RichardObi / medigan

medigan - A Python Library of Pretrained Generative Models for Medical Image Synthesis
https://medigan.readthedocs.io/en/latest/
MIT License
115 stars 13 forks source link

[Documentation] Missing documentation for 00023_PIX2PIXHD_BREAST_DCEMRI #67

Open LHamnett opened 3 weeks ago

LHamnett commented 3 weeks ago

Hi,

It seems that there has been a GAN added which has been trained on the Duke-Breast-Cancer-MRI. However in the documentation page, this GAN does not have information about the parameters this model accepts.

I am also working on a research paper that involves training a GAN on the DCEMRI dataset, can you confirm if this GAN was trained on all of the images regardless of metadata or if it is possible to return a trained GAN for a specific class eg cancer type.

Thanks. Leon

LHamnett commented 3 weeks ago

@RichardObi for visibility

RichardObi commented 3 weeks ago

Hi @LHamnett, thank you very much for pointing out this issue. Model 23 was missing in the docs due to the .readthedocs.yaml requiring an update to version 2. The documentation for model 23 will now become available here. Apart from that, you may also find more detailed info for training this model in this repo and in this paper. The GAN was trained on 668 cases (mixed cancer types) out of the 922 cases available in the Duke-Breast-Cancer-MRI. You can find the 254 cases excluded from GAN training here.

Hope this helps and feel free to reach out anytime.