search

RawthiL / PET_DeepAttCorrection

Whole Body Positron Emission Tomography Attenuation Correction Map Synthesizing using 3D Deep Networks
MIT License
22 stars 3 forks source link

NAC PET to AC PET #1

Open sarabanoo opened 1 year ago

sarabanoo commented 1 year ago

Hi, can I use this codes ( cycle-GAN or U_net ) for another dataset. I want to convert NAC PET to AC PET ?

RawthiL commented 1 year ago

I think that you can adapt the code to go directly from NAC-PET to AC-PET. There is some literature about it. You can modify this code to achieve that. Keep in mind that the result can contain hallucinated structures. This was one of the reasons that we had to avoid going directly from NAC-PET to AC-PET. We wanted mainly to estimate the attenuation image to then improve the effectiveness of simultaneous emission and attenuation reconstruction by providing a high quality prior.

P.S.: I have finished the thesis since my last update to this code, you might want to check it out here.

sarabanoo commented 1 year ago

thank's RawthiL . Do you have an opinion about which neural network is better for going directly from NAC-PET to AC-PET? The dataset that I have has scattered corrections applied to it and only the attenuation correction has not been applied to dataset.

RawthiL commented 1 year ago

You will need to review current bibliography, but as of 2020 the best approaches were to use GAN (like this topologies) You can start looking references to this work to see if there is any new one. https://pubmed.ncbi.nlm.nih.gov/31869826/

If you are looking for released/public code for the task of translating NAC-PET to AC-PET , I'm not aware of any. The experience I had trying to reach other authors for code releases or testing their models on new data was never successful.