at 15:09

1 Answer 1

Reset to default
3
$\begingroup$

My personal feeling is that you should do each things separately and compare the results. For example, take your MRI dataset and denoise using "standard algorithm 1", "standard algorithm 2" and "neural network algorithm 1". I would keep things reasonably simple unless you have a good justification of doing "standard algorithm 2 and neural network algorithm 1". Doing two in a row is going to be harder to justify in a paper.

(I think the other question that is difficult is how to tell if one denoising algorithm is "better" than another. I have written a couple papers on standard image processing denoising in MRI and it is always a difficult question to answer. Also, if the output of the denoising is a dataset that is to be interpreted by a radiologist then other consideration would have to happen vs denoising as a pre-step to some scientific quantification.)

One thing you could do is look through papers, for example at https://arxiv.org/ to see if there are other neural network denoising examples that can help guide you in parameter selection.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Wow that helped a lot. Thanks. Could you maybe share some of your published papers in a similar field so that I have a better idea? $\endgroup$
    – anonymous2718
    Jun 7, 2019 at 20:25
  • 1
    $\begingroup$ Well, for example: onlinelibrary.wiley.com/doi/pdf/10.1002/mrm.10492, but as I said, the format of the paper will depend a lot on what you are going to do after the smoothing / de-noising. For example in this paper the post-processing was to quantify myelin water maps. Good luck! $\endgroup$
    – brechmos
    Jun 9, 2019 at 1:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.