limsungjoo/Myocardial_domain_adaptation

SPECT-to-CT Image Translation

Myocardial Domain Adaptation

I have studied various methods to solve domain adaptation for Myocardial 3D images. The following models have been developed and the results of the models were compared.
Submitted for a paper titled "Clinical Feasibility of Deep Learning-based Attenuation Correction Models for 201Tl-Myocardial Perfusion Single-Photon Emission Computed Tomography"

• Requirements
• Reg GAN
• Diffusion models
• Swin-CycleGAN
• Modified U-Net
• Experiments

---

Requirements

• pytorch: 1.10.2+cu113
• torchvision: 0.11.3+cu113
• CUDA version: 12.0
• Python: 3.8.8

Reg GAN

Results

Diffusion models

Results

Swin-CycleGAN

Results

Modified U-Net

Results

Results

Experiments

Comparison of results for several domain adaptation networks (Gray scale images)

Networks	PSNR	SSIM
Diffusion models	28.3914	0.9706
Swin-CycleGAN	28.5760	0.9602
Modified U-Net	33.6580	0.9903

Comparison of results for several domain adaptation networks (Color scale images)

Networks	PSNR	SSIM
Reg GAN	18.7497	0.7910
Modified U-Net	20.1862	0.8378