End-to-end autoencoding architecture for the simultaneous generation of medical images and corresponding segmentation masks

TL;DR

This paper introduces an end-to-end HVAE-based architecture that improves the quality of generated medical images and segmentation masks, especially under limited data conditions, outperforming GANs in this setting.

Contribution

The novel use of Hamiltonian Variational Autoencoder for simultaneous medical image and mask generation, enhancing data augmentation in scarce data scenarios.

Findings

Higher quality image generation compared to traditional VAEs

Outperforms GANs in data-scarce conditions

Effective across multiple medical imaging datasets

Abstract

Despite the increasing use of deep learning in medical image segmentation, acquiring sufficient training data remains a challenge in the medical field. In response, data augmentation techniques have been proposed; however, the generation of diverse and realistic medical images and their corresponding masks remains a difficult task, especially when working with insufficient training sets. To address these limitations, we present an end-to-end architecture based on the Hamiltonian Variational Autoencoder (HVAE). This approach yields an improved posterior distribution approximation compared to traditional Variational Autoencoders (VAE), resulting in higher image generation quality. Our method outperforms generative adversarial architectures under data-scarce conditions, showcasing enhancements in image quality and precise tumor mask synthesis. We conduct experiments on two publicly available datasets, MICCAI's Brain Tumor Segmentation Challenge (BRATS), and Head and Neck Tumor Segmentation Challenge (HECKTOR), demonstrating the effectiveness of our method on different medical imaging modalities.