Domain-Adaptive 3D Medical Image Synthesis: An Efficient Unsupervised Approach

TL;DR

This paper introduces an efficient unsupervised domain adaptation method for 3D medical image synthesis using a 2D variational autoencoder, significantly improving synthesis accuracy across unseen domains.

Contribution

It proposes a novel 2D VAE-based domain adaptation approach specifically designed for 3D image synthesis, addressing the challenge of domain shift in medical imaging.

Findings

Significant improvement in synthesis accuracy on unseen domains.

Effectiveness of the approach depends on the amount of adaptation data.

Hyper-parameter tuning impacts adaptation performance.

Abstract

Medical image synthesis has attracted increasing attention because it could generate missing image data, improving diagnosis and benefits many downstream tasks. However, so far the developed synthesis model is not adaptive to unseen data distribution that presents domain shift, limiting its applicability in clinical routine. This work focuses on exploring domain adaptation (DA) of 3D image-to-image synthesis models. First, we highlight the technical difference in DA between classification, segmentation and synthesis models. Second, we present a novel efficient adaptation approach based on 2D variational autoencoder which approximates 3D distributions. Third, we present empirical studies on the effect of the amount of adaptation data and the key hyper-parameters. Our results show that the proposed approach can significantly improve the synthesis accuracy on unseen domains in a 3D setting. The code is publicly available at https://github.com/WinstonHuTiger/2D_VAE_UDA_for_3D_sythesis