Learning Disentangled Representations in the Imaging Domain

Abstract

Disentangled representation learning has been proposed as an approach to learning general representations even in the absence of, or with limited, supervision. A good general representation can be fine-tuned for new target tasks using modest amounts of data, or used directly in unseen domains achieving remarkable performance in the corresponding task. This alleviation of the data and annotation requirements offers tantalising prospects for applications in computer vision and healthcare. In this tutorial paper, we motivate the need for disentangled representations, revisit key concepts, and describe practical building blocks and criteria for learning such representations. We survey applications in medical imaging emphasising choices made in exemplar key works, and then discuss links to computer vision applications. We conclude by presenting limitations, challenges, and opportunities.

Accepted by Medical Image Analysis. This paper follows a tutorial style but also surveys a considerable (more than 260 citations) number of works

References in corpus (28)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Bootstrap your own latent: A new approach to self-supervised Learning
• Unsupervised Domain Adaptation by Backpropagation
• Measuring and testing dependence by correlation of distances
• NICE: Non-linear Independent Components Estimation
• Barlow Twins: Self-Supervised Learning via Redundancy Reduction
• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges
• Towards Out-Of-Distribution Generalization: A Survey
• Learning Disentangled Representations with Semi-Supervised Deep Generative Models
• Discovering Hidden Factors of Variation in Deep Networks
• Generalizing to Unseen Domains: A Survey on Domain Generalization
• Towards Causal Representation Learning
• ALICE: Towards Understanding Adversarial Learning for Joint Distribution Matching
• Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style
• GeneGAN: Learning Object Transfiguration and Attribute Subspace from Unpaired Data
• Independently Controllable Features
• Deep Learning based Multi-modal Computing with Feature Disentanglement for MRI Image Synthesis
• Learning Disentangled Representation by Exploiting Pretrained Generative Models: A Contrastive Learning View
• Harmonic Unpaired Image-to-image Translation
• Adversarial Pseudo Healthy Synthesis Needs Pathology Factorization
• GLOWin: A Flow-based Invertible Generative Framework for Learning Disentangled Feature Representations in Medical Images
• Disentangling Influence: Using Disentangled Representations to Audit Model Predictions
• Information-based Disentangled Representation Learning for Unsupervised MR Harmonization
• Measuring Unintended Memorisation of Unique Private Features in Neural Networks
• Order-Guided Disentangled Representation Learning for Ulcerative Colitis Classification with Limited Labels
• Lung Swapping Autoencoder: Learning a Disentangled Structure-texture Representation of Chest Radiographs
• A Deep Network for Joint Registration and Reconstruction of Images with Pathologies
• Decomposing Normal and Abnormal Features of Medical Images into Discrete Latent Codes for Content-Based Image Retrieval

Cited by in corpus (4)

• A survey of recent methods for addressing AI fairness and bias in biomedicine
• SE(3)-Equivariant and Noise-Invariant 3D Rigid Motion Tracking in Brain MRI
• Disentangling representations of retinal images with generative models
• Approaching an unknown communication system by latent space exploration and causal inference