Latent Diffusion -Synthetic along with Original Image -Train UNET for Image Segmentation

[agokli]

Hello

I am using RSNA (3D stacked DICOM images) along with UNET to do image segmentation. I am using Transformers to do data augmentation. I will like to use Latent Diffusion model (3D) to generate synthetic data. Is it feasible to use concatenation (original images and synthetic data from Latent Diffusion) to train UNET model to image segmentation? I am successfully able to generate synthetic data using Latent Diffusion.

Any point3er or tutorial will be beneficial.

Thanks