Multiscale Metamorphic VAE for 3D Brain MRI Synthesis
This work addresses the problem of realistic 3D brain MRI synthesis for medical imaging applications, representing an incremental advance over existing generative models.
The paper tackles the challenge of generating high-fidelity 3D brain MRIs with good coverage of the data distribution by proposing a multiscale metamorphic VAE that uses composable morphological transformations on a reference image. The result shows substantial improvements in FID scores while maintaining comparable or better reconstruction quality than prior VAE and GAN methods.
Generative modeling of 3D brain MRIs presents difficulties in achieving high visual fidelity while ensuring sufficient coverage of the data distribution. In this work, we propose to address this challenge with composable, multiscale morphological transformations in a variational autoencoder (VAE) framework. These transformations are applied to a chosen reference brain image to generate MRI volumes, equipping the model with strong anatomical inductive biases. We structure the VAE latent space in a way such that the model covers the data distribution sufficiently well. We show substantial performance improvements in FID while retaining comparable, or superior, reconstruction quality compared to prior work based on VAEs and generative adversarial networks (GANs).