Adaptive Latent Diffusion Model for 3D Medical Image to Image Translation: Multi-modal Magnetic Resonance Imaging Study
WACV 2024
Summary
This paper proposes a method to translate the modality (e.g., T1, T2) of 3D MRI data using 3D VQGAN and a diffusion model. By utilizing the Multiple Switchable Spatially Adaptive Normalization (MS-SPADE) block, it can perform various modality translations with a single model and achieves state-of-the-art (SOTA) performance across all translations.
Questions
Q: What is the part proposed by the author in the VQGAN part?
A: MS-SPADE block was added to change the latent to fit the corresponding modality, and cycle loss was added.
Q: Does the latent enter the Diffusion model in a vector-quantized state?
A: The latent obtained through the trained codebook and the latent obtained through the MS-SPADE block are used as inputs.
Paper
Summary
This paper proposes a method to translate the modality (e.g., T1, T2) of 3D MRI data using 3D VQGAN and a diffusion model. By utilizing the Multiple Switchable Spatially Adaptive Normalization (MS-SPADE) block, it can perform various modality translations with a single model and achieves state-of-the-art (SOTA) performance across all translations.
Questions
Q: What is the part proposed by the author in the VQGAN part?
A: MS-SPADE block was added to change the latent to fit the corresponding modality, and cycle loss was added.
Q: Does the latent enter the Diffusion model in a vector-quantized state?
A: The latent obtained through the trained codebook and the latent obtained through the MS-SPADE block are used as inputs.